NURSING WORKLOAD AND STAFFING: IMPACT ON PATIENTS AND STAFF

Transcription

1 MARCH 2009 THINK. CHANGE. DO. THINK.CHANGE.DO CENTRE FOR HEALTH SERVICES MANAGEMENT UTS: NURSING WORKLOAD AND STAFFING: IMPACT ON PATIENTS AND STAFF

2

3 Nursing Workload and Staffing: Impact on Patients and Staff Professor Christine Duffield Michael Roche Professor Linda O Brien-Pallas Professor Donna Diers Chris Aisbett Kate Aisbett Professor Caroline Homer ISBN UNIVERSITY OF TECHNOLOGY, SYDNEY 3

4 4 UNIVERSITY OF TECHNOLOGY, SYDNEY

5 Roles of Contributors The roles of contributors during the project were as follows: Professor Christine Duffield (Centre for Health Services Management UTS) o Project Director. Cross-sectional design. Cross-sectional sample definition. Interpretation of cross-sectional and longitudinal analysis. Report development. Michael Roche (Centre for Health Services Management UTS) o Longitudinal data collection. Cross-sectional sample definition. Crosssectional data collection and entry. Analysis and interpretation of crosssectional data. Report development. Professor Linda O Brien-Pallas (Nursing Health Services Research Unit University of Toronto and Adjunct Professor UTS) o Cross-sectional design and supply of instruments, syntax for cross-sectional analysis. Analysis and interpretation of cross-sectional data. Interpretation of the longitudinal data. Professor Donna Diers (Yale New Haven Health System [USA] and Adjunct Professor UTS) o Longitudinal study outcomes design. Interpretation of longitudinal data. Analysis of cross-sectional data and the integration of both methods. Report development. Chris Aisbett (Laeta Pty Ltd) o Collation and editing of longitudinal data. Analysis and interpretation of longitudinal data. Report development. Kate Aisbett (Laeta Pty Ltd) o Analysis and interpretation of longitudinal data. Report development. Professor Caroline Homer (Centre for Family Health & Midwifery UTS) o Cross-sectional design. Report development. UNIVERSITY OF TECHNOLOGY, SYDNEY 5

6 6 UNIVERSITY OF TECHNOLOGY, SYDNEY

7 Acknowledgements The investigators wish to acknowledge the commitment of ACT Health to improving patient safety and the working lives of nurses through funding this study. The ongoing involvement of and input from senior staff in ACT Health and its two hospitals has been critical to the success of this project. We would also like to acknowledge the support and guidance provided by the Senior Nurses associated with this project throughout its duration: the Chief Nurses, Adjunct Professor Jenny Beutel for her commitment to ensuring this project was funded and Ms Joy Vickerstaff to whom this Report was handed; and the Directors of Nursing, Ms Joy Vickerstaff and Ms Sue Hogan who facilitated access to their hospitals and data collection. The additional assistance and support of Leonie Johnson, Michelle Cole, and other staff of the Canberra Hospital Research Centre, and of Sue Minter of Calvary Hospital was also gratefully received. Without the assistance of all the staff in the Nursing and Midwifery Office, particularly Sonia Hogan and Heather Austin, in their responses to our numerous requests for assistance, this project would not have been completed. The team also acknowledges the extraordinary diligence of Dianne Pelletier who coordinated the cross-sectional data collection process and acted as the trouble-shooter and liaison throughout the project. The research team is indebted also to the generous assistance provided by Dr Barbara McCloskey in allowing us to use her SAS (analytic software) code for the outcomes algorithms, Sping Wang and Xiaoqiang Li of The Nursing Health Services Research Unit (University of Toronto) for the use of their SPSS syntax, Nancy van Doorn of Laeta Pty for her extensive work in data cleaning and analysis, Christine Catling-Paull for her comprehensive review of the literature, and Jane Ewing for her preliminary data analysis. In addition, the assistance of ACT Health and Calvary Information Technology staff in the extraction of workforce data was indispensable. Finally, the researchers wish to recognise and acknowledge the support provided by the nursing profession throughout the Territory and in particular, those nurses who willingly gave of their time to complete the surveys, tolerated our intrusions and answered our questions. UNIVERSITY OF TECHNOLOGY, SYDNEY 7

8 8 EXECUTIVE SUMMARY

9 Executive Summary This study was commissioned by ACT Health to inform future policy decisions on managing nursing workload in the Territory. The Australian Capital Territory (ACT) is the smallest of Australia s six states and two territories. However it has the highest population density and is the only state or territory without a sea border. The health needs of its residents are served by only two public hospitals, The Canberra Hospital and Calvary Public Hospital, as well as three private hospitals. Planning and sample definition commenced during late Cross-sectional data collection commenced September 2006 and was completed by November Longitudinal patient data were collected from the ACT Administrative Data System for two years ( ) and nursing payroll (workforce) data where possible for the same years, hospitals and wards. The study of hospital (N=2) nursing wards (N=16) used longitudinal data held at Territory levels to associate nursing workload and nursing skill mix (defined as the percentage of RNs) to patient outcomes from In-depth cross-sectional data collected from 16 medical-surgical wards in the two hospitals in 2006 amplified the findings. In addition, a variety of relationships between the work environment of nurses and patient outcomes were examined, as were nurses job satisfaction and intention to leave. The small sample across only two hospitals means that comparisons with other studies (for example similar work conducted for NSW Health), must be viewed with caution. NSW and ACT are different health systems and should not be compared without careful analysis of admission and case-typing practices. Administrative divisions such as acute, sub-acute, non-acute, daycase, admitted ED patient, nonadmitted ED and Outpatients are not standardised across health systems. However where relevant, comparisons have been made. The focus of this study was on medical and surgical nursing wards/units, the operational unit where the work of patient care and cure happens, where innovation can be most readily introduced with real consequences for patients and staff, and where the relationship between hospital resources and patient outcomes needs to be studied. UNIVERSITY OF TECHNOLOGY, SYDNEY 9

10 This study was designed to: a) Improve understanding of what constitutes nurses workload in medical and surgical units across the two hospitals in the Australian Capital Territory. b) Examine whether patient acuity and length of stay (LOS) have changed over time, and the impact on nurses workload. c) Examine the impact of skill mix (the proportion of registered nurses to total clinical nurse staffing) on patient outcomes as adverse patient circumstances (casemix controlled in longitudinal data). d) Determine the impact of the nursing work environment on patient and nurse outcomes. This information would assist ACT Health to: 1. Identify and implement innovative models of practice and care where applicable; 2. Identify how best to meet the health service needs of the community; 3. Identify how to achieve the capacity and capability required to meet high standards of practice and safe outcomes. Nursing Workload Across Australia, the nursing work environment and consequently nursing workload, has changed considerably over the past few years. This trend is also evidenced in the ACT data where the ever increasing patient turnover rate is impacting on nursing hours required to meet workload. In the longitudinal component, nursing workload on the ward is composed of patient requirements measured as AR-DRGs, plus movement of patients on and off wards. Nursing workload is also influenced by the amount of time patients spend on nursing wards length of ward (and hospital) stay. Shorter length of stay compresses nursing work. In the cross-sectional component nursing workload was measured using a standardised and validated measure, the PRN-80, which estimates the hours of care required for a patient for the coming 24 hours. Information was collected from the uncoded medical record by trained data collectors. 10 EXECUTIVE SUMMARY

11 Staffing levels have increased overall at Canberra Hospital during the study period. Most ward staffing is matched to acuity adjusted patient load (workload). In contrast, there has been an increase in the workload of nurses Calvary Hospitals during the study period. Using longitudinal data, the average number of different case types (AR-DRGs) per ward was calculated. The number ranges from a low of 164 to a high of 459, from a possible range of 613. The wider the range of DRGs cared for in a ward the greater the workload as nurses who work on these medical and surgical units must understand the care requirements, the pharmacology, the treatments, the protocols and preferences of specialist medical staff for an increasingly various patient assignment. There is a growing awareness of the impact that the movement of patients to and from nursing wards has on nursing workload (churn). Churn includes the effect of admission to Emergency Departments (ED) so increased rates of admission to wards through ED increases churn. Increased throughput, combined with strategies that result in the movement of patients as space becomes available on the most appropriate ward for their diagnosis, also increases churn. This bed movement is in addition to patient transfer required by the treatment regimen itself from ward to imaging, back to ward, and so forth. Each new admission, transfer, or discharge, requires documentation, orientation, clinical assessment and management review, and other tasks associated with the patient. Accompanying a patient to another ward or service may take a nurse away from his/her assignment of patients or tasks for an unknown period of time. In the longitudinal study patients visited 1.24 and 1.32 wards per episode at the two hospitals in an average length of stay (LOS) in hospital of 2.9 and 3.2 days respectively. When attention was restricted to patients who had some contact with the wards in the study the ward visit figures became 1.64 and 1.84 respectively and the average LOS figures were 8.9 and 6.3 respectively. Either way, the ward visits were less than the 2.26 wards per episode found in the NSW Health study. In the crosssectional study patients per bed was calculated per ward by dividing the number of patients per day by the number of beds. This calculation does not include bed movements within the ward. The mean was one patient per bed per day, again less than the 1.25 found in NSW. Both these results may reflect better bed management strategies. UNIVERSITY OF TECHNOLOGY, SYDNEY 11

12 Nursing hours per patient day (NHPPD) provided varied considerably on a per day basis (mean 6.5, range ) and were reasonably normally distributed though the data, indicating significant variation between and within wards. When patient needs vary significantly, staffing is more difficult to predict and can result in an increased workload for nurses because staffing may fail to match patient needs. The cross-sectional study used the PRN-80 (see Table 13, page 39 for further explanation), a standardised and validated tool (Chagnon, Audette, Lebrun, & Tilquin, 1978; O'Brien-Pallas et al., 2004) which measures the minutes of care (later translated into hours) required (both direct and indirect) per patient for the coming 24 hours. Information was collected from the un-coded medical record by trained data collectors. By comparing the hours of care required (using the PRN-80) and the hours of staffing provided taken from the ward roster, on average, approximately one half hour per day of additional care is required to meet each patient s needs. In addition, there was considerable variation across the sample. The difference between the minimum and maximum requirements per ward-day ranged from just over 4 hours to 10.7 hours. This degree of variability in care needs makes it difficult to predict the staffing required, and the discrepancy between hours needed and available hours may impact on workload, quality of care and the work environment. Nurses self-reported an average of 1.3 tasks per nurse per shift delayed and 1.5 tasks per nurse per shift not completed. The tasks not done include a range of care and comfort measures: talking with patients, pressure area care, oral hygiene and patient/family teaching, mobilisation and turning patients, adequate documentation and the taking of vital signs. Just over one-third (34.3%) of nurses reported they were unable to comfort and talk to their patients on the most recent shift. A small response rate was seen for night shift so statistical comparisons could not be made, but an apparently similar rate of tasks delayed was found, with a lower rate of tasks not done. Similar factors were influential in regard to both tasks delayed and tasks not completed. The proportion of nurses indicating less time available to deliver care, the amount of additional time required to complete care this shift, and the proportion of hours worked by agency staff were common elements. As these factors increased so did the rate of tasks delayed or not done. Additional predictors were identified in regard to the rate of tasks not done. These included the proportion of patients admitted from a care facility and the amount of involuntary overtime reported. An increase in the proportion of patients admitted from a care facility led to an increase in tasks delayed. 12 EXECUTIVE SUMMARY

13 In terms of indirect or additional nursing care activities, nearly half of the respondents reported that these included delivery or retrieval of patient meal trays (47%), cleaning (46%) or clerical duties (45%). Over one-third (36%) of nurses order, co-ordinate or perform ancillary work; 29% arrange discharge referrals and transport, while 9% transport patients. Starting IVs (35%), undertaking routine phlebotomy (17%) or ECGs (14%) were also undertaken by nurses. Nurse Staffing and Skill Mix Using the longitudinal data, nursing skill mix, defined as the proportion of registered nurses (RNs) to clinical nurse staffing (Shullanberger, 2000), is highly variable across the sample wards ranging from 49% to 80% at Canberra Hospital and 57% - 89% at Calvary Hospital in the final period of analysis. Skillmix was lower in wards with aged or rehabilitation casemix, higher in specialty surgical wards. Several wards at Canberra Hospital have had a steady increase in hours worked. At Calvary Hospital all wards have had an increase in hours worked, although as noted previously this has not matched increases in workload. In the cross-sectional data most wards had between 60% and 80% RN staff. Only twelve ward-days over six different wards employed nurses which were other than RN and EN categories and the percentage of these other nurse hours worked ranged from %, with two outliers at 22.4 and 24.5%. There were considerable differences in the proportion of full-time to part-time, casual or agency hours worked. There were two wards which had less than 40% full-time staff. Part-time staff ranged from % and casual staff ranged from 1 3%. Four wards in the sample employed no agency staff at all, while the remaining 10 wards employed between 1 8% agency staff. However, there is considerable variation in these figures when reported on a ward-day basis. The lowest percentage of full-time hours worked on one ward-day was 10.5% and the highest percentage was 93.3%. There were ten ward-days which had less than 40% full-time staff and two ward-days which had more than 80% full-time staff. There were great variations in the proportion of hours worked per ward-day by grade. RN L1 staff worked on average 51.6% of the hours with a large range from %; RN L2 staff worked on average 16.8% with a range of between 0 51%; and ENs worked 29.9% of hours, also with a large range of 0 66%. UNIVERSITY OF TECHNOLOGY, SYDNEY 13

14 Patient Outcomes Twelve clinical Outcomes Potentially Sensitive to Nursing (OPSN) were examined in the study. In addition, failure to rescue (death following certain OPSN) was counted in the longitudinal data. In the cross-sectional study data were collected from un-coded patient records or the ward reporting system and included falls (with and without injury) and medication errors (with and without patient consequences), events that cannot be captured in administrative data. The statistically significant findings supported the hypothesis that more nursing hours per patient reduces patient length of stay, but the size of the effect was small. It was found that if the two hospitals were to increase their RN hours by 10%, only a minor reduction of 1-2% in patient length of stay would result. However when patient outcomes as Outcomes Potentially Sensitive to Nursing (OPSN) were examined, it was found that increasing RN hours by 10% could produce decreases in the adverse event rates studied from 11% to 45%. In the cross-sectional study 26 (4.3%) patients in the study were found to have experienced a fall with or without injury, and some of these patients had experienced both types of fall. Two patients experienced medication errors without consequences. Out of the 601 patients studied, 34 (5.7%) experienced time-based medication errors, lower than found in the NSW study. Falls also were lower in the B1 hospital but higher in the A hospital than in NSW data. As a result of the low rates of adverse events, no relationships could be established. Work Environment The cross-sectional design provided insight into nurses perceptions of their working environment, their ability to practice comfortably, and the relationship between nurses perceptions and patient outcomes. Most nurses (88%) rated the quality of care as excellent or good over the past shift. When asked to indicate whether the quality of care given over the last 12 months had changed on their wards, 80% of respondents indicated that it had improved or remained the same. 14 EXECUTIVE SUMMARY

15 Results from the Nursing Work Index-Revised (NWI-R) indicate that on four of the five measures, that is, nurse autonomy, nurse control over practice, nurse-doctor relationships and resource adequacy, nurses in ACT scored higher than did nurses in NSW. Nurse leadership was slightly lower in the ACT data than NSW. Higher levels of autonomy, control over practice and nurse-doctor relations correlated with lower discrepancy between nursing demand and supply (hours of care required compared to those provided). Conversely, a high nursing demand/supply figure (indicating wider discrepancy between hours of care required and that supplied) related to lower levels of autonomy, control over practice and nurse-doctor relations. When asked whether they had experienced a physical or emotional threat or actual abuse during the last five shifts, 33% of respondents experienced emotional abuse but up to a maximum of 58% of staff on a ward did. In terms of threat of violence only 21% experienced this and while there were wards where no staff experienced a threat of violence, up to a maximum of 67% of staff on a ward did. The results are similar for physical violence where 15% of staff experienced this in the past five shifts and up to 58% of staff on a ward did so. The source of violence was nearly exclusively patients and families. Patients and families were responsible for most physical assaults (96.6%) and threats of assault (95.1%) and emotional abuse (69.7%). Nurse Outcomes 71.5% nurses were satisfied with their job and even more (79.5%) were satisfied with the profession. Furthermore 74% do not intend to leave their current job in the next 12 months. Job satisfaction increased with greater satisfaction with nursing, resource adequacy and total nursing hours provided, while decline in job satisfaction was related to increases in the number of shifts missed and increased age of the respondent. Nurses who were satisfied with their job and who perceived they had adequate resources were more likely to be satisfied with their profession, while those in temporary employment were less satisfied with nursing. A higher patient turnover also predicted satisfaction with nursing. Nurses were more likely to intend to leave their current job if they were required to re-sequence their work frequently, if there was a higher proportion of agency hours worked on their ward and if demand for nursing care per day exceeded supply. Nurses who had worked longer and who were satisfied with their job were less likely to plan to UNIVERSITY OF TECHNOLOGY, SYDNEY 15

16 leave. Nurses indicating they had more time to deliver care per shift were more likely to leave. Those working on wards with a higher proportion of patients waiting for a care facility were less likely to intend to leave. There was considerable variability between the wards. Overall, the study provides insight into patterns in nursing staffing, the work environment and patient outcomes in ACT public hospitals. The results suggest that to successfully manage a hospital system requires an understanding of the nature of the work and a commitment to matching resources to workload. 16 EXECUTIVE SUMMARY

17 UNIVERSITY OF TECHNOLOGY, SYDNEY 17

18 Table of Contents. 1. Introduction Purpose and Objectives Organisation of the Report Glossary Literature Review Study Design & Ethics Approval Study Design Ethics Approvals Samples and Data Collection Longitudinal Component Cross-sectional Component Data Analysis Longitudinal Analysis Cross-sectional Analysis Findings Longitudinal Findings Patterns in Skill Mix Patterns in Staffing Levels Findings for OPSN other than ALOS Conclusion Cross-sectional Findings Patient Characteristics Nurse Characteristics Ward Characteristics Skill Mix Characteristics Nursing Workload Work Environment Quality of Care Violence Experienced Satisfaction and Intention to Leave Patient Outcomes Outcome Predictors Nurse Outcomes Limitations Summary and Discussion References Appendices EXECUTIVE SUMMARY

19 UNIVERSITY OF TECHNOLOGY, SYDNEY 19

20 1. Introduction Nurse staffing in Australian hospitals has received greater attention recently with projections that the current shortage of nurses is unlikely to abate, particularly as the workforce ages. An overall annual increase in demand for nurses of 2.56% until 2010 has been predicted, with 180,552 Registered Nurses (RNs) being required by that time. A shortfall of approximately 40,000 is expected (Access Economics, 2004a; Karmel & Li, 2002). Current workforce predictions indicate that the retirement of large numbers of nurses in the baby boomer age bracket and the lower age at which female nurses retire will exacerbate current shortages (Schofield & Beard, 2005). It is possible that half the nursing workforce will be retired within 15 years (ARHRC, 2005). Efforts to recruit more people into the profession without addressing retention will not have a sustainable impact unless measures are undertaken to understand and address nursing workload and the quality of the work environment for nurses. These factors have been shown to have a significant impact on patient outcomes. Much of the nursing workforce comprises general (although still highly specialised) medical and surgical nurses. Not only are the majority of hospitalised patients found in general medical/surgical wards, but also, it is frequently these nurses who move to more specialised clinical areas such as intensive care, midwifery or mental health where there are already documented shortages (AHWAC, 2002a, 2002b, 2004; VDHS, 1999). This study was commissioned to examine factors which impact on nurses workload, particularly at the ward/unit level (medical and surgical) but in addition, examines the relationships between patient outcomes, the nursing work environment, nursing skill mix and workload. Study at the ward level enables a greater understanding of the relationships between the factors mentioned above but more importantly, can provide greater insight for those charged with responsibility for managing staff and patient care. 20 INTRODUCTION

21 Purpose and Objectives This study examined several questions fundamental to the design and implementation of optimal models of nurse staffing within ACT, collecting data from two time perspectives longitudinal and cross-sectional, in order to: a) Improve understanding of what constitutes nurses workload in medical and surgical units across the two public hospitals in the Australian Capital Territory. b) Examine whether patient acuity and length of stay (LOS) have changed over time, and the impact on nurses workload. c) Examine the impact of skill mix (the proportion of registered nurses) on patient outcomes as adverse patient circumstances (casemix controlled in longitudinal data). d) Determine the impact of the nursing work environment on patient and nurse outcomes. This information would provide a basis for ACT Health to: 1. Identify and implement innovative models of practice and care where applicable; 2. Identify how best to meet the health service needs of the community; 3. Identify how to achieve the capacity and capability required to meet high standards of practice and safe outcomes. Organisation of the Report The longitudinal context provided by two years of ACT administrative data grounds understanding of data collected at the coal face in the cross-sectional design in one eight week period of time. In the interests of readability, most of the detail about data acquisition, management and measurement are contained in Appendices. Throughout the Report, we will move from descriptions of patients and their experiences to nursing workforce as skill mix, hours of care and back to patient outcomes. Nursing resources cannot be understood without understanding the context UNIVERSITY OF TECHNOLOGY, SYDNEY 21

22 in which nursing is practiced the work environment, the patients who require care and the staff providing that care. To assist the reader, a glossary of terms used in the various methodologies is presented on the following pages. Glossary TABLE 1 DEFINITION OF WARD TYPES Type Cross-sectional Study Longitudinal Study Medical Wards designated as Specialty Medical or Medical by the hospital Wards with a casemix of predominantly Medical AR-DRGs. Calculated per year. Surgical General, Mixed Medical-Surgical Wards designated as Specialty Surgical or Surgical by the hospital Wards designated as Medical- Surgical by the hospital Wards with a casemix of predominantly Surgical AR-DRGs. Calculated per year. Wards with no predominant casemix. Calculated per year. Other N/A Other ward types such as Intensive Care Units, Emergency Departments, and Day Units Ward type selection in the longitudinal component was made for fairly broad AR- DRG case-types and overnight stays. Please note the difference in definitions between the two methods. One of the difficulties in this study was recognising that what a hospital defined as a medical or surgical ward for example, might well be an historical label not supported by casemix analysis. 22 INTRODUCTION

23 TABLE 2 LONGITUDINAL COMPONENT: OUTCOMES POTENTIALLY SENSITIVE TO NURSING (OPSN) DEFINITIONS* Item Central Nervous System (CNS) Complications Deep Vein Thrombosis/Pulmonary Embolism (DVT/PE) Decubitus Ulcer (Pressure ulcer) Gastrointestinal Bleeding (Ulcer/GIB) Pneumonia Sepsis Shock/Cardiac Arrest Urinary Tract Infection (UTI) Failure To Rescue (FTR) Physiologic/Metabolic Derangement Pulmonary Failure Surgical Wound Infection Mortality Length Of Stay (LOS) * Adapted from Needleman, et al. (2001) Detail Complications such as confusion or delirium. Nurses intervene to create a supportive environment, such as structuring sleep patterns etc. Thromboses (blood clots) are frequently related to periods of immobility. Early and frequent mobilisation of patients is an important activity performed by nurses. Decubitus ulcers are caused by prolonged pressure on skin areas, usually due to immobility. Mobilisation and positioning of patients are central activities performed by nurses. In most cases, gastrointestinal ulcerations and bleeding are stressrelated complications in hospital patients. Nursing plays a role in relieving psychological stress and detecting ulcers at an early stage. Two key risk factors for hospital-acquired pneumonia are prolonged immobility, which leads to inadequate ventilation of parts of the lungs, and inappropriate or failure to perform pulmonary hygienic techniques. Nursing care influences both risk factors. Sepsis, defined as life-threatening and systemic infection, can result when a hospital-acquired infection is left untreated. The two most frequent hospital-acquired infections (UTI and pneumonia) are considered to be nursing sensitive. Both pulmonary failure and cardiac arrest are endpoints to a continuous deterioration in a patient s status. UTI is a frequent complication in hospitalised patients, particularly those with indwelling urinary catheters. Infection can result from inattention to sterile techniques when placing indwelling urinary catheters or from inadequate attention to time consuming toileting programs for incontinent patients. Defined as mortality of patients who experienced a hospital-acquired complication, studies have shown failure to rescue to be related to hospital quality and nursing. The underlying rationale is that excellent care is more likely to prevent patients with complications from dying. Operationally defined here as death following sepsis, shock, GI bleeding or DVT. Imbalances in electrolytes and blood sugar levels are very common in hospital patients. Given the central role of nurses in patient monitoring and reporting abnormal lab values to the treating team, slight imbalances can be caught quickly and corrected in a timely manner in well-staffed hospitals. Both pulmonary failure and cardiac arrest are endpoints to a continuous deterioration in a patient s status. Because nurses are responsible for pre-operative preparation of patients, which includes skin cleansing and administration of antibiotics, surgical wound infections could be influenced by the quality of nursing care. A number of studies have related mortality to nurse staffing patterns in hospitals. Nurses play an important role in discharge planning. They can ensure that a patient is not discharged prematurely or kept in the hospital for too long and thereby expose them to hospital acquired complications. UNIVERSITY OF TECHNOLOGY, SYDNEY 23

24 TABLE 3 CROSS-SECTIONAL STUDY: NURSING WORK INDEX - REVISED FACTORS Factor Possible Sample Items from NWI-R Score Range * Autonomy 6-24 Freedom to make important patient care and work decisions Not being placed in a position of having to do things that are against my nursing judgment A nurse manager or supervisor who backs up the nursing staff in decision making, even if the conflict is with a physician Control Over Practice Nurse-Doctor Relations 7-28 Adequate support services allow me to spend time with my patients Enough time and opportunity to discuss patient care problems with other nurses Patient care assignments that foster continuity of care 3-12 Collaboration between nurses and physicians A lot of team work between nurses and physicians Physicians and nurses have good working relationships Leadership A nurse manager or immediate supervisor who is a good manager and leader Support for new and innovative ideas about patient care A clear philosophy of nursing that pervades the patient care environment Resource Adequacy 4-16 Enough registered nurses on staff to provide quality patient care Enough staff to get work done * Higher scores indicate the factor was stronger TABLE 4 CROSS-SECTIONAL STUDY: ENVIRONMENTAL COMPLEXITY SCALE FACTORS Factor Re-sequencing of work in response to others Unanticipated changes in patient acuity Composition and characteristics of the care team Possible Score Range * * Higher scores indicate the factor was stronger Sample Items from ECS 0-10 Clarifying doctor's orders Medications, supplies and narcotic keys missing Completing work of others 0-10 Stat blood work Extra vital signs Greater demand for psychosocial support for patient 0-10 Students on the unit required supervision and assistance Students wanted access to charts, equipment and supplies Scheduled unit staff absent this shift 24 INTRODUCTION

25 TABLE 5 CROSS-SECTIONAL COMPONENT: OUTCOME DEFINITIONS Item Falls with Injury Falls without Injury Falls (any) Medication Errors with Patient Consequence Medication Errors without Patient Consequence Medication Errors (any) Time-based Medication Error Definition The patient experienced a fall occasioning an injury The patient experienced a fall without injury The patient experienced a fall, with or without injury (ie number of patients who experienced any type of fall ) The patient experienced a nurse medication error that occasioned adverse consequences The patient experienced a nurse medication error without adverse consequences The patient experienced a nurse medication error with or without adverse consequences Medication delivered more than 30 minutes outside the prescribed time TABLE 6 CROSS-SECTIONAL COMPONENT: DATA ANALYSIS TIME PERIOD DEFINITIONS Item Ward Ward Day Shift Shift-period Definition Data for the sample period from a single hospital ward Sample period = 5 days: Monday-Friday Data for a 24 hour period from a single hospital ward ECS and related data collected per (self-reported) shift Three equal shift-periods calculated from ward staffing (roster) data: (Morning); (Evening); (Night) TABLE 7 CROSS-SECTIONAL COMPONENT: OTHER DEFINITIONS Item Staffing hours Hours of nursing care required Definition Data from the ward roster for the 24 hour period, excluding leave and other hours off-ward (see also collection form page 151) Hours of nursing care needed per patient for the next 24 hours; data collected by trained data collectors with the validated PRN-80 instrument (see also Table 13 Instruments, page 39) TABLE 8 CROSS-SECTIONAL COMPONENT: PROPORTION HOURS WORKED GRADE CATEGORIES Item RN RN L1 RN L2 EN AIN Definition Registered Nurse: Sum of RN L1 & RN L2 Hours Registered Nurse Level 1 Hours Registered Nurse Level 2 Hours Enrolled Nurse Hours (levels not differentiated in all ward roster data) Assistant in Nursing Hours UNIVERSITY OF TECHNOLOGY, SYDNEY 25

26 TABLE 9 STATISTICAL TERMS Term Probability Estimate/Statistical Significance Correlation Coefficient Regression Regression Coefficient Beta (β) Weight R 2 Value/Adjusted R 2 Value/Pseudo R 2 Value Cronbach's Alpha (α) Description* Significance is the percent chance that a relationship found in the data is random. A probability estimate of 0.05 = 5% chance. Lower values indicate a lower chance of a random relationship. Correlations measure how variables are related. Values range from 0 (no or random relationship) to 1 (perfect relationship: "The more the x, the more the y, and vice versa.") or -1 (perfect negative relationship: "The more the x, the less the y, and vice versa."). It is a symmetrical value, not providing evidence of which way causation flows. Regression is used to account for (or predict) the variance in a dependent variable, based on combinations of independent variables. Multiple regression can establish that a set of independent variables explains a proportion of the variance in a dependent variable. Logistic regression is a form of regression used when the dependent variable is dichotomous. The average amount the dependent variable increases when the independent variable increases one unit and other independents are held constant. The larger this coefficient the more the dependent variable changes for each unit change in the independent. If all independent variables are measured on the same scale then regression coefficients are directly comparable; but if not then beta (β) weights may be calculated. The average amount the dependent variable increases when the independent increases one standard deviation and other independent variables are held constant. They display the relative predictive importance of the independent variables. Betas weights reflect the unique contribution of each independent variable, but do not account for the importance of a variable which makes strong joint contributions to the regression model. R 2 is the percent of the variance in the dependent variable explained uniquely or jointly by the independent variables (i.e. the model overall). A large value indicates that a large fraction of the variation is explained by the independent variables. Adjusted R 2 is a conservative reduction to R 2. It adjusts for the effect of a large number of independent variables that may artificially increase R 2. Pseudo R 2 provides an approximate measure of the explanatory power of Poisson regression models used in this analysis. Not considered equivalent to R 2 or Adjusted R 2. A commonly used measure of scale reliability. Higher values are better. Values above 0.70 are acceptable in the social sciences. -2 Log Likelihood Value Measure of goodness of fit. Used to assess the relative fit of each regression model. * (Garson, 2005; Goldstein, 2003; Sheshkin, 2000) 26 INTRODUCTION

27 Literature Review The current nursing shortage in Australia has been well documented (AHWAC, 2002a, 2002b, 2004). In 2006, estimates of up to 12,270 new nurses were needed to enter the profession to keep up with health care needs (AHWAC, 2004), and a shortfall of approximately 40,000 nurses is expected by 2010 (Access Economics, 2004b; Karmel & Li, 2002). This scenario will likely be detrimental to patient outcomes and nurse turnover rates as workloads increase, job satisfaction rates decrease and nurses find alternative employment (Duffield, O'Brien-Pallas, & Aitken, 2004). In light of these projections it is becoming more important to employ strategies to help retain nursing staff by addressing issues of work environment, skill mix, workload, job satisfaction, and the relationship between these and patient outcomes. Without efforts to sustain the existing nursing workforce, attempts to recruit more nurses will likely be short-lived and unsuccessful. Nursing work has changed considerably in recent years and a range of factors have been identified which impact on nurses workload. These include an increased ageing population (including both nurses and patients), increased patient acuity, new diseases, treatments and technologies, and changing employment patterns (AIHW, 2005; Karmel & Li, 2002). Nurse managers have had to become more creative in staffing and patient allocations to try to maintain standards of care and positive patient outcomes as skill mix and the workforce profile have changed. Skill mix The different categories of health care workers who provide care to patients is termed skill mix or staff mix (McGillis-Hall, 1997). Skillmix is defined as the proportion of registered nurses to total clinical nurse staffing (Aiken, Sochalski, & Anderson, 1996; Shullanberger, 2000). It is argued that a lesser qualified skill mix may result in increased nurse turnover and unproductive time (Orne, Garland, O'Hara, Perfetto, & Stielau, 1998), and others have tried to clarify roles of unlicensed and untrained personnel (McKenna, Hasson, & Keeney, 2004). Other large studies have found that a higher proportion of RNs on medical and surgical wards was associated with better outcomes in terms of morbidity and mortality (Estabrooks, Midodzi, Cummings, Ricker, & Giovannetti, 2005; O'Brien-Pallas et al., 2004; Tourangeau et al., 2006). Critical in these is the proportion of registered nurse hours worked as compared to other categories of employee regulated nurses such as enrolled nurses or licensed UNIVERSITY OF TECHNOLOGY, SYDNEY 27

28 practical nurses or unregulated workers such as health care assistants, assistants in nursing. Work environment There is increasing emphasis on the work environment of nurses because of its potential in retaining nurses and ensuring positive patient outcomes. Many years ago in the United States (USA), a number of hospitals were labelled Magnet institutions good places for nurses to work. Nurses in these facilities were deemed central to the hospital and as a result of this philosophy, had higher job satisfaction and retention rates (Kramer & Schmalenberg, 1991). These institutions were found to have a 4.6% lower patient mortality when compared with non-magnet hospitals (Aiken, Smith, & Lake, 1994). A more recent study also found that attractive organisational characteristics are key factors in nurse retention. An increased workload and having to leave basic nursing tasks undone were also found to be fundamental to nurses levels of job satisfaction and retention rates (Aiken et al., 2001). A collegial working environment, opportunities for nurse education, a richer skill mix and continuity of care have also been linked to lower patient mortality levels (Baumann, O'Brien-Pallas et al., 2001; Estabrooks et al., 2005). Nurses job satisfaction is affected by the perception of control over their work (Finn, 2001; Laschinger, Finegan, Shamian, & Wilk, 2004; Rafferty, Ball, & Aiken, 2001; Stamps & Piedmont, 1986; Tillman, Salyer, Corley, & Mark, 1997). The Nursing Work Index Revised (NWI-R), used in the ACT study, is a measure of the work environment. It has 49 items that measure nurse autonomy, control over practice, nurse-doctor relations, nursing leadership and resource adequacy. The NWI-R was first developed in the US and has since been refined and used widely including in Australia (Aiken & Patrician, 2000; Aiken & Sloane, 1997; Aiken et al., 1994; Estabrooks et al., 2002; Kramer & Hafner, 1989). Also used in this study was the Environmental Complexity Scale (ECS) (O'Brien-Pallas, Irvine, Peereboom, & Murray, 1997) used previously in Australia (Duffield et al., 2007). This tool has three domains: resequencing of work in response to others requests; unanticipated changes in patient acuity; and characteristics and composition of the caregiver team. Nurses are also asked whether nursing interventions were left undone or delayed due to lack of time. Use of both of these tools provides a comprehensive measurement of nursing work and the factors impacting on it. 28 INTRODUCTION

29 Nursing care environments and the organisation of nursing care have been linked to adverse patient outcomes such as medication errors, increased length of stay and mortality (American Nurses' Association, 1997; Czaplinski & Diers, 1998; Estabrooks et al., 2005; Grillo-Peck & Risner, 1995; Needleman, Buerhaus, Mattke, Stewart, & Zelevinsky, 2002; Tourangeau, 2002; Tourangeau et al., 2006). Recent research suggests that adverse patient events and nurses emotional exhaustion are directly affected by the quality of the work environment (Laschinger & Leiter, 2006). Aiken, Clarke & Sloane (2002) report that understaffing leads to greater nursing turnover because nurses are being prevented from providing the quality of care that they wish, compromising patient care. Clarke and Aiken (2006) also argue that nurse productivity could improve if there were improved work environments. Workload In Australia, there are many ways of allocating nursing resources which are not related to types of patient or ward specialty (except intensive care and high dependency units) (Duffield, Roche & Merrick, 2006). Some measures used include nursing hours per patient day (NHPPD) (Western Australia). A nurse to patient ratio has been adopted in Victoria which is designed to promote equal workload amongst nurses (Plummer, 2005). Unruh & Fottler (2006) found this method may underestimate nursing workload, and Graf et al. (2003) suggest such a method may produce inflexibility which could exacerbate staffing and quality issues. Other methods that measure nursing workload are patient dependency or patient acuity systems. In the early 1980s in Australia, PAIS (Patient Assessment and Information System) was introduced into Victoria (Hovenga, 1996). The resources required (hours of nursing) for a given PAIS category had been developed from a number of work sampling studies and included time for administrative work and indirect nursing activities (Goodwin & Hawkins, 1990; Hovenga, 1996). These nursing activities include direct patient care and indirect nursing care such as documentation and within the PAIS model, patients are classified on a per shift, daily, weekly, monthly, random or ad hoc basis to reflect the workload at a particular point in time. Software packages, such as E-care (D. E. Goldstein, 2003) and TrendCare (Trend Care Systems Pty Ltd, 2004), involve nurses using care plans or clinical pathways, determining the time necessary for each unit of care, and establishing patient requirements from these parameters. UNIVERSITY OF TECHNOLOGY, SYDNEY 29

30 Nursing workload can be impacted by many factors such as the number of case types (Diagnostic Related Groups [DRGs]) nurses have to care for (Diers & Potter, 1997); the degree of patient turnover and churn (movement of patients between and within wards) (Duffield et al. 2007); the increased throughput of patients (Unruh & Fottler, 2006); their length of stay and acuity (Birch, O'Brien-Pallas, Alksnis, Murphy, & Thomson, 2003); and staff shortages (Buerhaus, 1997). The decreased length of patient stay in hospital and the concentration of, and increase in nursing work that this requires, has not been widely studied (Graf et al., 2003). Diers and Potter (1997) present a case study of an overspent and difficult to manage ward. It became apparent that a large number of different DRGs (casemix) contributed to the apparent disorganisation. Some studies argue for similar patient types to be organised on specialised wards to enhance expert nursing care (Aiken, Lake, Sochalski, & Sloane, 1997; Czaplinski & Diers, 1998; Diers & Potter, 1997). The argument is that it is unreasonable for nurses to be expert in all manner of patient types/specialities, and that by narrowing the demands on their expertise, they would work more efficiently and improve patient outcomes. Case mix cohorting may help managers predict nursing care requirements more efficiently, because when patient needs vary in intensity on a day-to-day basis, nurse staffing requirements are more difficult to anticipate: patient needs may not be met. The nursing work environment, and consequently nursing workload, has changed considerably over the past few years. As a result of technology and efficiency policies that target length of stay, nurses have a more complex patient load (Baumann, Giovannetti et al., 2001; Birch et al., 2003). The increased turnover of patients or churn intensifies the nursing workload further. Birch (2003) found that after hospital restructuring in Ontario (Canada) there was an increased number of severity-adjusted patients using fewer beds cared for by fewer nurses. Patient throughput increased by 12% and inpatient episodes per bed increased by over 25%. Unruh & Fottler (2006) found that patient turnover (in their sample of up to 205 hospitals) significantly increased from 1994 to 2001 and that as a consequence, staffing requirements and workload for nurses may be underestimated. Admission and discharge of patients means extra documentation, educational, general nursing and organisational duties, thereby increasing nursing workload. The movement of patients within wards is also a factor in nursing workload, and one that is harder to quantify. However some wards will have systems of management whereby it is necessary to move patients from area to area on a regular basis (eg. from high to low acute areas). Nurses are also called upon 30 INTRODUCTION

31 to assist with these when transferring patients between wards, and, depending on resources, can be required to move the bed themselves. Nursing workload can be further increased by nurses needing to accompany patients for investigations in other departments (eg. CT or MRI scans), leaving their allocated patients in the care of a colleague who already has his/her own patient load. Another factor impacting on nursing workload is a general shortage of allied health professionals in Australia (DEWR2006). This includes occupations such as physiotherapists, occupational therapists, speech pathologists, radiographers and pathologists. This shortage of staff may cause delays in patient treatment, and an increased workload as nurses try to incorporate into their day the types of care patients should ideally receive from these professionals. Patient Outcomes / Outcomes Potentially Sensitive to Nursing (OPSN) Nurses are the health professionals that are most directly involved with patients. They monitor patients progress, assess clinical changes, intervene when appropriate and are central to communication and coordination among the allied health team. Patient safety has been defined as freedom from accident, or, more broadly, avoiding injuries to patients from the care that is intended to help them (IOM, 1999, 2001). Ingersoll (1998) defined patient outcomes as the end result of treatment or care delivery. Outcomes Potentially Sensitive to Nursing (OPSN) have been the focus of a number of studies (Buerhaus, 1999; McCloskey & Diers, 2005; Needleman et al., 2002; Needleman, Buerhaus, Mattke, Stewart, & Zelevinsky, 2001). Needleman et al. (2001) found that lower levels of RNs were linked to higher rates of urinary tract infections, pneumonia, shock and cardiac arrest, upper gastrointestinal bleeding, failure to rescue (FTR), and length of hospital stay in both medical and surgical patients treated in hospitals. FTR has been suggested as a better gauge of care quality than complications alone (Clarke & Aiken, 2003), the term having been introduced by Silber et al. (1992) to describe how patients are rescued from events that complicate their health by nurses and other health care professionals. FTR is operationally defined as death following adverse events such as sepsis, DVT, GI bleeding, cardiogenic shock and hospital-acquired pneumonia. UNIVERSITY OF TECHNOLOGY, SYDNEY 31

32 The meta-analysis by Kane et al. (2007), established that an increase in RN staffing was associated with a reduction in patient mortality, adverse events and FTR. This study found that for surgical patients, an increase of one full-time RN a day was associated with a reduction in the relative risk of FTR, and nosocomial bloodstream infections. Similarly, in intensive care facilities, a similar increase in staffing consistently decreased rates of cardiopulmonary resuscitation, unplanned extubation, pulmonary failure and nosocomial pneumonia. In the USA, the Nursing Care Report card (1997) was developed to monitor nursing care in acute care settings. It was based on data collected by state agencies in 1992 and 1994 from 502 hospitals in California, Massachusetts, and New York. The purpose of the study was to quantify nurse staffing, patient incidents, and lengths of stay at the hospitals, as well as the relationship between these variables. Upon evaluation, the American Nurses Association (1997) found that preventable conditions, such as pressure ulcers, pneumonia, post-operative infections and urinary tract infections were inversely related to RN skill mix and nurse staffing. Similar results were found by Kovner & Gergen (1998) and more recently, Cho et al. (2003). The Institute of Medicine (2004) suggested that lower levels of nursing staff (especially RNs) are related to increases in length of stay, hospital acquired infections and the incidence of pressure ulcers. Tourangeau (2006) found that by increasing the percentage of RNs by 10%, there were six fewer deaths for every 1000 discharged patients. In New Zealand an increase in the percentage of RNs together with a decreased number of nursing hours per patient per day increased negative patient outcomes (McCloskey & Diers, 2005). Recently there has been a subtle change in language from OPSN to nursing (or nurse ) sensitive outcomes. This originated in the USA and is now seen to be the accepted term (Kane et al., 2007; Person et al., 2004). Accurate information about safe and optimal ward staffing catering to different patient types is only possible on a ward-level shift-by-shift basis. Study at this level gives a clearer understanding of the ward environment and its effect upon nursing practice and patient outcomes. To date, few published studies have been based at the ward level (Boyle, 2004; Diers, Bozzo, Blatt, & Roussel, 1998; Diers & Potter, 1997). 32 INTRODUCTION

33 Context The ACT is the smallest of Australia's six states and two territories, but has the highest population density and is the only state or territory without a sea border. At 30 June 2006, the Australian Capital Territory (ACT) had an estimated resident population of 334,200 persons, with the majority residing in Canberra and nearby surrounds. The Canberra-Queanbeyan Statistical District had a population of 381,400 persons at June This is 1.8% of Australia's total population making it the eighth largest major population centre in Australia, larger than the capital cities of Hobart and Darwin (Australian Bureau of Statistics, a, b). Public in-patient hospital services in the ACT are provided at The Canberra Hospital and Calvary Public Hospital. In-patient hospital services for private patients in the ACT are provided by Calvary Private Hospital, John James Memorial Hospital and the National Capital Private Hospital. According to the Australian Institute of Health and Welfare (AIHW, 2007), there were 72,136 public hospital separations in the ACT during , 1.6% of the nearly 4.5 million public hospital separations nationally. UNIVERSITY OF TECHNOLOGY, SYDNEY 33

34 2. Study Design & Ethics Approval Study Design The study was designed to include both longitudinal data extracted from administrative data systems for the two year period and cross-sectional data collected within this time frame for medical and surgical wards in ACT hospitals. The longitudinal component of the study included: Patient data extracted from the ACT Administrative Data System for two years ( ) Nursing payroll (workforce) data where possible for the same years and hospitals. These data allow the determination of the relationship of nursing resources as paid hours worked, to patient outcomes as Outcomes Potentially Sensitive to Nursing (OPSN) (Needleman et al., 2001) controlling for casemix as AR-DRGs and hospital type. The nursing payroll data allow specification of nursing resources by skill mix, in the context of patient load as case type, patient volume, ward type (medical-surgical or other). The key aspects of the data collected from the cross-sectional sample of hospital wards are: Ward organisation/environmental characteristics Nursing workload and environmental complexity Nurse outcomes as intent to stay/leave present job or the profession Patient characteristics Patient outcomes as adverse events that cannot be captured in administrative data (falls with and without injury and medication errors with and without consequences). The use of two compatible methodologies provides a powerful design in which the known inadequacies of administrative data can be balanced by the cross-sectional data collection and the known issues of labour intensive but small sample data collection 34 STUDY DESIGN & ETHICS APPROVAL

35 can be informed by the use of large, longitudinal datasets (Jiang, Stocks, & Wong, 2006). A conceptual model based in General Systems Theory guided the study. The model is presented in Appendix 1. Both a process and an outcome approach were taken in the study. Ethics Approvals Ethics approval was sought and gained from the Human Research Ethics Committee, University of Technology, Sydney, from ACT Health and Community Care Human Research Ethics Committee, and from Calvary Health Care ACT Human Research Ethics Committee. Approval from all committees included cross-sectional and longitudinal components of the study. Participants were assured that no individual or ward would be identified in any report or publication derived from the study, although it is not possible to disguise the two participating hospitals completely. Where data were analysed and reported at ward level, wards were deideintified using alphanumeric codes. UNIVERSITY OF TECHNOLOGY, SYDNEY 35

36 3. Samples and Data Collection Longitudinal Component The data sources for this study were owned by the two ACT hospitals involved; The Canberra Hospital and Calvary Public Hospital. Data on patients were held by ACT Health as part of its mandatory hospital morbidity collection and patient level ward history data. Data pertaining to the nursing workforce, specifically nurse rostering and payroll data were held by the two individual hospitals. Data of the two types were received from both hospitals for the period Aug/Sep 2004 to Oct/Dec 2006, inclusive. Data were available for a total of 398 ward months. Details of the patient (Table 10) and nursing (Table 11) data sample are presented below (see also Table 31, page 62). TABLE 10 LONGITUDINAL COMPONENT PATIENT DATA Hospital Separations ALOS on Sample WARD (hrs) ALOS in HOSP (hrs) Total Patient Hours Ward per Episode (churn) ,744, ,469, TABLE 11 LONGITUDINAL COMPONENT NURSE DATA Hospital No. Nursing Shifts RN Hrs EN Hrs AIN Hrs Total Nursing Hours , ,136 1,146, ,727, ,939 89, , ,242 Cross-sectional Component Sixteen medical-surgical hospital wards consented to participate, 12 from The Canberra Hospital where 158 nurses participated, and four at Calvary Hospital where 42 nurses participated in the study (see Table 12). No data were collected from obstetric, paediatric or psychiatric wards, nor from ED or outpatient areas or theatre. 36 SAMPLES & DATA COLLECTION

37 Data collection commenced on the 20 September 2006 at The Canberra Hospital and 14 November at Calvary Hospital, and was completed by 12 October and 30 November 2006 respectively. Five experienced nurses were seconded from the hospitals under study, and were trained to undertake data collection with support from UTS staff. No eligible wards declined the invitation to participate. Each ward had one week of data collection randomly assigned within the sampling period allocated for each hospital. Orientation sessions were held with each ward in the week before data collection and nurses consent obtained. Staff unable to attend and casual or agency staff were given an information sheet, consent form and copy of the survey to complete and return to a marked box at the nurses station or by reply-paid post. Nurses were given a study identification (ID) number. All nurses on the 16 nursing wards selected were invited to participate. The Nurse Survey captured information on nurse demographics, the work environment and organisational attributes. At the end of each shift, nurses were asked to complete the Environmental Complexity Scale which acquired information on ward factors that influence nurses ability to provide the required care for patients, in addition to details of nursing interventions delayed or not done and indirect care activities. The data collector completed the PRN-80 form which measured patient acuity daily for each patient on the ward. This instrument lists nursing interventions that nurses complete during a 24 hour period. This instrument provided the total minutes of care (later converted to hours) required for that patient for the coming 24 hour period. The data collector or the Clinical Nurse Consultant (CNC) completed the Daily Unit Staffing Profile and Unit and Hospital Profile, providing roster data and information on the ward. Table 13 (page 39) lists the instruments used in the cross-sectional part of the study along with their psychometric properties and where appropriate, inter-rater reliability (see also Appendix 7). Table 12 outlines the details of cross-sectional data collection, and the number of responses for each instrument. Two wards were not able to provide complete roster data for the sample period, and one ward did not provide a unit profile. These wards were omitted from description or analyses requiring those data. However, in order to provide as complete a report as possible, data were included where available. The number of wards used for each analysis or description is indicated. UNIVERSITY OF TECHNOLOGY, SYDNEY 37

38 TABLE 12 CROSS-SECTIONAL COMPONENT DATA COLLECTION Instrument * Collection Frequency Response Nurse Survey: Revised Nurse Work Index Scale (NWI-R); Nurse Demographics & Work Environment Environmental Complexity Scale (ECS); Nursing interventions delayed or not done, and indirect care activities Ward Staffing Form Once per nurse Once per nurse per shift Once per ward-day 200 nurses (71% of all consenting nurses) (158 [75.2%] Canberra Hospital 42 [58.3%] Calvary Hospital) 612 shifts 14 wards, 67 ward-days, 1292 shift-periods Ward Adverse Events Profile Once per ward 16 wards Unit & Hospital Profile Once per ward 15 wards Patient Data Form Once per patient 601 patients Workload Measurement (PRN 80) Once per patient-day 1768 patient-days * See also Table 13, and Appendix 7 38 SAMPLES & DATA COLLECTION

39 UNIVERSITY OF TECHNOLOGY, SYDNEY 39 TABLE 13 INSTRUMENTS Patient Characteristics & Outcomes Instrument Details Present study statistics Source Revised Nurse Work Index Scale (NWI-R) Nurse Demographics & Work Environment Ward Staffing Form Ward Adverse Events Profile Unit & Hospital Profile Patient Data Form Identifies organisational attributes leading to positive patient, nurse and institutional outcomes. The four sub-scales of the NWI-R and their reliability are: nursing unit-nurse autonomy (Cronbach s alpha = 0.85), nurse control (0.91), nurse physician relations (0.84) and organisational support (0.84), with overall (aggregated) scale reliability of 0.96 (Aiken & Patrician, 2000). Units with higher subscale scores demonstrate higher patient satisfaction, lower mortality rates, lower nurse emotional exhaustion, and lower incidences of needlestick injuries (Aiken et al., 1997). Measures nurses perceptions about their work environment and the quality of care on the unit. It also measures demographics, job satisfaction and intent to leave. This instrument (adapted from Aiken et al., 2001; O'Brien-Pallas, Doran et al., 2001) allowed us to examine links between nurse staffing, workload and types of nursing activities. Used to record nurse staffing, and skill mix on each unit every shift each day during the sampling period. Key variables include: patient census, number/mix of staff working, number of agency/casual staff, nurse absenteeism, number of staff floated to/from the unit, number of staff on orientation, and nurse patient ratios. Number of medications given 30 minutes outside prescribed time. Information on hospital/unit size, use of clinical pathways and standard nursing care plans, presence of an educator, and hours of cleaning/clerical/auxiliary support available to the unit. The specific medical conditions creating the demand for nursing care and the outcomes of that care. Key variables include primary and secondary diagnoses and the medical condition most responsible for hospital stay. Since AR-DRGs are not assigned until after hospital medical records coding and patient discharge, patient records were matched to HIE data after the longitudinal data were acquired. Cronbach s alpha: Autonomy (0.63); Control over practice (0.69); Nurse-doctor relationships (0.67); Leadership (0.80); Resource adequacy (0.71). Nurse survey, administered once to each nurse in the sampled units Ward rosters retrieved by data collectors Adverse events reporting system on the unit Ward CNC by interview Patient record accessed by data collectors; supplemented by HIE data.

40 40 SAMPLES & DATA COLLECTION Nursing Workload Instrument Details Present study statistics Source PRN Workload Measurement (PRN 80) Environmental Complexity Scale (ECS) See also Appendix 7: Instruments Lists 214 indicators or tasks nurses complete for patients during a 24-hour period. Each indicator has a standard point value reflecting time involved completing tasks for patients; each point represents 5 minutes, and a higher total point value indicates greater amounts of nursing care required. PRN methodology has been tested extensively with several iterations since its development in 1972, and its content validity has been established by nurse experts. Chagnon et al. (1978) established the construct and predictive validity of the PRN. Recent work (O'Brien-Pallas et al., 2004) found no significant differences in workload estimates between the PRN-80 and other established systems (Grasp and Medicus), providing further support for its reliability and validity. Measures tensions nurses experience in providing care to patients to a standard outlined in nursing care plans. It taps three domains: unanticipated delays in response to others leading to re-sequencing of work; unanticipated delays due to changes in patient acuity; characteristics and composition of the caregiver team (O'Brien-Pallas et al., 1997). O Brien-Pallas et al. (2002), found Cronbach s alpha for each subscale of: 0.80 for unanticipated delays and re-sequencing of work; 0.85 for changes in patient acuity; and 0.92 for composition and characteristics of the care-giving team. This instrument also collects information per nurse-shift on the quality of care, nursing interventions delayed or not done due to time pressures, and indirect care activities. Inter-rater reliability: 87.8% Cronbach s alpha: Re-sequencing of work (0.68); Unanticipated changes in patient acuity (0.80); Composition and characteristics of the care team (0.61). Patient record accessed by data collectors Nurses on sampled wards, once per nurse per shift

41 Data Analysis Longitudinal Analysis The aim of this research was to study the relationship between nursing inputs and patient needs (e.g. nursing workload) with a focus on outcome measures as a means of assessing the adequacy of care. The data were longitudinal, allowing assessment of variation in the relationship over time and hence an assessment of the relative adequacy of nurse staffing levels at various times during the study period. It related to two public institutions and a number of ward areas in each, allowing a degree of generalisation to a range of circumstances arising on a ward. The methods used in the research employ controlling for workload (through AR- DRG casemix and activity variables) and then reviewing the impacts of staffing level. That is, it considers the impact of changes in staffing and skill-mix relative to a fixed workload. However it also offers a method for determining what staffing has been more or less successful for a given workload from a range of workloads encountered during the study period. Data Preparation Two types of patient data were requested from ACT Health. The first were coded morbidity records at patient episode of care level. These data, known as admitted patient care data, were provided in the format shown in Appendix 2. These gave data elements such as hospital of treatment, start and end dates and times for the episode of treatment, basic demographic information on the patient, along with diseases and procedures as coded under the Australian version of the International Classification of Diseases (ICD-10-AM) 5th Ed. and the Australian Classification of Health Interventions (ACHI) 5th Ed. respectively. In addition, information was available on mode of separation/type of ending of episode. The data also uniquely identified each episode of care without identifying the patient. The second type of patient data, termed ward history data, was provided in the format shown in Appendix 3. These identified ward area, start and end times and a unique morbidity data identifier of every patient having contact with the ward (and its staff). It should be noted that short absences from the ward do not generate new ward episode data, however prolonged absences such as visits to theatre and recovery, do. UNIVERSITY OF TECHNOLOGY, SYDNEY 41

42 All records in the admitted patient care data were linked to the ward episode data to provide a detailed ward history of the patient. Data on nurse rostering and payroll for particular wards were provided by the two study hospitals. These came from the computerised nurse rostering systems (PROACT) in two formats, both reflecting the actual assignment of nurses to ward areas rather than the planned assignment. The roster data included information on the skill level of each nurse on a shift as well as their start and finish times. The nursing data and patient data were then linked by ward to provide a detailed patient and nurse profile for the ward. Although the ward identifiers used in the nursing data were not a direct match to the ward identifiers used in the ward history collection, links could be made between the two. These links were either made or confirmed by the staff of the hospitals, project staff in the field or information systems staff in ACT Health. The links settled on are in Appendix 4. There was an inconsistency in the data as originally matched, which was resolved by combining two ward areas (ward codes 1AF & 1AI in The Canberra Hospital). The roster data reflected the shifts of nurses working on a ward during a given pay period. However, for both the staffing and patient data, the focus of the study was the wards and the events occurring there. Therefore the data were reorganised to be a sequence of events of specified nature occurring at a specified time on the ward, for example, the commencement of a shift by a RN qualified staff member or the transfer to the ward of a patient in a particular AR-DRG with a particular number of hours already spent in hospital. These reorganised data are referred to technically as transaction records, but we treated and referred to them as Time Series. Time series data allowed the construction of measures that could be used to assess changes in workload. This included cumulative patient hours spent on the ward, and patient hours spent in hospital before admission to the ward, or after discharge from the ward. Similarly, for nursing data, measures included a cumulative count of nurses being rostered on and off the ward, as well as the number of hours worked by the nurses. Patient data covered a wider range of wards (n = 76), compared to the nursing data (n =15). All wards from the nursing data were matched to corresponding patient data. Data from wards 1AF and 1AI (Hospital 82) were combined and treated as a single ward. In total a full nursing and patient profile was able to be provided for 14 wards areas listed below. 42 FINDINGS

43 TABLE 14 HOSPITALS AND WARDS PROFILED Hospital Code Roster Ward WARD Start Date WARD End Date 82 1AA 09/09/ /02/2007* 82 1AB 09/09/ /02/2007* 82 1AD 09/09/ /02/2007* 82 1AF & 1AI 09/09/ /10/ AG 09/09/ /10/ AH 09/09/ /10/ AK 09/09/ /10/ AL 09/09/ /02/2007* 82 1AM 09/09/ /02/2007* 82 1AO 09/09/ /02/2007* 83 2AC 26/08/2004 7/03/2007* 83 2AE 26/08/2004 7/03/2007* 83 2AJ 26/08/2004 7/03/2007* 83 2AN 26/08/2004 7/03/2007* *cut off at 31/12/2006 Matched nurse and patient data relating to a ward covered approximately 2.5 years. The exact periods are shown in Table 14 above. It was found in the patient records that the majority of data with separation date after the 31/12/2006 were not yet coded, therefore the cut off point for both nursing and patient data became 31/12/2006. Data Considerations and Controlling for Workload Patients commonly make contact with more than one ward area during a hospital episode and indeed often have multiple hospital stays during a 2.5 year period. These multiple contacts result in repeated measures on the same patient. In our time series analysis we have ignored the presence of multiple hospital stays (in common with most large dataset studies) and have used the patient s AR-DRG and prior hospital stay (in hours) to reduce the interdependence of their consecutive ward episodes. However the dependency that remains cannot be ignored over short periods. Therefore the study elements were chosen to be 28 day (roster period) segments of the time series of each study ward. The patient and workforce data for these roster periods (ward months) were then linked and records not overlapping the study the period discarded. The final data were a full patient and nursing profile (by ward month) for each of the 14 wards. The data used in this study have a limitation that potentially affects the strength of effects found. It is that the adverse events data (captured in the morbidity record) is at UNIVERSITY OF TECHNOLOGY, SYDNEY 43

44 hospital episode level and does not attribute an event time or place. Therefore such occurrences were attributed to a ward area in proportion to exposure. We felt biases could arise through the transfer of injured patients from one ward area (for example a short stay ward) to another ward area where they recovered. Therefore we controlled for ward workloads during the contact period and placed the staffing in the role of experimental variable. The controlling approach used was based on clusters methodology. There were two matchings of ward month used. The first, the load cluster, was based on the profile of the ward months measured through: Total patient hours for each AR-DRG Total admissions to ward for each AR-DRG Total hours in hospital before admission to ward for each AR-DRG Total patient hours (a redundant variable used for consolidation) Total ward separations The second clustering was by assess cluster which matched ward months on a profile of: Total admissions to ward for each AR-DRG Total hours in hospital before admission to ward for each AR-DRG These methods produce relatively similar clusters of wards by the clinical characteristics embedded in AR-DRGs and are therefore a form of risk adjustment. Both these matchings ignore the size of the wards; they only use the patterns in the profile variables. Other statistical controlling techniques, such as linear regression and casemix index methodology, were used within clusters to strengthen the analyses reported below. The Outcome Measures A recent development in the nursing literature has been the adoption of statistical measures referred to as Outcomes Potentially Sensitive to Nursing (OPSN). The OPSN algorithms were originally developed by Needleman and Buerhaus (2001). Dr Barbara McCloskey developed the cross walks from the American ICD-9 to the Australian/NZ ICD-10 for use with New Zealand data (McCloskey and Diers, 2005). The OPSN definitions can be found in Appendix 5. Mapping tables from the National 44 FINDINGS

45 Centre for Classification in Health were used to find the comparable outcome codes. NZ other exclusions were used and Version 3.1 AN-DRGs were mapped to AR-DRG Version 5.1 on the basis of the Grouper logic (Laeta Pty Ltd is a Commonwealth Certified Grouper Developer). Workforce (nursing hours by skill level) was then correlated with outcomes potentially sensitive to nursing (OPSN) whilst controlling for caseload (patient hours on wards by case-type and other features). The method used to control for caseload was the combination of DRG casemix and matching through clustering of ward months with like patient profiles discussed under Data Considerations above. Interpretation of the results of OPSN analyses requires familiarity with the data and methods used. Therefore we draw an extract from our earlier report to NSW Health to explain the standard approach (see Duffield et al. 2007, pp.43-44). The episodes of care were compared with the criteria found in Appendix 5, defining Outcomes Potentially Sensitive to Nursing (OPSN) that are reasonably well supported by administrative collections such as the ACT Health admitted patient care data. The work by Needleman and Buerhaus (2001; 2002) and McCloskey and Diers (2005) has led to the development of the following measurable concepts. TABLE 15 OUTCOMES POTENTIALLY SENSITIVE TO NURSING Code OPSN 1 Urinary Tract Infection 2 Decubitus 3 Pneumonia 4 Deep Vein Thrombosis/Pulmonary Embolism 5 Ulcer/Gastro-Intestinal Bleeding 6 Central Nervous System Complications 7 Sepsis 8 Shock/Cardiac Arrest 9 Surgical Wound Infection 10 Pulmonary Failure 11 Physiological/Metabolic Derangement 12 Failure to Rescue* * Deaths following sepsis, pneumonia, GI bleeding, or shock All definitions are subject to the following filter (exclusion rules) on records, and these apply to all comparator sets and records counted to form denominators in rates: UNIVERSITY OF TECHNOLOGY, SYDNEY 45

46 MDC = 14,15,19 or 20 (maternity, newborn, mental illness, substance abuse) Paediatrics (i.e. age <18) LOS < 1 day LOS > 90 days Error DRGs (i.e. DRG = 961Z, 962Z, 963Z) Each OPSN category is supported by a list of ICD-10 diagnosis codes (some also include ACHI surgery codes) for inclusion of cases, and a set of exclusion rules that apply to both the codes selected for the presence of codes and those in scope of the concept (denominator). For example Category 1, UTI, is defined as either diagnoses N39.0 or T83.5 or as a secondary diagnosis (but not as a primary diagnosis) and the case is not grouped to any of MDC 11 through to MDC 15 inclusive nor to MDC19 or MDC 20 (mental illness and substance abuse), and nor is the rubric of the principal diagnosis A40, or A41. Another simple OPSN is Category 9, surgical wound infection, where either of the diagnosis codes T79.3 or T81.4 appears as a secondary diagnosis, but neither as a principal diagnosis gives membership of the category. See Appendix 4 for detailed definition of category membership. The denominators used to form the rates for either of these indicators are the count of cases restricted to the same set of MDCs and with the principal diagnosis being other than one excluded by the OPSNs definition. In practice, the both the Numerator and Denominator counts are restricted to being either of medical or surgical DRGs and a medical and a surgical version of the OPSN is produced. Failure to rescue (FTR) is death following an adverse event of sepsis, pneumonia, GI bleeding, or shock (Silber et al., 1995; Silber et al., 1992), Therefore the denominator is the count of these particular adverse events. OPSN have been investigated in a number of studies (Beurhaus, 1999; McCloskey & Diers, 2005; Needlemen et al., 2001, 2002). They were also investigated in the recent NSW report by our team. The analysis of OPSN is complicated in these data, and in general, because the measures were initially intended to be applied at hospital level to quite similar hospitals, or the same hospital over a number of time periods. However we bring the analysis to bear on the units of our study, ward months. One of the most immediate consequences of this shift is that the casemix seen on a ward will affect the rate of adverse outcomes in an unbalanced way. Therefore we used the load cluster to match ward months. 46 FINDINGS

47 Another technical complication arises in the analysis of OPSN because the rates of these events in a typical ward over a 28 day period are numerically low, so that the counts of events do not suit Analysis of Variance based on the Normal Distribution. The Statistical literature contains a number of relevant examples of analyses of counts data based on Generalised Linear Modelling with Poisson distribution. In particular SPSS Version 15.0 has implemented the approach so that it could be applied to our OPSN data. We needed to replace the OPSN values by their nearest integer value because the Poisson method expects count data. OPSN analyses were performed using Generalised Linear Modelling with Poisson distribution. A range of different models were tested using the following factors: Cluster Cluster, NH:PH Cluster, RN:PH, EN:PH Cluster, RN:NH Cluster, RN:NH, NH:PH Cluster, RN:NH, RN:PH Cluster, RN:NH, RN:PH, EN:PH Where Cluster = group which the ward month falls into dependent upon the number of hours of care by each AR-DRG etc NH = total nursing hours PH = total patient hours RN: total hours worked by Registered Nurses EN: total hours worked by Enrolled Nurses The best model for each individual OPSN was selected dependent upon the significance of the Omnibus test, and Model Effects Type III Chi-Square results (produced by SPSS Version 15 (SPSS Inc., 2006)). Once the best model was chosen, the direction of the parameter estimates was noted. This indicated whether the parameter was having a positive or negative effect on the incidence of OPSN. Review of the SPSS output made it clear that the effect of rounding the OPSN may affect findings, so a subsidiary testing process was put in place. This secondary approach was guided by the standard method for testing the difference of proportions and by the Gauss Markov Theorem. We only applied it to testing for RN Proportion Effect. UNIVERSITY OF TECHNOLOGY, SYDNEY 47

48 We start by taking the underlying rate for an OPSN in a ward month to be that of its load cluster under the null hypothesis that only Cluster has an effect. This is estimated by summing the OPSN across the cluster, summing the patient hours on ward across the cluster and then dividing the former by the latter. We then predict the number of OPSN for each ward month by multiplying its estimated underlying rate by its patient hours on ward. In keeping with the standard tests of proportions we then divide each ward month s OPSN number by its predicted value. It is at this point we bring Gauss Markov and the underlying Poisson distribution to bear and weighted each ward month ratio by the square root of its predicted value. If RN proportion has no effect, each weighted ratio (GME) is an unbiased, unit variance predictor of unity. Under the null hypotheses there will be no regression of GME on RN proportion. Under the alternative there will be and negative slope will be associated with better outcomes. The actual testing process included a modification, which was to conduct the regression while controlling for cluster effects. The latter could be induced by the differing RN proportion across Cluster, and hence needed to be controlled for. An important methodological point here is that while this second approach does not take full advantage of the Poisson error distribution, use is made of Gauss Markov. Further, under the Poisson analysis our model for the parameter b is not identified: the absolute size of the anti-logged cluster effects is completely confounded with the absolute value of b. We also found it necessary to adopt some sample statistics for the cluster effects when the largest attributed OPSN count was less than 0.5 for a whole Cluster. We conducted the follow up test described above to strengthen our findings and report these results along with the formal method results. Poisson analysis allows the assessment of the statistical significance of a factor and the direction of its effect, but not a readily interpretable measure of its size. This gap in understanding needs to be filled using other methods. The follow-up testing approach assists in this but is biased by the weighting applied to form GME. In addition the clusters have different average proportions of RN hours say. However use of the General Linear Model with fixed effects of Cluster, Intercept set to zero and weighted least squares (using the expected OPSN number as weight variable) offers an approximate approach consistent with the Gauss Markov based approach. This follows from the fact that the unweighted ratios are unbiased estimators of 1 with variance equal to the inverse of the variance of the observed value. 48 FINDINGS

49 The regression slope for an experimental variable in this new type of analysis needs interpretation which we now offer. If b is the regression parameter for RN hours as a proportion of nursing hours (for example), then we see the effect of increasing the RN hours proportion by 10% as changing the rate of the OPSN by b times 10%. So if b were -3 then a 10% increase in the RN hours proportion would reduce the rate of the OPSN to 70% of its current value. In this report we extend our investigations to include length of hospital stay (LOS) as an OPSN variable. LOS is responsive to the quality of nursing care (as well as other factors) and therefore ward months associated with patients who have longer than expected stays may also be those where the quality of nursing care is lower. One of the obvious factors affecting LOS is the patient s illness and medical intervention. These are not nursing dependent. Therefore LOS as an OPSN needs to be controlled for the patient s AR-DRG V5.1. The standard approach for doing this is to form casemix indices, where the LOS performance of a particular ward is compared with that to be expected if it had the same average LOS for each AR-DRG as seen in the whole dataset. Another method for dealing with these factors is by matching ward months (through load clusters) before considering the effects of nurse staffing and skill mix. To be particularly careful, we combined these approaches and a further linear regression approach to adjust for prior exposure to risk. ALOS as an OPSN Methodology As discussed above, each ward month had been assigned to a load cluster and an assess cluster. Taking each assess cluster at a time, casemix adjusted indices for both the time spent in hospital before encountering a ward month and time spent in hospital after contact with the ward month were calculated. The use of casemix adjustment within assess cluster was to make sure LOS precursors and outcomes were being compared like with like. The next step was conducted load cluster by load cluster, thereby controlling for workload on the ward at the time of the assess cluster patient contacts. The within load cluster processing was the conduct of linear regression involving the logarithms of indices calculated in the previous step. These indices were each ward month s index for after contact hours of stay (After Index) and its index of before contact hours of stay UNIVERSITY OF TECHNOLOGY, SYDNEY 49

50 (Before Index). The regression predicted the logarithm of After Index based on the logarithm of Before Index. After the regressions had been run for each load cluster, it was possible to calculate the difference between each ward month s observed logarithm of After Index and its predicted value. These residuals are referred to as performances. The anti-logarithm of a performance provides a measure of the care hours after ward contact as a proportion of the care after contact expected in a ward in the same assess cluster, in the same load cluster, with the same casemix and the same patient pre-contact history. The methodology for assessing the effects of nursing hours per patient hour, and proportion of RN nursing care hours could thus be based on the correlations and regressions of performance on the experimental variables. It was safe to assume that the statistical dependence between the ward months performance statistics could be ignored as there were many raw data points and 398 ward months. Cross-sectional Analysis Cross-sectional data were entered into a Microsoft Access (Microsoft Corporation, 2003) database and extracted to SPSS versions 14 and 15 (SPSS Inc., 2005, 2006) for analysis. Where data were missing at the patient or nurse level, they were imputed as the ward mean calculated from the non-missing values on that ward. Where more than 10% of data were missing at the patient or nurse level, that variable was not used in regression analyses. Complete staffing data were not available on two wards. These wards were consequently excluded from analyses that used staffing data. Subscale scores and alpha reliabilities for the instruments used were generated using syntax provided from the Canadian study (O'Brien-Pallas et al., 2004). Correlation analysis (Pearson s r or Kendall s tau b [τ], depending on the nature of the data (Sheshkin, 2000) was used to explore relationships between variables at the individual and ward level. Data collected at the patient and nurse level were aggregated to ward level for some analyses, using mean values, rates or proportions. Some patient level data were converted to percentage of patients per ward, for example, adverse patient outcomes such as falls and medication errors. In similar studies, multilevel modelling (MLM) has been used for analysis of hierarchical or clustered data (Duffield et al., 2007; H. Goldstein, 2003). That approach 50 FINDINGS

51 is considered appropriate where some variables are measured at the individual level (patient or nurse) and others measured at the ward level. Data are therefore not aggregated, but rather retained at the measurement level. However, the number of wards with complete data in this study (14) does not provide sufficient statistical power to undertake this type of analysis. Data were therefore used at the most appropriate level of aggregation for each analysis. Some data from the Environmental Complexity Scale (ECS) in the cross-sectional component were further analysed at the shift-period level (see Table 6, page 25). In this case, hours of nursing care data were apportioned to three conventional time periods: morning ( ); evening ( hours) and night ( hours), using the individual nurses shift start and end times. For all regression modelling explanatory variables were added in sequence to the statistical models. The order of entry of variables into the statistical modelling process was consistent with the theoretical framework described in Appendix 1. In order to address potential multicollinearity, a univariate regression analysis on each individual explanatory variable identified all significant predictors, and a factor analysis was conducted. This identified 17 variable groupings. The significant univariate predictors were then identified within the different groups. All predictor variables for each outcome variable were put into a stepwise regression model, whereby the properties of each model were compared to the previous one using the -2 Log Likelihood value. The output for that model was then considered in terms of its position among the 17 components to ensure that any two predictor variables did not fall into the same group. In order to compare the relative contributions of the independent variables to the models, beta (β) weights were calculated. In the case of linear models, the adjusted R 2 value was also calculated to provide an estimate of overall model fit (see also Glossary, page 22). Linear regression models for tasks delayed and not done were developed with data at the ward-day level. This level of data provides outcome variables that are an aggregate of responses for that ward for that day. Analysis at this level of data for these outcomes is more meaningful as it accounts for the overall picture of the ward for a given day, and the impact of workload and other variables for that period. Analyses for the nurse outcome variables job satisfaction, satisfaction with nursing, and intention to leave the current job, were conducted with these variables measured UNIVERSITY OF TECHNOLOGY, SYDNEY 51

52 at the nurse level. Data collected at shift level (Environmental Complexity Scale) were aggregated to nurse level to permit matching with nurse data. However, not all data could be matched, leaving a reduced dataset of 149 cases. As these outcomes are dichotomous, logistic regression models were developed. In summary, longitudinal data were examined for changes in the relationship between the amount and type of nursing resources and OPSNs across the two year period, at a ward level. Cross-sectional data were analysed for relationships between variables, and models were developed to determine the variables that significantly impact on outcomes. Comparison with similar research in NSW was made where data were available, either as overall figures or by hospital grouping. Where possible in both components of the study, estimates of the strength of each model and of the relative contribution of each variable were calculated. 52 FINDINGS

53 4. Findings Longitudinal Findings Descriptive Patterns in Skill Mix TABLE 16 CANBERRA AND CALVARY TIME PERIODS Canberra Calvary Period Date Start Date End Date Start Date End 1 9/9/04 8/3/05 26/8/04 25/2/05 2 9/3/05 8/9/05 26/2/05 25/8/05 3 9/9/05 8/3/06 26/8/05 25/2/06 4 9/3/06 8/9/06 26/2/06 25/8/06 5 9/9/06 21/2/07 26/8/06 25/2/ /2/07 7/3/07 Note that calculations were adjusted for the final periods which were shorter than 6 months. Also three wards in Canberra have a final period shorter than the other wards, ending on 18/10/06 instead of 21/2/07. This has been noted under relevant tables ( Ward 1AH, Ward 1AK and Ward 1AG ). Table 17 to Table 26 show the RN and EN hours for each ward from Canberra Hospital included in the study. Table 27 to Table 30 show results for Calvary Hospital. Notes on each ward are below each ward table. A summary of how wards compare can be found in text following Table 26 for Canberra and Table 30 for Calvary. Wards are described by type as indicated (see Longitudinal Analysis, page 41 and Table 31 page 62). Three time series (1, 3, 5) cross the Christmas/January period which may impact on staffing and patient levels. UNIVERSITY OF TECHNOLOGY, SYDNEY 53

54 Canberra Hospital TABLE 17 NURSE SKILL MIX FOR CANBERRA WARD 1AB MEDICAL TYPE FROM 09/09/2004 TO 21/2/ Month Period Hours Worked No. of Personal Shifts Ratio Hours Ratio Shifts Total Hours EN RN EN RN EN 53% 50% 49% 47% 51% RN 47% 50% 51% 53% 49% EN 53% 50% 48% 46% 51% RN 47% 50% 52% 54% 49% EN/RN Table 17 shows 51% EN and 49% RN hours worked over the given time period in Ward 1AB. There is a small increase in RN and total hours worked between periods 2 and 3. TABLE 18 NURSE SKILL MIX FOR CANBERRA HOSPITAL WARD 1AL 'MEDICAL TYPE' FROM 09/09/2004 TO 21/02/ Month Period Hours Worked No. of Personal Shifts Ratio Hours EN RN EN RN EN 45% 43% 37% 36% 37% RN 55% 57% 63% 64% 63% Ratio Shifts Total Hours EN 44% 43% 36% 36% 36% RN 56% 57% 64% 64% 64% EN/RN Note that AIN worked one shift (6 hours) in Period 5. Table 18 shows a sustained, gradual increase in the proportion of RN hours worked in the Medical Type Ward 1AL throughout the whole period (from 55% to 63%), levelling off in the last three periods. The largest increase in the proportion of RN hours worked occurred between period 2 and period 3 (7%). 54 FINDINGS

55 TABLE 19 NURSE SKILL MIX FOR CANBERRA HOSPITAL WARD 1AD 'MEDICAL TYPE' FROM 09/09/2004 TO 21/02/ Month Period Hours Worked No. of Personal Shifts Ratio Hours Ratio Shifts Total Hours EN RN EN RN EN 21% 21% 19% 20% 22% RN 79% 79% 81% 80% 78% EN 20% 20% 18% 19% 20% RN 80% 80% 82% 81% 80% EN/RN Table 19 shows that there is a far greater proportion of RN hours worked on ward 1AD than both wards 1AL and 1AB above. The proportion remains steady around 22% to 78% for EN to RN hours across the whole study period. TABLE 20 NURSE SKILL MIX FOR CANBERRA HOSPITAL WARD 1AO 'MEDICAL TYPE' FROM 09/09/2004 TO 21/02/ Month Period Hours Worked No. of Personal Shifts Ratio Hours EN RN EN RN EN 36% 36% 35% 35% 36% RN 64% 64% 65% 65% 64% Ratio Shifts EN 35% 36% 35% 34% 35% RN 65% 64% 65% 66% 64% Total EN/RN Hours Note that AIN worked 14 shifts (91 hours) in Period 5 (9/9/06 21/2/07) Table 20 shows a steady proportion of 36% EN and 64% RN ratio hours in the ward 1AO over the study period. This is more than 1AB and 1AL but less than 1AD. UNIVERSITY OF TECHNOLOGY, SYDNEY 55

56 TABLE 21 NURSE SKILL MIX FOR CANBERRA HOSPITAL WARD 1AH SURGICAL TYPE FROM 09/09/2004 TO 21/02/ Month Period Hours Worked No. of Personal Shifts Ratio Hours Ratio Shifts Total Hours EN RN EN RN EN 27% 24% 23% 29% 26% RN 73% 76% 77% 71% 74% EN 25% 23% 22% 28% 25% RN 75% 77% 78% 72% 75% EN/RN Note that period 5 ends earlier than most other Canberra wards (21/2/07). Table 21 above shows a consistent 26% to 74% ratio between EN and RN staff hours worked in the Surgical Type ward 1AH throughout the study period. TABLE 22 NURSE SKILL MIX FOR CANBERRA HOSPITAL WARD 1AM 'MEDICAL TYPE' FROM 09/09/2004 TO 21/02/ Month Period Hours Worked No. of Personal Shifts Ratio Hours EN RN EN RN EN 40% 40% 34% 29% 27% RN 60% 60% 66% 71% 72% Ratio Shifts EN 40% 40% 33% 29% 27% RN 60% 60% 67% 71% 72% Total EN/RN Hours Note that AIN worked 24 shifts (330 hours) in Period 5. Table 22 shows a consistent increase in the proportion of RN to EN hours worked in the ward 1AM across the study period (from 60% RN 40% EN, to 72% RN and 27% EN) with the total number of nursing hours also increasing by 37% since the start of the period. Although there is an increase in hours, the number of hours worked by EN rises at first (period 2) and then steadily declines to be lower than the start of the study period (14829 compared to 16074). 56 FINDINGS

57 TABLE 23 NURSE SKILL MIX FOR CANBERRA HOSPITAL WARD 1AA 'MEDICAL TYPE' FROM 09/09/2004 TO 21/02/ Month Period Hours Worked No. of Personal Shifts Ratio Hours Ratio Shifts Total Hours EN RN EN RN EN 24% 28% 26% 25% 20% RN 76% 72% 74% 75% 80% EN 23% 26% 26% 24% 19% RN 77% 74% 74% 76% 81% EN/RN Table 23 shows a fair bit of instability in skill mix for ward 1AA, but a distinctly higher RN ratio (80% RN, 20% EN) in the final period. TABLE 24 NURSE SKILL MIX FOR CANBERRA HOSPITAL WARD 1AK 'MEDICAL-SURGICAL TYPE' FROM 09/09/2004 TO 18/10/ Month Period Hours Worked No. of Personal Shifts Ratio Hours Ratio Shifts Total Hours EN RN EN RN EN 34% 30% 30% 31% 32% RN 66% 70% 70% 69% 68% EN 33% 29% 29% 31% 32% RN 67% 71% 71% 69% 68% EN/RN Note that period 5 ends earlier than most other Canberra wards (21/2/07). Table 24 shows an increase in the proportion of RN hours worked between period 2 and 3 for ward 1AK, accompanied by an increase in the number of total nurse hours between period 1 and 4 (increase of 4192 hours, or 16.7%). Note that period 5 is only one month long. Overall the proportion of EN to RN hours remains steady at about 32% EN to 68% RN. UNIVERSITY OF TECHNOLOGY, SYDNEY 57

58 TABLE 25 NURSE SKILL MIX FOR CANBERRA HOSPITAL WARD 1AG 'MEDICAL-SURGICAL TYPE' FROM 09/09/2004 TO 18/10/ Month Period Hours Worked No. of Personal Shifts Ratio Hours Ratio Shifts Total Hours EN RN EN RN EN 31% 28% 29% 29% 26% RN 69% 72% 71% 71% 74% EN 32% 28% 29% 30% 27% RN 68% 72% 71% 70% 73% EN/RN Note that period 5 ends earlier than most other Canberra wards (21/2/07). Table 25 shows a steady ratio between EN and RN hours of 28.6% to 71.4% in this Medical-Surgical Type Ward. Note that period 5 is only one month long. TABLE 26 NURSE SKILL MIX FOR CANBERRA HOSPITAL WARD 1AF MEDICAL-SURGICAL TYPE AND WARD 1AI SURGICAL TYPE FROM 09/09/2004 TO 21/02/2007 * 6 Month Period Hours Worked No. of Personal Shifts Ratio Hours Ratio Shifts Total Hours EN RN EN RN EN 40% 37% 38% 35% 34% RN 60% 63% 62% 65% 66% EN 40% 36% 38% 35% 34% RN 60% 64% 62% 65% 66% EN/RN * Note that these data were combined from 2 wards in order to retain reasonable stability in the time series, so should be viewed with caution. Table 26 shows a statistically significant increase in the proportion of RN hours worked over the period of the study, but little change in the total number of nursing hours. RN hours increase over this period, while EN hours decline. 58 FINDINGS

59 Summary of Canberra Wards EN to RN Ratios Medical Type ward 1AB maintained a steady 50% EN to 50% RN ratio over the study period and ward 1AL demonstrated an increase from 55% to 63% over the study period. Medical type ward 1AD had the highest proportion of RN hours from the wards studied (20% to 80% comparing EN to RN). Ward 1AA was the highest of the remaining wards with a 25% to 75% ratio. Most of the remaining wards held a ratio between 26% to 74% and 40% to 60% of EN to RN hours. A number of wards showed increases in total nursing hours, of up to 39%. Calvary Hospital TABLE 27 NURSE SKILL MIX FOR CALVARY HOSPITAL WARD 2AJ 'SURGICAL TYPE' FROM 26/08/2004 TO 7/03/ Month Period Hours Worked No. of Personal Shifts Ratio Hours Ratio Shifts Total Hours EN RN EN RN EN 11% 9% 19% 20% 20% 11% RN 89% 91% 81% 80% 80% 89% EN 11% 9% 19% 20% 20% 10% RN 89% 91% 81% 80% 80% 90% EN/RN Note that period 6 ends earlier than most other time frames. Table 27 shows a clear decrease in the ratio of RN hours between period 2 and period 3 (from 90% to 80%) in the Surgical Type ward 2AJ. This ratio remains consistent until the end of period 5 (20% to 80%). Total hours during this time increase with additional EN and RN hours worked between period 2 and 3 and decrease slightly in 4 and 5. The final total remains higher than the starting amount. UNIVERSITY OF TECHNOLOGY, SYDNEY 59

60 TABLE 28 NURSE SKILL MIX FOR CALVARY HOSPITAL WARD 2AE 'SURGICAL TYPE' FROM 26/08/2004 TO 7/03/ Month Period Hours Worked No. of Personal Shifts Ratio Hours Ratio Shifts Total Hours EN RN EN RN EN 20% 19% 19% 23% 22% 21% RN 80% 81% 81% 77% 78% 79% EN 20% 19% 19% 23% 22% 21% RN 80% 81% 81% 77% 78% 79% EN/RN Table 28 shows a steady ratio between EN and RN hours worked of 20% to 80% over the study period for the Surgical Type ward 2AE. The greatest difference occurs in period 4 with an increase in EN hours worked of 4% proportionally. Total numbers increased by 100% over the same time, with RN and EN numbers increasing in the same proportion. TABLE 29 NURSE SKILL MIX FOR CALVARY HOSPITAL WARD 2AN 'MEDICAL TYPE' FROM 26/08/2004 TO 7/03/ Month Period Hours Worked No. of Personal Shifts Ratio Hours Ratio Shifts Total Hours EN RN EN RN EN 45% 38% 38% 55% 51% 43% RN 55% 62% 62% 45% 49% 57% EN 45% 38% 38% 56% 51% 43% RN 55% 62% 62% 44% 49% 57% EN/RN Table 29 shows a variable pattern for EN to RN work hour ratios for the Medical Type ward 2AN within the study period. Total work hours increase by over 100% during the study period but not at the same rate for EN and RN. The ratio moves from close to 45% EN to 55% RN, to almost 40% EN to 62% RN then 55% EN 45% RN in 60 FINDINGS

61 period 4, back toward 50% EN and RN in period 5 and 43% EN to 57% RN at the end of the period. TABLE 30 NURSE SKILL MIX FOR CALVARY HOSPITAL WARD 2AC 'MEDICAL TYPE' FROM 26/08/2004 TO 7/03/ Month Period Hours Worked No. of Personal Shifts Ratio Hours Ratio Shifts Total Hours EN RN EN RN EN 36% 32% 30% 33% 31% 37% RN 64% 68% 70% 67% 69% 63% EN 36% 32% 31% 33% 31% 37% RN 64% 68% 69% 67% 69% 63% EN/RN Table 30 shows a slight increase in the proportion of RN hours worked over the time of the study from 64% to 69% for ward 2AC. Most of the increase in total hours over the period is due to an increase in RN hours. Calvary Summary Both the Medical Type wards 2AC and 2AN have the lowest ratio of RN to EN hours and are the most variable over the study period, both showing a steady increase in total hours over the period. Surgical Type wards 2AE and 2AJ have the highest proportion of RN hours (20% to 80% EN to RN). All wards showed an increase in total hours over time, with the nursing skill mix ratio remaining fairly steady. UNIVERSITY OF TECHNOLOGY, SYDNEY 61

62 Patterns in Staffing Levels Staffing level is expected to change with patient acuity, including features related to age and length of stay. We would also expect the actual care hours delivered on the ward during a ward month to be the major determinant of staffing numbers. Table 31 below gives basic utilisation data on the wards studied. TABLE 31 WARD STATISTICS Hospital Ward Ward Type Separations Avg LOS on ward (days) Avg LOS in hospital (days) Avg Age (Yrs) 82 1AA Medical AB Medical AD Medical AF & 1AI Medical-Surgical & Surgical AG Medical-Surgical AH Surgical AK Medical-Surgical AL Medical AM Medical AO Medical AC Medical AE Surgical AJ Surgical AN Medical Figure 1 below illustrates the relationship between staffing numbers and patient load in the wards from Canberra Hospital. The data are shown for consecutive roster periods. Further, the methodology for investigating ALOS as a OPSN has been used to 62 FINDINGS

63 assess amount of nursing required to achieve the average level of ALOS outcome given the patient load and casemix. This level is plotted as Typical Nurse Hrs indicating that it is the level that leads to the average risk for patients of this type. There is no supposition that typical means appropriate, however the plot allows assessment of variation from the empirical norm established by the software. The accuracy of this assignment would be improved by the addition of further ward month data to the method s learning (reference) set. Figure 1 shows staffing levels were similar to typical hours over most of the period. Staffing shows a general match to patient load and acuity adjusted patient load (workload). This observation is based on the typical plot which is well matched to the actual. There has been no significant change in the workload of nurses in these wards in Canberra Hospital during the study period. FIGURE 1 CANBERRA HOSPITAL STAFFING AND PATIENTS Figure 2 shows the same measures for Calvary Public Hospital. The typical plot is increasing over time with respect to the actual nursing provided. This means the nurses workload has increased over the period in this hospital. UNIVERSITY OF TECHNOLOGY, SYDNEY 63

64 FIGURE 2 CALVARY PUBLIC HOSPITAL STAFFING AND PATIENTS Acuity, measured as the ratio of typical nursing hours to patient hours, has remained static in Calvary Public Hospital. In the Canberra Hospital there there has been a statistically significant decline in acuity over the study period, although it would take 10 years of the current trend to halve the current level of acuity. We now look at the study wards in turn. The first feature we look at is the complexity of their caseloads as measured by the number of different AR-DRGs seen during the period. Note that the figure for Ward 1AF & 1AI should be disregarded as it is an artefact of our need to combine the two areas in order to retain reasonable stability in the time series. The other data show that the wards see a wide range of casemix and hence complexity in matching care to care requirements. It also illustrates the need for casemix adjustment (of the type we have employed) in the comparative analysis of wards and even ward months of the same ward. 64 FINDINGS

65 TABLE 32 AR-DRGS CARED FOR OVER THE STUDY PERIOD Hospital Ward No DRGs seen (out of possible 613) 82 1AA AB AD AF/1AI AG AH AK AL AM AO AC AE AJ AN 188 We have devoted Appendix 6 to plots for each study ward. The plots show the same measures as used in Figure 1 and Figure 2, and so allow demonstration of the changes in acuity adjusted workload, patient load and nurse staffing level. We note that there are significant differences in (acuity adjusted) staff to patient ratios between wards. We accept that part of the explanation of ward level variation in acuity adjusted staffing is the result of use the AR-DRG system to classify patients not in an acute phase of their illness. Therefore the absolute level of agreement between Typical Nurse Hrs and Actual Nurse Hrs will be affected by the presence of sub-acute and/or non-acute patients on some wards, for example aged care units. If there were a question of whether nursing availability drives the patient load or the patient load drives the nursing allocation, then the charts in Appendix 6 would indicate that both apply at different times. Sometimes the staffing falls away and then patient numbers decline (nursing leads), sometimes changes happen together, and other times the patient numbers lead. What we do see however is a strong relationship between all three series plotted in each chart. The combined wards 1AF and Ward 1AI show a decline in activity over the period with a peak and then large step down in patient hours over ward months 10 and 11. Acuity adjusted staffing estimate Typical fits the actual staffing quite well, and any UNIVERSITY OF TECHNOLOGY, SYDNEY 65

66 large deviation is towards better actual staffing. There has been no real change in acuity adjusted patient hours per nursing hour. Ward 1AL has very stable series but so is the difference between typical and actual nursing, with the actual nursing only about 60% of the former. This ward exemplifies the issue of clinical acuity as an influence on AR-DRG assessed nursing requirement. The Medical Type patient load includes some less acute patients than AR-DRG is designed to classify. We may conclude however that there has been no real change in acuity adjusted patient hours per nursing hour. Ward 1AD has a high staff to patient ratio but one which is fully supported in acuity adjusted terms. The trend is towards reduced workload for the nurses. Ward 1AB is a little less stable than Ward 1AL, but is also a lower acuity type ward. The patient load and staffing series track quite well, however there are quite dramatic up-changes in acuity adjusted patient load (as reflected in the Typical series) which are not matched by changes in staffing. This means the nurses on this ward face very variable workloads, but no clear trend over time. Ward 1AO shows the interdependence of the patient and nursing series very clearly. However no clear trend over time (between Typical and Actual Nursing Hrs) emerges. Ward 1AH shows a large variation in patient hours around ward months 18 and 22, with concurrent changes (although of lesser magnitude) in staffing hours at the same time. There is no trend in regard to workload. Ward 1AM shows a growth of activity during the period. There is a trend for reduced workload for its nurses. Ward 1AA shows a disconnect between its acuity adjusted nursing measure and its actual staffing level. No trend emerges. Ward 1AK also displays a disparity between the actual staffing level and acuity adjusted nursing hours, although the two measures are more closely matched from ward month FINDINGS

67 Ward 1AG shows a close match between actual staffing and the acuity adjusted measure, except for two ward months (14 and 15). Ward 2AJ shows considerable variation in patient hours that are not clearly reflected in staffing numbers. The acuity adjusted measure and actual staffing track each other quite well with no clear trend. The low staffing levels in this ward mean that variations, such as that observed at ward month 17 must place a great deal of strain on the nurses. Ward 2AE would appear very difficult to manage and it is clear that reduction in staffing to meet reduced patient numbers is more easily achieved than staffing up to meet added patient load. This leads to some very distinct staffing shortfalls, for example ward month 24. However no time trend emerges. From the nurses point of view, this would mean unpredictable patient assignments. Ward 2AN shows seasonal effects (including closure 1 ) which mask a major step up in activity. The staffing level lags behind the increase in patient care requirements leading to a massive increase in workload. Ward 2AC also shows the effect of a slow-down around ward month 18 and evidence of increased activity in the later months, but its staffing tracks the activity change well. There is no trend in nursing workload. Findings on Outcomes Potentially Sensitive to Nursing (OPSN) Findings for OPSN other than ALOS Our approach to this technically difficult area has been described in detail in the Longitudinal Analysis section, page 41. We first set about using a counts data approach found in the literature and our findings from these analyses are summarized as Poisson or Approach 1 (A1). We then adopted a less rigorous approach that accommodated the fact that our OPSN data were not actual count data and could be non-integers, and in particular between 0 and 1. We report the findings from this approach under the label GM Ratios or Approach 3 (A3). Finally we adopted a 1 Note that closure and slow-down were only strongly evident in the Calvary data. UNIVERSITY OF TECHNOLOGY, SYDNEY 67

68 weighted least squares approach aimed at finding approximations for the effects of interest measured in a way that could be interpreted in Nurse Staffing terms. These are presented under the label Local Rate Reduction Effects or Approach 2 (A2). The inclusion of the word local is to reinforce the understanding that the estimated effect only applies to practical levels of change in the staffing variable. For example a slope estimate that applies for RN hours as a proportion of total nursing hours would not make sense for a change between no RNs to all RNs on a ward, but would make sense for a 5% decrease in RN share. Table 33 shows the statistical significance of factors in the various models tested for each OPSN, for the three types of analyses used. These findings are presented to show the degree of consistency in results between the analytic methods and hence the amount of weight that can be attached the estimated values presented in Table 34. It must be borne in mind that the different analyses are estimating different things even if an experimental factor bears the same name. We can be most confident if a factor comes up as significant in each analysis of a model for OPSN, but must expect contradictions. TABLE 33 TEST FOR SIGNIFICANT MODELS OF STAFFING ON OPSN OPSN Model Model Model Model Model Model Model Method A1 8 CNS A1 A1 A2 A1 A2 A1 A2 A3 None None None Yes Decubitus A1 None A2 A1 A2 A3 A2 A2 A1 Yes DVT A1 None A2 None A2 A2 None Yes FTR A1 None A1 A2 A1 A2 A3 None None None No GI Bleed A1 None A1 A2 A1 A2 A3 None None None No PM Derangement A1 None None None None None None Yes Pneumonia A1 None A2 None A2 A2 None Yes Pulm Failure A1 None A2 A2 A3 A2 A2 None No Sepsis A1 None A2 None A1 A2 A2 None Yes Shock A1 None A2 None A2 A2 None No UTI A1 None A1 A2 A1 A2 A3 None None None Yes Wound Infection A1 None A2 A2 A3 None None None No 1 Load Cluster 2 Nursing Hours per Patient Hour Adjusted for Model 1 3 RN:PH, EN:PH Adjusted for Model 1 4 RN Proportion Adjusted for Model 1 5 RN:NH, NH:PH Adjusted for Model 4 6 RN:NH, RN:PH Adjusted for Model 4 7 RN:NH, RN:PH, EN:PH Adjusted for Model 3 8 Estimates all clusters 68 FINDINGS

69 The final column in Table 33 indicates whether the Poisson model used data from all available clusters (Yes) or ignored sections of the data because no OPSN values greater than 0.5 were recorded. The cells that are marked up in Table 33 correspond to the best model selected on analysis of deviance criteria and appreciation of the superiority of Approach 1. With the exception of DVT and PM Derangement, the selected models include RN hours as a proportion of total nursing hours, i.e. the skill mix variable. We note that Nursing Hours which was always a candidate in A1 and A3 is never selected (in the best choice of model) on its own; there is always a skill mix component. TABLE 34 PARAMETER ESTIMATES FOR OPSN OPSN Model RN Proprtion RN Hours EN Hours Nursing Hours A1 Finding CNS N/A N/A N/A Yes Decubitus N/A N/A N/A Yes DVT 3 N/A NS * N/A No FTR N/A N/A N/A Yes GI Bleed N/A N/A N/A Yes PM Derangement 1 N/A N/A N/A N/A Yes Pneumonia N/A N/A No Pulm Failure 6 NS * N/A N/A No Sepsis N/A N/A Yes Shock 6 NS N/A N/A No UTI N/A N/A N/A Yes Wound Infection N/A N/A N/A No * Note that the entry NS in Table 34 means the parameter estimate was not significantly different from zero even though the inclusion of the effect in the model was supported. This occurs in DVT and Shock and suggests that these parameter estimates are not useful. It is clear from Table 34 that parameter estimates based on models not supported by Poisson (Approach 1) should be treated with scepticism. The alternative Approach 2 is biased by tendencies for people subject to different risks being nursed at different intensities and skill mix. With this in mind, we have chosen to ignore the parameter estimates for DVT, Pulmonary Failure, Shock and Wound Infection. UNIVERSITY OF TECHNOLOGY, SYDNEY 69

70 TABLE 35 LOCAL RATE REDUCTION EFFECTS OF INCREASING RN SHARE OF NURSING TIME TO BE 10% MORE OF NURSING TIME OPSN Current ACT Mean Rate for 84 hr Stay Current ACT Standard Deviation of Rate for 84 hr stay New ACT Mean Rate for 84 hr Stay New Rate as Percent of Current Rate CNS 0.63% 0.44% 0.35% 55% Decubitus 0.50% 0.18% 0.40% 81% DVT 0.47% 0.18% N/A N/A FTR 0.23% 0.13% 0.17% 73% GI Bleed 0.17% 0.09% 0.11% 63% PM Derangement 2.30% 0.85% N/A N/A Pneumonia 0.47% 0.16% 0.42% 89% Pulm Failure 0.18% 0.07% N/A N/A Sepsis 0.48% 0.24% 0.41% 85% Shock 0.06% 0.04% N/A N/A UTI 1.05% 0.59% 0.70% 66% Wound Infection 0.21% 0.14% N/A N/A Table 35 demonstrates that increasing the skill level of the ward s nurses improves patient outcomes across a broad range of measures. The choice of an 84 hour base in this table is to make the rates relate to the average patient stay in hospital. Thus the figures relate to episodes of care as well as hours of care. An appropriate response to an unacceptably high level of an OPSN is to increase the skill mix rather than to increase the nursing hours per patient day. ALOS as an OPSN The analysis was conducted using software that tested and partitioned the data into a successful outcome group of ward months and a not successful outcome group. The successful group (partition) of the ward months was made up of all those ward months for which the (casemix controlled) total hospital bed-days of patients the 70 FINDINGS

71 software recognised as less than the expected bed-days after adjusting for prior beddays in each AR-DRG. The not successful partition were the remaining ward months. This analysis suggested an association between increased nursing hours and decreased LOS, although it was not statistically significant. It is possible that a clear result will be obtained when there is a larger data set of ward months with which to work. In particular, the setting of meaningful performance thresholds in the current data lead to problems of unreliable sample numbers. Findings on Workforce and Skill Mix Stability The tables at the start of this section illustrate the variances between wards over time in staffing factors. Since time period considered in the tables (6 months) is quite long, any time trend in the RN proportion would be systemic rather than random in nature. The regressions of RN proportion against time found significant improvements in skill mix for ward 1AM and ward 1AF but no other trend was significant. The second group of tables and graphs in this section (and Appendix 6) show changes in staff to patient ratios over time. They are based on ward months which are quite a long period in line management terms. When we consider the very short term fluctuations in nurse to patient ratios we find the effects of shifts generating highly variable data. Figure 3 below is typical for Canberra Hospital. It takes a sample of nursing event times (e.g. start of shift) for ward 1AO and shows the ratio between the nurses and patients on the ward after the event is completed. The sample was chosen based on the simultaneous staff change and patient movement, with the aim of seeing what the staff to patient ratio looks like at these busy change points. We see that the extreme variation is more towards the case of many nurses per patient rather than in the other direction. This may be expected with overlapping shifts and handover time. We also see that even over short time periods, the wards are staffed above a positive minimum level. The data from Calvary Hospital was also highly variable but did not exhibit the lower threshold found in the Canberra Hospital wards. Another feature of the Calvary wards is that they do not show such a distinction in the nurse to patient ratio between shifts. UNIVERSITY OF TECHNOLOGY, SYDNEY 71

72 Nurse:Patient Ratio NURSING WORKLOAD AND STAFFING: IMPACT ON PATIENTS AND STAFF FIGURE 3 STAFFING AT WARD EVENT TIMES L6A L6A Jun-04 Aug-04 Oct-04 Dec-04 Feb-05 Apr-05 Jun-05 Aug-05 Oct-05 Dec-05 Feb-06 Apr-06 Jun-06 Aug-06 Oct-06 Dec-06 Feb-07 Apr-07 There is a possibility that generally satisfactory nurse to patient ratios may be calculated over a long time period masking periods of low staffing. So we look at the ward months again, but this time using the output from the Nursing Model Software. We set this up by looking at the suggested levels of RN staffing. Firstly the Model does not determine adequate levels of nursing; it only assesses what levels are most likely to achieve the threshold (average) standard of LOS outcome and indicates the change (up or down) in staffing required to match these levels. The average outcomes may be quite substandard, so if the Model (with this threshold set) indicates that RN staff may be removed from the ward there is no reason to do so. However if the model indicates that RN staff need to be added to the ward, then there are genuine reasons to be concerned. Figure 4 demonstrates the (cumulative) distribution of ward months and of patient hours against the RN staffing adjustment (as RN hours per patient hour) that would bring LOS outcome expectations in line with the average. A negative adjustment means that the ward month would have been expected to perform above average while a positive adjustment means that the ward month would tend to have a poor performance. 72 FINDINGS

73 CumlativeTotal Patient Hours Cumlative Ward Months NURSING WORKLOAD AND STAFFING: IMPACT ON PATIENTS AND STAFF FIGURE 4 PERFORMING TO AVERAGE LOS RN Workforce Adjustment Spread Total Patient Hours Ward months RN worforce adjustment We see that in both the ward months and patient hours distributions the median adjustment is near zero, however more patient hours are found in the region with better staffing than average. The OPSN work shows that increasing the proportion of RNs by 10% gives good gains and we note that 8% of patient care hours and 11% of ward months are delivered in ward months where the Modelled RN staffing adjustment exceeds 10%. Looking at the better staffed ward months, we see that about the same proportions of patient hours and ward months fall into the range with adjustments below So it is possible to improve outcomes within available resources. The modelled adjustments were analysed at ward month level to find any time trends. Wards 1AL and 1AM had decreasing adjustment (improved RN nursing) over time. No other patterns emerged at ward level or in data with all wards combined. Conclusion The findings of this research do not include evidence of a hospital systems failure. In particular there is no evidence of the feed-forward loop resulting from adverse extended LOS that one would expect in a system in which adequate corrective nursing action cannot be delivered. We can conclude that in ACT this adverse effect is never allowed to run out of control for long. We see that in the relative stability (after casemix UNIVERSITY OF TECHNOLOGY, SYDNEY 73

74 adjustment) of skill mix and staffing levels over the medium to longer term (though not at ward level). We note that there has been an increase in acuity adjusted workload in both hospitals. This increase is more evident at Calvary Public Hospital where nursing workloads are approaching unsustainable levels for an environment where the provision of quality patient care is important. A positive but weak association between adverse LOS and low nursing numbers was shown by running the Nurse Staffing Model software. The effect does not seem large. This may be the result of the masking brought about by the fact nursing levels are never allowed to remain critically low for extended periods and when the levels are low the nurses compensate by giving more of their time. Findings from the crosssectional study support a safety valve model. A relationship between better OPSN outcomes and higher skill mix was found. It is strong enough to encourage the further skilling of the ACT nursing workforce. This is particularly the case because the OPSN are only indicator values that are likely to markedly understate the true rate of avoidable adverse events, and because our analysis was limited by the data on adverse events. We would expect the true gains to be much higher. This hypothesis should be confirmed by analysing data in which the time and place of the adverse events were recorded. Then time periods shorter than ward months could be used as the basis of staffing and skill mix evaluation, removing the masking referred to above. 74 FINDINGS

75 Cross-sectional Findings Patient Characteristics The tables below describe the patient sample and characteristics for both patients and nurses in the cross-sectional study. Data were obtained on 601 different patients and 1768 patient-days using the PRN-80 tool (Table 36). Table 37 outlines the patient characteristics obtained from the patient record in the cross-sectional study. TABLE 36 DATA COLLECTION RESPONSE PATIENT DATA Patient Data Patient Data Form PRN-80 Total 601 (patients) 1768 (patient-days) In this sample of patients (n=601) 88.9% had a caregiver at home. The majority (96.9%) were under the care of a GP (LMO); 16.1% were referred for homecare; 16.8% were waiting for a care facility (this may impact the average length of stay for the ward); 1.4% had been admitted for respite care. Only 24.7% of patient admissions were planned with 13.4% admitted from a care facility. Pre-admission clinics had been attended by 13.7% of patients. Finally, 14.3% of patients had been hospitalised for the same condition in the last three months. TABLE 37 PATIENT CHARACTERISTICS Frequency Percent Patient has a caregiver at home Patient has a GP (LMO) Patient attended pre-admission clinic Referral for homecare Planned admission Patient hospitalised, same condition, past 3 months Patient admitted for respite care Patient waiting for a care facility Patient admitted from a care facility N=1768 (Patients) UNIVERSITY OF TECHNOLOGY, SYDNEY 75

76 Nurse Characteristics As indicated earlier 200 nurses responded to the nurse survey. The staff classifications for which self-reported data were collected included registered nurses level 1 and 2 (RNL1 & RNL2), enrolled nurses (ENs) endorsed enrolled nurses (EENs), trainee enrolled nurses (TENs) and assistants in nursing (AINs). In addition nurses in charge of the wards/units, clinical nurse consultants (CNCs), were asked to participate. Definitions of Nurse Categories RNL1 means a registered nurse who delivers nursing and/or midwifery care to patients/clients in any practice setting and is provided with, or has access to, guidance from more experienced nurses or midwives and, who provides support and direction to enrolled nurses and nursing and midwifery students. RNL2 is a registered nurse who has demonstrated competence in advanced nursing or midwifery practice, provides guidance to RNL1, enrolled nurses, and nursing and midwifery students in the delivery of nursing and/or midwifery care; and acts as Team Leader in the absence of the Clinical Nurse Consultant. An EN is an enrolled nurse who completes one year of training within the Vocational Education and Training (VET) sector. The VET sector consists principally of government-funded Technical and Further Education (TAFE) institutes (McKenna et al. 2000). An EEN is an enrolled nurse who has completed a 6- month post-enrolment medication administration certificate. An AIN assists in the performance of nursing duties and other duties incidental and related to the provision of nursing care services. The AIN is under the direct or indirect supervision of a RN. A Clinical Nurse Consultant (CNC) is responsible for the quality of clinical nursing care provided in a ward or clinical unit or to a specified group of patients (ACT Health / ANF, 2007). When the profile of respondents (n = 200) in the cross-sectional sample is compared to AIHW data (Table 38) the sample had 12% fewer registered nurses and 10% more enrolled nurses; 13% fewer part time nurses and 7% more full time nurses; and 4% more male nurses than ACT data (AIHW, 2006). 76 FINDINGS

77 TABLE 38 NURSE SURVEY RESPONSE COMPARISON WITH AIHW DATA GRADE, GENDER, EMPLOYMENT STATUS ACT Average * Cross-sectional Data Registered nurses (RN) 82.9% 70.5% Enrolled nurses (EN) 17.1% 27.0% Female 93.8% 89.5% Male 6.2% 10.5% Full time 51.9% 58.5% Part time 48.1% 35.0% * (AIHW, 2006) The characteristics of the respondents to the Nurse Survey in regard to employment grade, status, permanency of position and their highest qualification are described in the following tables. Table 39 indicates that the cross-sectional sample consisted of nearly 71% registered nurses (levels 1 and 2) and 27% enrolled nurses (including EENs) with the remaining 2.5% distributed over CNCs (1.5%) and AINs (1%). More than half (58.5%) were employed full-time (n = 117), while approximately one-third (35%) were part-time. Casual and agency staff accounted for the remaining 6.5% of respondents, either RNs or ENs. Table 40 shows that the majority (89%) of the 200 respondents were permanent employees (i.e. not employed on temporary contracts). TABLE 39 NURSE GRADE & EMPLOYMENT STATUS (SELF-REPORTED) AIN EN EEN RNL1 RNL2 CNC Total N Total % Full time Part time Casual Agency Total N Total % % TABLE 40 PERMANENT AND TEMPORARY NURSING STAFF (SELF-REPORTED) Frequency Percent Permanent Temporary Total % Asked for their highest nursing education qualification (Table 41), most RN respondents (49.5%) reported holding a degree or diploma, 23.5% an EN certificate, 13% an RN hospital certificate. Very few reported having a post-registration UNIVERSITY OF TECHNOLOGY, SYDNEY 77

78 qualification ranging from post basic certificates (2.5%) to 6.5% with postgraduate qualifications (i.e. graduate certificate, graduate diploma or a master level degree). In addition, more than half the respondents (n = 119, 59.5%) report that they hold a nonnursing qualification (see Table 42). TABLE 41 HIGHEST NURSING QUALIFICATION (SELF-REPORTED) Frequency Percent EN Certificate EEN Certificate RN Hospital Certificate Post Basic Certificate RN Diploma RN Degree Graduate Certificate Graduate Diploma Master of Nursing No Qualification Total % TABLE 42 HIGHEST NON-NURSING QUALIFICATION (SELF-REPORTED) Frequency Percent TAFE Certificate Diploma Degree Graduate Certificate Graduate Diploma Masters Degree PhD Other No Qualification Total In terms of age the youngest respondent was 20 while the oldest was 67 years. The mean age of 39.2 years is less than the reported average age of employed nurses for the ACT (45.3 years) (AIHW, 2006). Forty-five percent (45%) of the 200 nurses have children living at home, with only 7% having other dependents living with them (Table 44). On average respondents reported having worked as a nurse for almost 12 years, had worked at the present hospital for almost five years, and had worked on the current ward for almost three years (Table 43). 78 FINDINGS

79 TABLE 43 NURSES AGE & EXPERIENCE Mean SD Min Max Age Years worked as a nurse Years worked as a nurse at present hospital Years worked as a nurse on current unit Years worked as a casual nurse N = 200 (Nurses) TABLE 44 CHILDREN & OTHER DEPENDENTS Frequency Percent Children living with you Other dependents living with you 14 7 N = 200 (Nurses) On average respondents worked 32.4 hours per week (range 0 50) at the current hospital, 6 hours on another ward in the same hospital, and 2.1 hours in another job (self-reported, see Table 45). TABLE 45 NURSES HOURS WORKED AVERAGE PER WEEK OVER THE PAST YEAR SELF-REPORT Mean SD Min Max Hours worked per week in this hospital Hours worked per week in other jobs Hours worked per week on other wards in this hospital N=200 (Nurses) Nearly 11 percent (10.8%) of respondents reported having missed a morning or afternoon tea break in the current shift and almost 7% reported that they had missed lunch (Table 46). TABLE 46 NURSES WHO MISSED BREAKS DURING THE CURRENT SHIFT Frequency Percent Morning or afternoon tea Lunch N = 612 (Shifts) UNIVERSITY OF TECHNOLOGY, SYDNEY 79

80 The median number of shifts missed per respondent over the over the past year was 5, and the median number of occasions missed was 3. Approximately 13% of nurses reported missing no work during this period. The most common reason for missing work was physical illness (Table 47). TABLE 47 REASON FOR MISSING WORK Frequency Percent Physical illness Other Injury (work related) Mental health day FACS leave Unable to get requested day off N=200 (Nurses) Ward Characteristics In terms of allied health (Table 48) two-thirds of the wards had a dedicated social worker, physiotherapist (60%), occupational therapist (33%), dietician (20%) or speech therapist (20%). 80% of wards had access to a dietician and a speech therapist, twothirds had access to an occupational therapist and 40% had access to a physiotherapist. As for ancillary staff, 60% had a dedicated clerical assistant and 40% a dedicated ward assistant. 60% had access to a ward assistant and 40% access to a clerical assistant. There was a mean of 6.5 hours of housekeeping support per ward (range 4 8, data not shown). TABLE 48 WARD CHARACTERISTICS: ALLIED HEALTH & ANCILLARY SUPPORT Access Dedicated N % N % Physiotherapist % % Occupational Therapist % % Social Worker % % Dietician % % Speech Therapist % % Ward Assistant % % Clerical Assistant % % N=15(Wards) (one ward did not complete a ward profile, see also Sample definition, page 36) 80 FINDINGS

81 Table 49 shows the level of nursing support at ward level. While all wards (100%) had technical or medical support, 93.3% also had support from specialist nurses, 60% had access to a nurse educator and 53.3% had critical pathways or clinical guidelines. TABLE 49 WARD CHARACTERISTICS: NURSING SUPPORT N Percent Specialist Nursing (diabetes, wound, stomal, chemo, podiatry) % Technical or Medical (MET, I/V, Path, ECG, Pain) % Nurse Educator (access) % Critical Pathways or Clinical Guidelines % N=15 (Wards) (one ward did not complete a ward profile, see also Sample definition, page 36) There was an average of 24.5 (range 16-34) beds per ward. The average number of patients seen each day per ward during the sample period was 22.9 (range ) (Table 50). TABLE 50 WARD CHARACTERISTICS Mean SD Min Max Beds Patients on ward (mean per day) N=15 (Wards) (one ward did not complete a ward profile, see also Sample definition, page 36) Skill Mix Characteristics Nursing hours worked A skill mix profile was derived from nursing hours worked and was calculated from the complete record of ward nursing roster data. These data were aggregated into proportion of hours worked by employment status (full-time, part-time, casual and agency); by grade categories (RN level 1, RN level 2, EN [including EENs as it was not possible to differentiate on all ward rosters], and AIN). These profiles include hours worked on the ward only and excluded CNCs and CNEs. Table 51 below outlines these staffing data. UNIVERSITY OF TECHNOLOGY, SYDNEY 81

82 TABLE 51 CROSS-SECTIONAL STAFFING DATA Staffing data Ward-shifts * Ward-days Daily ward staffing profile * Data for a 24 hour period from a single hospital ward Data for a shift-period from a single hospital ward See also Table 6, page 25 Figure 5 summarises the percentage of RN, EN, and other nurse hours worked per ward, and hence provides an overview of the skill mix across the sample. As mentioned, staffing data were not complete for two wards. Analyses were restricted to reporting on a per ward and ward-day basis due to the small number of wards (i.e. 14 wards; see also sample details, page 36). The skill mix ranged from 44% RN and 55% EN staff, to 82% RN and 18% EN staff. Most wards had between 60% and 80% RN staff (Figure 5). There were three wards with a mix of fewer than 60% RN staff and three wards with greater than 80% RN staffing. FIGURE 5 PERCENTAGE OF NURSE HOURS WORKED PER WARD, BY GRADE, ORDERED BY RN% 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Ward RN EN Other Table 52 indicates that the cross-sectional staffing profile was within 10% of longitudinal data on all but ward 1AF, suggesting that the sample was generally representative of staffing for each ward (see also Table 58, page 89). 82 FINDINGS

83 TABLE 52 COMPARISON OF LONGITUDINAL & CROSS-SECTIONAL STAFFING PROFILE,, BY WARD RN% EN% Ward Code Cross-sectional Longitudinal Cross-sectional Longitudinal 1AA AB AD AF AG AH AI AK AM AO AC AE AJ AN * Cross-sectional data recorded other nursing staff as follows: 1AA:4.6%; 1AG:2.9%; 1AH:1.1%; 1AK:4.0%; 2AC: 0.8%; 2AN:10.2%. These data were not collected in the longitudinal component. Staffing data for Ward 1AF was atypical when compared to longitudinal data. On a ward-day basis (Figure 6) there were 39 (58%) ward-days that had between 60-80% RN hours worked and one that had 100% RN hours. Fifteen ward-days had fewer than 60% of hours worked by RNs and 13 had 80% or more. Twenty ward-days had greater than 35% EN hours worked. Only twelve ward-days over six different wards employed nurses which were other than RN and EN categories and the percentage of these other nurse hours worked ranged from %, with two outliers at 22.4 and 24.5%. UNIVERSITY OF TECHNOLOGY, SYDNEY 83

84 FIGURE 6 PERCENTAGE OF NURSE HOURS WORKED PER WARD-DAY, BY GRADE, ORDERED BY RN% 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Ward-Day RN EN Other As indicated earlier part-time nurses were under-represented in this sample when compared to Territory figures (48.1% part-time). This may be an artefact of the data collection process. Most data collection was undertaken in a six to eight hour period between 0700hrs and 1800hrs which might have provided less opportunity to engage those nurses working fewer days or outside these hours. However, there is no effect on staffing data as they were obtained from the ward roster. In addition to the data from the ward roster, the Nurse Survey asked respondents to report on for example, employment status and hours worked. Figure 7 and Figure 8 present ward level and ward-day level information on employment status. The percentage of full-time, part-time, casual and agency hours worked per ward for the 14 wards surveyed are shown below (Figure 7). The lowest percentage of full-time hours worked on one ward was 39.4% and the highest percentage was 75.38%. There were two wards which had less than 40% full-time staff. Part-time staff ranged from % and casual staff ranged from 1 3% in five wards with a maximum of 26.1%. Four wards in the sample employed no agency staff at all, while the remaining 10 wards employed between 1 8% agency staff. However, there is considerable variation in these figures when reported on a ward-day basis. 84 FINDINGS

85 FIGURE 7 PERCENTAGE OF NURSE HOURS WORKED PER WARD, BY EMPLOYMENT STATUS, ORDERED BY FT% 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Ward Full time Part time Casual Agency Table 53 and Table 54 show the proportion of hours worked per ward and ward-day (see Glossary, page 22) respectively by employment status. The mean for each category of staff per ward and per ward-day is comparable although the range and consequently the SD are larger in the ward-day data. For example per ward, full-time staff comprised 53.8% (SD = 11.22%), part-time staff comprised one-third (SD = 11.75%), casual staff 9.9% (SD = 8.39%) and agency staff comprised 2.9% (SD = 2.81%) of the ward staffing (Table 53). Table 54 shows that the range in the proportion of full-time ( %) and part-time (0 79%) hours was considerably greater at the ward-day level. TABLE 53 PROPORTION OF HOURS WORKED PER WARD BY EMPLOYMENT STATUS Mean SD Min Max Full-time Part-time Casual Agency N=14 (Wards) (two wards did not provide staffing data, see also Sample definition, page 36) Table 54 shows that 54% of the hours worked per ward were by full-time nurses and a third (33.3%) of the hours were worked by part-time nursing staff. The remaining 12% were casual and agency hours. These employment status categorisations were not UNIVERSITY OF TECHNOLOGY, SYDNEY 85

86 available in the longitudinal data, and they therefore provided a more detailed understanding than would be available using administrative data alone. TABLE 54 PROPORTION OF HOURS WORKED PER WARD-DAY BY EMPLOYMENT STATUS Mean SD Min Max Full-time Part-time Casual Agency N=67 (Ward-Days) Percentages of full-time, part-time, casual and agency hours worked per ward-day are shown below (Figure 8). The lowest percentage of full-time hours worked on one ward-day was 10.5% and the highest percentage was 93.3%. There were ten warddays which had less than 40% full-time staff and two ward-days which had more than 80% full-time staff. One ward-day had 71.7% full-time nurses, no part-time or agency staff at all and instead filled this gap with 28.3% casual staff. Apart from this particular ward-day, the remaining ward-days had part-time staff ranging from % of their staff. Twenty-three ward-days reported no casual staff, while the remaining 44 (66%) ward-days had between % casual staff. On 43 ward-days in the sample no agency staff were employed at all, while for the remaining 24 ward-days between % agency staff were employed. This analysis indicates that there is considerable variation in staffing across many wards each day. 86 FINDINGS

87 FIGURE 8 PERCENTAGE OF NURSE HOURS WORKED PER WARD-DAY, BY EMPLOYMENT STATUS, ORDERED BY FT% 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Ward - Day Full time Part time Casual Agency Table 55 below shows great variation in the proportion of hours worked per wardday by grade. RN L1 staff worked on average 51.6% of the hours with a large range from %; RN L2 staff worked on average 16.8% with a range of between 0 51%; and ENs worked 29.9% of hours, also with a large range of 0 66%. TABLE 55 PROPORTION OF HOURS WORKED PER WARD-DAY, BY GRADE Mean SD Min Max RN L1 * RN L2 * EN Other N=67 (Ward-Days) * See Glossary, page 22 When the same data are presented per ward (Table 56) the means are comparable but as expected, the standard deviations and the range for each grade are smaller. The maximum percentage hours worked per ward for ENs is 55.9% and 74.5% for RNL1s. UNIVERSITY OF TECHNOLOGY, SYDNEY 87

88 TABLE 56 PROPORTION OF HOURS WORKED PER WARD, BY GRADE Mean SD Min Max RN L RN L EN Other N=14 (Wards) (two wards did not provide staffing data, see also Sample definition, page 36) Staffing data were also examined for skill mix across three equal shift-periods (see Table 6, page 25), referred to as morning ( ), evening ( ) and night ( ) shift-periods. Raw staffing data were apportioned to these periods to provide an indication of the relative availability of these staffing categories during the day, evening or night (see also Table 6, page 25). Similar proportions of all categories were found on morning and evening shift-periods, while the night period showed an increased presence of ENs, significantly fewer RNL1 hours, and slightly fewer RNL2 hours (Table 57). TABLE 57 PROPORTION OF HOURS WORKED PER SHIFT-PERIOD BY GRADE % Hours Evening Morning Night RNL1 54.0% 54.0% 41.8% RNL2 17.6% 16.6% 15.9% EN 27.5% 27.4% 41.1% Other 1.0% 1.9% 1.2% N=1292 (Shift-periods) A comparison of cross-sectional and longitudinal staffing data indicated that there were similar mean proportions of staffing hours per ward (eg RN = 68.5% and 68.4% respectively), although with slightly greater variation in the cross-sectional data (eg RN SD = and 9.72 respectively, see Table 58). 88 FINDINGS

89 TABLE 58 COMPARISON OF LONGITUDINAL & CROSS-SECTIONAL STAFFING PROFILE, BY GRADE Crosssectional Longitudinal N * Mean SD Min Max RN EN Other RN EN Other * Two wards did not provide cross-sectional staffing data, see also Sample definition, page 36 Nursing Workload The movement of patients through the ward is referred to as patient churn. Each new admission, transfer, or discharge, requires documentation, orientation, clinical assessment and management review, and other tasks associated with the patient. In order to provide an indication of the amount of churn per ward, Patients per bed was calculated per ward by dividing the number of patients per day by the number of beds (Equation 1). This calculation does not include bed movements within the ward. While EQUATION 1 PATIENTS PER BED NumberofPatientsOnWardPerDay PatientsPerBed= NumberofBedsPerDay the mean was one patient per bed per day there was considerable variation between wards, with the range from less than one patient (0.7) per bed per day and the maximum 1.2 (Table 59). When examined on a day by day basis, rather than averaged across the ward sample period, the maximum rose to 1.4, with a larger range ( ). TABLE 59 PATIENTS PER BED N Mean SD Min Max Patients per bed by ward 14 * Patients per bed by ward-day * Two wards did not provide cross-sectional staffing data, see also Sample definition, page 36 Nursing hours per patient day & hours of care required per patient day Nursing hours per patient day (NHPPD, Equation 2) provided varied considerably on a per day basis (mean EQUATION 2 NURSING HOURS PER PATIENT DAY NursingHoursWorkedOnWard NHPPD= NumberofPatients UNIVERSITY OF TECHNOLOGY, SYDNEY 89

90 Frequency Frequency NURSING WORKLOAD AND STAFFING: IMPACT ON PATIENTS AND STAFF 6.5, range ) and were reasonably normally distributed though the data indicate significant variation between and within wards (Table 61, page 91). The PRN-80 determines the minutes of care (later transferred into hours) required by patients for 24 hours using data from the medical record (see Table 7, page 25 & Table 13, page 39). An average of the hours of care required per patient per day (determined by the PRN-80) was calculated. Across the sample of 67 ward-days there was considerable variability (Figure 9). The average requirement per ward-day (using PRN-80) was 7.02 hours. The difference between the minimum and maximum requirements per ward-day (range) was considerable; from just over 4 hours to nearly 11 hours (10.7 hours). This degree of variability in care needs makes it difficult to predict the staffing required, and the discrepancy between hours needed and available hours may impact on workload, quality of care and the work environment. At the patient-day level there was also great variability over 24 hours (mean 7 hours 5 mins; range 1 hour 4 mins 21 hours 11 mins) indicating great variation between individual patient care requirements per day (Table 60). FIGURE 9 HOURS OF CARE REQUIRED Histogram Histogram Mean =7.02 Std. Dev. =1.486 N = Mean =7.09 Std. Dev. =1.389 N = Mean hours of nursing care required for 24 hours (ward-day) Mean hours of nursing care required for 24 hours (ward) 90 FINDINGS

91 TABLE 60 HOURS OF CARE REQUIRED N Mean SD Min Max Patient-Day Ward-Day Ward 14 * * Two wards did not provide staffing data, see also Sample definition, page 36 When the hours of care required (measured using the PRN-80) are compared to those provided (Table 61), on average, approximately one half hour per day of additional care is required to meet each patient s needs (see Table 13 for explanation of the use of this tool). There was considerable variation per ward day over the entire sample period, as displayed in Figure 10. TABLE 61 NURSING HOURS PER PATIENT DAY; NURSING CARE REQUIRED; NURSING DEMAND/SUPPLY Hours of nursing care required per patient day Mean SD Min Max Percentiles Nursing hours per patient day Nursing demand/supply N=67 (Ward-Days) FIGURE 10 NURSING HOURS PER PATIENT DAY & NURSING CARE REQUIRED (ENTIRE SAMPLE PERIOD 67 DAYS) Hours of nursing care required per patient day Nursing hours per patient day Ward Day UNIVERSITY OF TECHNOLOGY, SYDNEY 91

92 An additional factor, referred to here as the nursing demand/supply figure (Equation 3), is calculated by dividing the required hours of care (derived from the PRN-80) by the hours of care provided (O'Brien-Pallas et al., 2004). Nursing demand/supply figures over 100 indicate that more care is required by patients than is provided. Table 61 indicates that only a quarter of the ward days EQUATION 3 NURSING DEMAND/SUPPLY FIGURE HoursofCareRequiredperDay %NursingDemand/Supply= 100 NursingHoursPerPatientDay sampled are in balance for nursing resources: That is, the supply of nursing hours equals or is less than that of the hours of nursing care required by patients on only 25% of days in the sample (i.e. nursing demand/supply = 91.4 at the 25 th percentile). For the remaining 75% of days there is an imbalance nursing hours required exceed those provided. These data were further analysed by hospital peer group, using ward mean data (Table 62), and were compared to similar data from NSW. This showed that, in the ACT, both the hours of nursing care required per patient day and nursing hours per patient day provided were higher in the A group hospital. However, there was a larger imbalance between care required and provided in the B1 hospital (A=110.8, B1=119.0). When compared with NSW figures, more hours of nursing care per patient day were required in A group, and fewer than NSW in B1. Nursing hours per patient day were substantially higher than NSW figures in A group (ACT=7.1, NSW=5.3), and slightly lower in B1 group (ACT=5.0, NSW=5.2). There was a lower imbalance between care required and provided in both groups in the ACT. TABLE 62 COMPARISON OF ACT & NSW NHPPD & CARE REQUIRED FIGURES, MEAN PER WARD, BY HOSPITAL PEER GROUP Hours of nursing care required per patient day Nursing hours per patient day provided Nursing demand/supply ACT * NSW Group Mean Min Max Mean Min Max A B A B A B * N=14 wards (10 in A group hospital, 4 in B1 group hospital) (two wards did not provide staffing data, see also Sample definition, page 36) N=65 wards (49 in A group hospitals, 16 in B1 group hospitals) 92 FINDINGS

93 Work Environment A range of factors in the work environment were measured. Results from the subscales of the Nursing Work Index Revised (NWI-R) and the Environmental Complexity Scale (ECS) were compared with prior research, and also included in regression models on patient and nurse outcomes. Nursing Work Index Revised Compared to the findings in the NSW study (Duffield et al., 2007) nurse autonomy, nurse-doctor relationships, resource adequacy and nurse control over practice were higher in the ACT data than the NSW study, while slightly lower for nurse leadership (Table 63) (see Glossary, page 22 for definitions). TABLE 63 NURSING WORK INDEX - REVISED NSW 2004/5 ACT Health Mean SD Mean SD Min Max Nurse autonomy Nurse control over practice Nurse-doctor relations Nurse leadership Resource adequacy N= 200 (Nurse Respondents) Associations were found between some factors of the NWI-R and the nursing demand/supply level (Table 64). A high nursing demand/supply figure (indicating wider discrepancy between hours of care required and that supplied) related to lower levels of autonomy, control over practice and nurse-doctor relations. TABLE 64 NURSING WORK INDEX REVISED & NURSING DEMAND/SUPPLY Kendall's τ Nursing Demand/Supply Nurse autonomy (mean) (*) Nurse control over practice (mean) (*) Nurse-doctor relations (mean) (**) Nurse leadership (mean) Resource adequacy (mean) * Correlation is significant at the 0.05 level (2-tailed) ** Correlation is significant at the 0.01 level (2-tailed) UNIVERSITY OF TECHNOLOGY, SYDNEY 93

94 Environmental Complexity Scale This instrument was completed by nurses per shift 2. Compared to results in the NSW study (Duffield et al., 2007), nurses in ACT scored slightly higher for unanticipated changes in patient acuity (6.4, SD = 1.02), and identically in the other two sub-scales. TABLE 65 ENVIRONMENTAL COMPLEXITY SCALE Re-sequencing of work in response to others Unanticipated changes in patient acuity Composition and characteristics of the care team * N = 612 (Shifts) N = 6839 (Shifts) NSW 2004/5 ACT * Mean SD Mean SD Min Max Items one and two on the Environment Complexity Scale referred specifically to the impact of students on the ward. In both instances, the majority of responses indicated that students were not present on that shift (Table 66 & Table 67). When students were present on the ward, over half of respondents suggested that their workload increased. TABLE 66 ECS ITEM 1: STUDENTS ON THE UNIT REQUIRED SUPERVISION AND ASSISTANCE Students required supervision/assistance Frequency Percent Increased Decreased Same N/A N = 612 (Shifts) 2 Note that the term shift indicates the shift as reported by the respondent. It is not the same as a shift-period derived from roster data (see Table 6, page 25). 94 FINDINGS

95 TABLE 67 ECS ITEM 2: STUDENTS WANTED ACCESS TO CHARTS, EQUIPMENT AND SUPPLIES Students wanted access to charts, etc Frequency Percent Increased Decreased Same N/A N = 612 (Shifts) Quality of Care Nurses were asked on the Environmental Complexity Scale How would you describe the quality of your nursing care delivered during this shift? The response choices were excellent, good, fair and poor. They were also asked on the Nurse Survey for their view of the changes in quality of care over the past 12 months (see Appendix 7, Instruments). Table 68 indicates the quality of care reported per shift. Of the 612 responses 88% of nurses rated the quality of care as excellent or good while 12% reported it as fair or poor over the past shift. When asked to indicate whether the quality of care given over the last 12 months had changed on their wards, 80% of respondents indicated that it had improved or remained the same, and 20% believed that it had deteriorated (Table 69). TABLE 68 QUALITY OF CARE PER SHIFT Frequency Percentage Excellent/good Fair/poor Total TABLE 69 QUALITY OF CARE OVER THE PAST YEAR Frequency Percentage Improved Remained same Deteriorated Total UNIVERSITY OF TECHNOLOGY, SYDNEY 95

96 Tasks delayed or left undone The Environmental Complexity Scale allowed measurement of nurses perceptions of tasks delayed or left undone. Respondents were asked Which of the following tasks were necessary but left undone because you lacked the time to complete them? and to Check all that apply from the list provided. Rates were calculated for each nurse per shift across the cross-sectional sample (Table 70). On average each nurse was delaying 1.3 tasks per shift and not completing 1.5 tasks per shift. A small response rate was seen for night shift so statistical comparisons could not be made, but an apparently similar rate of tasks delayed was found, with a lower rate of tasks not done. TABLE 70 TASKS DELAYED OR NOT DONE PER NURSE PER SHIFT Tasks Delayed Tasks Not Done Shift N Mean SD Min Max Morning Evening Night All Shifts Morning Evening Night All Shifts When compared by hospital peer group using ward means, a higher rate of tasks delayed was found in the A group hospital, while a higher rate of tasks not done was found in the B1 group hospital. These data were also compared by peer group with NSW data (Table 71). In regard to tasks delayed, ACT had a slightly higher rate in A group, and a lower rate in B1 group. Tasks not done were lower than NSW in both hospital groups. TABLE 71 COMPARISON OF ACT & NSW TASKS DELAYED OR NOT DONE, MEAN PER WARD, BY HOSPITAL PEER GROUP Tasks delayed Tasks not done * N=14 wards (10 in A group hospital, 4 in B1 group hospital) N=65 wards (49 in A group hospitals, 16 in B1 group hospitals) ACT * NSW Group Mean Min Max Mean Min Max A B A B FINDINGS

97 Detailed analysis of these data (Table 72) show that over the 612 shifts for which data were collected, routine vital signs, medications or dressings were reported as not done on 49 occasions (8%) and were delayed 165 times (27%). In addition, routine mobilisation or turns in bed were not done on 42 occasions (6.9%) and delayed 229 times (37.4%); delay in administering PRN (as needed) pain medication occurred 141 times (23%) and delayed response to patient bells occurred 282 times (46.1%). Necessary tasks left undone included routine teaching for patients and families which occurred 80 times (13.1%) and nurses acknowledged omitting preparing the patient and family for discharge on 71 occasions (11.6%). Comforting and talking to patients was not done 210 times (34.3%) and adequate documentation of nursing care was omitted 77 times (12.6%). Pressure area care was left undone 117 times (19.1%) and oral hygiene 128 times (20.9%). Most categories had similar or lower rates compared to recent NSW research (Duffield et al., 2007). TABLE 72 TASKS NOT DONE OR DELAYED DUE TO TIME PRESSURES NSW 2004/5 Tasks Not Done Tasks Delayed % Freq % Freq % Comforting/talking with patients Nursing care plan not done Oral hygiene Pressure area care Routine teaching for patients and families Adequately documenting nursing care Prepare patient and family for discharge Routine vital signs, medication Routine mobilisation Other Delay in responding to patient bell Delay in administering PRN pain medications N=612 (Shifts) When asked to specify other tasks delayed or left undone, 22 were cited. Analysis of these data (Table 73) shows that respondents reported a lack of time to complete patient hygiene tasks i.e. showering was thought of as necessary but left undone on five occasions (22.7%), dressings on three occasions (13.6%). Lack of time to complete and maintain fluid balance charts was mentioned separately by two respondents (9.1%), as was patient/family support, time to complete wound charts and UNIVERSITY OF TECHNOLOGY, SYDNEY 97

98 assessment/discharge activities were cited as being other tasks necessary but left undone. Finally, on one reported occasion a patient s enema was left undone. TABLE 73 OTHER TASKS NECESSARY BUT NOT DONE Frequency Percent Showering Other Dressings Patient/Family support Assessment/Discharge Fluid balance Wound charts Monitoring Patient enema Total % Time available to deliver care Respondents were asked Please rate the time available to deliver care on this shift compared to the last five shifts you have worked, with the choice of less, the same, or more time than usual. Table 74 shows that 54.4% of respondents had about the same amount of time as usual to deliver care on the current shift compared to the last five shifts worked. 22.7% reported they had more time than usually available to deliver care on their most recent shift, with 22.9% indicating they had less time than usual (Table 74). TABLE 74 TIME AVAILABLE TO DELIVER CARE PER SHIFT Response Frequency Percent Less time than usual About the same amount of time as usual More time than usual Total How much more time needed to deliver care Nurses were asked, Approximately how much more time do you need to give the type of care stated in the nursing care plan or your assessment of patients needs today? Respondents were asked to tick only one response. 98 FINDINGS

99 Table 75 shows that 26.3% of nurse respondents stated they needed no more time that shift to provide the type of care stated in the nursing care plan, 33.8% reported that up to 30 minutes more time was needed, and nearly 40% of respondents felt that more than 30 additional minutes were necessary to deliver care, 11% of whom felt they needed more than 60 minutes to do so. The additional time required may be offset by the use of support worker roles. TABLE 75 HOW MUCH MORE TIME NEEDED Response Frequency Percent No more time needed < 15 minutes minutes minutes minutes > 60 minutes Total An examination of these data by hospital peer group (Table 76) showed that there was more time required by nurses to complete their care per shift in the B1 hospital. Compared to NSW data, slightly more time was required in both groups. TABLE 76 COMPARISON OF ACT & NSW TIME NEEDED PER SHIFT, MEAN PER WARD, BY HOSPITAL PEER GROUP Tasks not done ACT * NSW Group Mean Min Max Mean Min Max A B * N=14 wards (10 in A group hospital, 4 in B1 group hospital) N=65 wards (49 in A group hospitals, 16 in B1 group hospitals) Indirect or Additional Care Activities Nurses were asked, Which of the following tasks did you perform during this shift. The response was to Check all [boxes] that apply. Across the entire sample of 612 shifts, a total of 1694 indirect care activities were completed. The average proportion of nurses required to undertake these tasks per ward-day is shown in Table 77. This table indicates that, by ward-day, 46% of nurses were required to deliver or retrieve patient meal trays, 34% order, co-ordinate or perform ancillary work, 42% UNIVERSITY OF TECHNOLOGY, SYDNEY 99

100 undertake cleaning and 43% clerical duties. 30% arrange discharge referrals and transport, while 9% transport patients. 38% of respondents state they are required to start IVs while performing ECGs was reported by 14% and routine phlebotomy by 16%. TABLE 77 PROPORTION OF NURSES UNDERTAKING INDIRECT CARE ACTIVITIES, PER WARD-DAY Mean SD Deliver/retrieve patient meal trays Order/coordinate/perform ancillary work Start IVs Arrange discharge referrals and transport Undertake ECGs Undertake routine phlebotomy Transport patients Undertake cleaning duties Undertake clerical duties N=67 (Ward-Days) Table 78 shows the proportion of the above tasks undertaken per shift. The majority were completed during the morning shift (64.1%), and fewer during the evening shift (31.5%). Relatively few were undertaken overnight (4.4%) with the exception of routine phlebotomies, of which nearly 10% occurred between at night. When these data are matched to the skill mix category of respondents to the nurse survey, approximately 75% of these tasks were reported by RNL1, with 20% by ENs or AINs (data not shown). TABLE 78 INDIRECT CARE ACTIVITIES BY SHIFT Morning Evening Night N % N % N % Deliver/retrieve patient meal trays Order/coordinate/perform ancillary work Start IVs Arrange discharge referrals / transport Undertake ECGs Undertake routine phlebotomy Transport patients Undertake cleaning duties Undertake clerical duties Total N=612 (Shifs) 100 FINDINGS

101 Violence Experienced Nurses were also asked about their experience of violence: In the last 5 shifts you worked, have you experienced any of the following while carrying out your responsibilities as a nurse. The response was yes or no to physical assault, threat of assault, and emotional abuse (Table 79). Emotional abuse was experienced by 33% of respondents but by up to a maximum of 58% of staff on one ward. In terms of threat of violence 21% experienced this and while there were wards where no staff experienced a threat of violence, up to a maximum of 67% of staff on one ward did. The results are similar for physical violence, where 15% of staff experienced this in the past five shifts and up to 58% of staff on a ward did so. TABLE 79 NURSES EXPERIENCING VIOLENCE IN THE LAST 5 SHIFTS Entire Sample * Average Per Ward Frequency Percentage Min% Max% Physical violence Threat of violence Emotional abuse * Proportion of nurses experiencing violence in entire sample (N=200 Nurses) Proportion of nurses experiencing violence per ward (N=16 Wards) Respondents were also given the opportunity to choose the source of violence from a list provided. Nurses indicated that patients and families were responsible for most physical assaults (96.6%) and threats of assault (95.1%). The majority of emotional abuse was also from patients and their families (69.7%) but was also reported from coworkers. These figures are similar to NSW data (Table 80). UNIVERSITY OF TECHNOLOGY, SYDNEY 101

102 Emotional Abuse % Threat of Violence % Physical Violence % ACT NSW ACT NSW ACT NSW NURSING WORKLOAD AND STAFFING: IMPACT ON PATIENTS AND STAFF TABLE 80 SOURCE OF VIOLENCE TOWARDS NURSES (COMPARISON WITH NSW [DUFFIELD ET AL., (2007)]) Patient Patient + family/visitor Family/visitor Nursing co-worker Patient + nursing co-worker Other Physician Patient + physician Patient + family/visitor + physician + nursing co-worker Patient + physician + nursing co-worker Family/visitor + nursing co-worker Physician + nursing co-worker Patient + family/visitor + nursing coworker Family/visitor + physician Patient + family/visitor + physician Number of nurses * * ACT N = 200 (Nurses); NSW N = 2278 (Nurses) Top 3 categories indicated in bold Satisfaction and Intention to Leave Most nurses (71.5%) were satisfied with their current job (Table 81), although more were satisfied with nursing as a profession (79.5%). Almost three-quarters of respondents (74%) were not intending to leave their current job (Table 82). 102 FINDINGS

103 TABLE 81 NURSES' SATISFACTION WITH CURRENT JOB & NURSING AS A PROFESSION Frequency Percent Satisfied with current job Dissatisfied with current job Satisfied with nursing as a profession Dissatisfied with nursing as a profession Total N=200 (Nurses) TABLE 82 NURSES PLANNING TO LEAVE THEIR CURRENT JOB Frequency Percent Do not intend to leave current job in the next 12 months Intend to leave current job in the next 12 months Total Patient Outcomes As described previously, patient outcomes in the cross-sectional data were collected from both the patient record and ward-level reporting mechanisms. The patient outcomes here were falls with and without consequences and medication errors with and without consequences. These data were aggregated to the ward level in order to conduct correlation analyses and regression models. The dependent variables (patient outcomes) were in all cases calculated as percentage of patients who experienced (the event) per ward. Regression models, either linear or logistic, were conducted and Beta (β) weights calculated where possible to indicate relativities between the factors. Adverse Events Adverse events were collected from the patient record or ward reporting system. Twenty six (4.3%) patients in the study were found to have experienced a fall with or without injury (Table 83), and some of these patients had experienced both types of fall. Two patients experienced medication errors without consequences. These adverse event rates were very low compared to other studies and may be indicative of the short sample period per ward, data collection issues, or unknown factors. UNIVERSITY OF TECHNOLOGY, SYDNEY 103

104 TABLE 83 PATIENTS EXPERIENCING ADVERSE EVENTS Frequency Percent Medication errors without patient consequences * Falls with injury Falls without injury Falls (any either with or without injury) N=601 (Patients) * No patients recorded medication errors with adverse consequence Some patients experienced both types of fall. See Glossary, page 22 These data were also calculated as the percentage of patients per ward who experienced these adverse events, by hospital peer group (Table 84). This showed a higher proportion of patients in the A group hospital experienced any type of fall, a fall with injury or medication error without consequences, and a higher proportion in the B1 hospital experienced falls without injury. Compared to NSW data, a lower proportion of patients in the ACT experienced medication errors without consequence in both groups, and falls with or without injury in the B1 group. In the A group, a greater proportion of patients in the ACT experienced falls. TABLE 84 COMPARISON OF ACT & NSW PATIENT OUTCOMES, MEAN % OF PATIENTS PER WARD, BY HOSPITAL PEER GROUP ACT * NSW Group Mean Min Max Mean Min Max Medication errors without A patient consequences * B Falls with injury Falls without injury A B A B Falls (any either with or A without injury) B * N=14 wards (10 in A group hospital, 4 in B1 group hospital) N=65 wards (49 in A group hospitals, 16 in B1 group hospitals) Although statistically significant correlations were found at the ward and ward-day level between these adverse events and a number of other variables, examination of scatter plots showed that this was an effect of the low rates, with the majority of data points clustered about zero and a few outliers influencing the results. Therefore, no relationships could be established. 104 FINDINGS

105 In addition to these data, late administration of medication (more than 30 minutes, see definitions Table 5, page 25) was recorded per patient-day. These time-based medication errors were recorded on 40 of the 1758 patient-days (Table 85). On most of these 40 patient-days between 1 and 4 errors were recorded, and on one patient day between 5-14 errors were recorded. It is possible these errors could be recorded in the aforementioned patient data collection also, so that only a summary of frequency is presented here. Out of the 601 patients studied, 34 (5.7%) experienced this type of error (data not shown). TABLE 85 TIME-BASED MEDICATION ERRORS Frequency Percentage 1-4 errors per patient-day errors per patient-day N=1758 (Patient-Days) Outcome Predictors Tasks Not Done & Tasks Delayed per Ward-Day Linear regression models for tasks delayed and not done were developed with data at the ward-day level. Analysis at this level of data for these outcomes is more meaningful as it examines the overall picture of the ward for a given day. Similar factors were influential in regard to both outcome variables (Table 86 & Table 87). The proportion of nurses indicating less time available to deliver care, the amount of additional time required to complete care this shift, and the proportion of hours worked by agency staff were common elements. As these factors increased so did the rate of tasks delayed or not done. Additional predictors were identified in regard to the rate of tasks not done (Table 86). These included the proportion of patients admitted from a care facility and the amount of involuntary overtime reported. Both models explained over 30% of the variance. UNIVERSITY OF TECHNOLOGY, SYDNEY 105

106 TABLE 86 LINEAR REGRESSION ON TASKS NOT DONE Direction β Weight Amount more time needed this shift Positive (+) Proportion of patients admitted from a care facility Positive (+) Proportion of hours worked by agency staff Positive (+) Average weekly overtime worked - involuntary paid Positive (+) Proportion of nurses indicating less time available to deliver care Positive (+) Adjusted R 2 = N= 67 (Ward-Days) p 0.05 TABLE 87 LINEAR REGRESSION ON TASKS DELAYED Direction β Weight Proportion of nurses indicating less time available to deliver care Positive (+) Amount more time needed this shift Positive (+) Proportion of hours worked by agency staff Positive (+) Adjusted R 2 = N= 67 (Ward-Days) p 0.05 Correlation of factors shown to be significant predictors of tasks delayed or not done were generally consistent with these regression models, although two items did not show a statistically significant correlation (Table 88). TABLE 88 CORRELATION OF FACTORS IN LINEAR REGRESSION MODELS ON TASKS NOT DONE & TASKS DELAYED Kendall's τ Tasks not done Tasks delayed Additional time needed this shift 0.260(**) 0.361(**) Proportion of patients admitted from a care facility 0.189(*) Proportion of hours worked by agency staff Average weekly overtime worked - involuntary paid Proportion of nurses indicating less time available to deliver care (**) * Correlation is significant at the 0.05 level (2-tailed) **Correlation is significant at the 0.01 level (2-tailed) The amount of additional time needed this shift was highly correlated with two outcome variables; tasks not done (τ =0.260) and tasks delayed (τ =0.361). As tasks not done or delayed increased, the amount of additional time reported as needed this shift also increased. Likewise an increase in the proportion of nurses indicating less time available to deliver care indicated an increase in tasks delayed (τ =0.244). Also an 106 FINDINGS

107 increase in the proportion of patients admitted from a care facility led to an increase in tasks delayed (τ =0.189). Nurse Outcomes Analyses were conducted for the nurse outcome variables - job satisfaction, satisfaction with nursing, and intention to leave the current job. These variables were measured at the nurse level. Analysis at this level is appropriate to examine the influence of workload and other variables on individual nurse outcomes. Job Satisfaction Nurses who were satisfied with their profession, had adequate resources to do their job, and who worked on wards with a higher overall amount of nursing hours were more likely to be satisfied with their current job. Older nurses, and those nurses missing a higher number of shifts, were less likely to be satisfied with their job (Table 89). TABLE 89 LOGISTIC REGRESSION ON JOB SATISFACTION Direction β Weight Number shifts missed work Negative (-) Satisfaction with nursing Positive (+) Resource adequacy Positive (+) Total nursing hours provided on the ward Positive (+) Age Negative (-) Pseudo R 2 =0.400 N=149 (Nurses) p 0.05 Satisfaction with Nursing Nurses who were satisfied with their job and who had adequate resources were more likely to be satisfied with their profession, while those in temporary employment were less satisfied with nursing. A higher patient turnover also predicted satisfaction with nursing (Table 90). UNIVERSITY OF TECHNOLOGY, SYDNEY 107

108 TABLE 90 LOGISTIC REGRESSION ON SATISFACTION WITH NURSING Direction β Weight Job satisfaction Positive (+) Temporary employment status Negative (-) Resource adequacy Positive (+) Patients per bed Positive (+) Pseudo R 2 = N=149 (Nurses) p 0.05 Intention to Leave Current Job Nurses were more likely to intend to leave their current job if they were required to resequence their work frequently, if there was a higher proportion of agency hours worked on their ward and if demand for nursing care per day exceeded supply. Nurses who had worked longer and who were satisfied with their job were less likely to plan to leave. Nurses indicating they had more time to deliver care per shift were more likely to leave; a finding worth further study. Those working on wards with a higher proportion of patients waiting for a care facility, were less likely to intend to leave (Table 91). TABLE 91 LOGISTIC REGRESSION ON INTENT TO LEAVE CURRENT JOB Direction β Weight Nursing demand/supply Positive (+) Proportion of patients waiting for a care facility Negative (-) Years worked as a nurse Negative (-) Job satisfaction Negative (-) Proportion of hours worked by agency Positive (+) Resequencing of work in response to others Positive (+) More time available to deliver care Positive (+) Pseudo R 2 =0.339 N=149 (Nurses) p 0.05 Correlation between the factors identified in the logistic regression analysis and the individual outcome variables showed similar relationships. Some variables, such as the proportion of hours worked by agency staff, displayed relationships with the outcomes even though they were not statistically significant in the regression models (Table 92). 108 FINDINGS

109 TABLE 92 CORRELATION OF FACTORS IDENTIFIED IN LOGISTIC REGRESSION ON INDIVIDUAL NURSE OUTCOMES Kendall's τ Job satisfaction Satisfaction with nursing Intent to leave current job Job satisfaction (**) -.195(*) Satisfaction with nursing.357(**) Intent to leave current job -.195(*) Number shifts missed work -.141(*) Resource adequacy.272(**).181(**) -.145(*) Proportion of hours worked by agency (*).253(**) Time available to deliver care Resequencing of work in response to others Temporary employment (**) Years worked as a nurse (*) Total nursing hours.232(**).161(*) Nursing demand/supply Patients per bed.150(*).197(**) Proportion of patients waiting for a care facility (*) -.244(**) * Correlation is significant at the 0.05 level (2-tailed) **Correlation is significant at the 0.01 level (2-tailed) As expected, job satisfaction was positively correlated with satisfaction with nursing and total nursing hours as found in the regression model. In addition, increases in resource adequacy were positively correlated with job satisfaction, while the number of shifts missed this week was negatively correlated with job satisfaction and were not included in the regression model. Increases in significant variables with a positive τ- value are likely to result in improved job satisfaction. However, the number of patients per bed was positively correlated with job satisfaction (τ=.150, p.05) reflecting earlier findings that nurses are happier and more satisfied when they are busier. Highly significant correlations between satisfaction with nursing and its predictor variables as in the regression model (Table 90) were as expected. In addition to these, satisfaction with nursing was positively correlated with total nursing hours and the proportion of patients waiting for a care facility. The proportion of hours worked by agency staff was negatively correlated with satisfaction with nursing. In regard to intention to leave current job, although resource adequacy was not a significant predictor in the regression model, it is significantly correlated with Intention UNIVERSITY OF TECHNOLOGY, SYDNEY 109

110 to leave (τ =-.145, p.05) and indicates that as resource adequacy improves the intention to leave the current job declines. 110 FINDINGS

111 5. Limitations Any study using standard administrative data is limited to what is in the data. In this instance, the administrative data mined in the longitudinal study were supplemented by the cross-sectional data collection to provide information on variables that are simply not part of standard data collection. These were particularly those variables concerned with the quality of the working environment and the nursing outcomes. In previous studies it has been shown that there is wide variation in a range of the variables captured in both the longitudinal and dross-sectional data. This potential variation should be considered when applying these findings outside the sampled hospitals. The longitudinal data were essentially the entire population of patients for the period studied and the entire record of nurses working for the periods available. Still, the data cover only about two years. Similarly, the cross-sectional data include all eligible nursing units after maternity, newborn, pediatric and psychiatric units were excluded. There are several limitations in regard to the longitudinal analysis: Limited amount of usable data Lack of large learning (reference) set for threshold contrast method Lack of direct link between ward and adverse event Potential seasonal effects for data time span As discussed earlier, the time and place of OPSN could not be determined in the data so attribution to the nursing unit is a limitation. Instrument reliability and validity have been reported. A high proportion of consenting nursing staff responded to the surveys overall (71%), but it is not known whether important responders declined to participate. The sampling period for the cross-sectional study was only one week per nursing unit and although it appeared to be similar to longitudinal data in terms of skill mix, it is not known how representative that week might have been in regard to patient type and the remainder of the the unit s life. This short sample period, unknown data collection issues or other factors may have been related to the very low number of patient adverse events collected. UNIVERSITY OF TECHNOLOGY, SYDNEY 111

112 6. Summary and Discussion Synopsis of Objectives, Design and Measures This study used a combination of longitudinal and cross-sectional data collection to examine nursing workload (and changes therein), patient acuity and length of stay, skill mix and the working environment and their relationships with patient outcomes in two hospitals in the Australian Capital Territory. The unit of analysis was the nursing ward, the business and operational unit of the hospital. The project was designed to provide information to assist policy development in the ACT toward innovations in care delivery. In particular it was to determine approaches to staffing which would provide for the health needs of the population, achieve high standards of care and enhance patient outcomes. The focus was on medical/surgical nursing units where the majority of nurses work. Patient data were obtained on all discharges from the two hospitals for two financial years, 2005 and 2006 (approximately 185,000 hospital morbidity records of which 40,538 contributed to the final analysis). Nursing payroll data were obtained for roughly the same period. Payroll data allowed tracking nurses to the wards on which patients were nursed. Eventually 398 ward months of data were used in the analysis. Casemix control to the ward level provided risk adjustment. Twelve Outcomes Potentially Sensitive to Nursing (OPSN) adverse events coded in administrative data were studied along with length of stay as an outcome. Cross-sectional data which involved original data collection at ward level, took place over a three week period at Canberra Hospital and two weeks at Calvary Hospital toward the end of the longitudinal data collection. Surveys of nurses collected data on job satisfaction, perception of the ward working environment (including environmental complexity), and perception of the extent to which work was accomplished fully and on time. Additional data collection at ward level obtained data on staffing and patient outcomes as falls and medication errors. Statistical treatment was designed to determine patterns of nursing resources and their relationship to patient outcomes, and in the case of cross-sectional data, nurse outcomes such as job satisfaction. Where appropriate, comparisons to a similar study conducted in New South Wales (NSW) were made. 112 SUMMARY AND DISCUSSION

113 Discussion of Results Sixteen medical/surgical nursing units were included in the sample. Relevant nursing data were available for 15 of these units, two of which were collapsed into one for analysis for statistical reasons leaving 14 units as the sample for the longitudinal analysis. There were 16 units in the cross-sectional study, but two did not provide complete roster data. Over time, nursing workload as measured by nursing hours per patient hour increased, especially in one of the hospitals; the ratio of nurse hours on ward to patient hours on ward decreased. Skill mix measured as the percentage of RN hours worked was quite variable ranging from 50% to 80% at one hospital and 54% to 84% at the other. Skillmix was lower in wards with aged or rehabilitation casemix, higher in specialty surgical wards. This is not an unexpected finding but it raises questions about the conventional wisdom that decrees a lesser skilled workforce for aged or infirm patients, many of whom may actually be more frail than surgical specialty patients. Patient movements can contribute to nursing workload. The findings here indicate the number of wards per patient episode over the two years (average length of stay = 4.0) were on average 1.24 and 1.32 at the two hospitals, considerably lower than the NSW result of In addition the number of patients per bed per day was on average one, compared to 1.25 in NSW. This may reflect better bed management strategies. In terms of the nursing hours required and provided, there was an average difference of 0.5 hours per patient day, less than in NSW data. Of interest is that in the ACT, both the hours of nursing care required per patient day and nursing hours per patient day provided were higher in the A group hospital than in NSW. The reverse is true of the B group hospital, where hours of nursing care required per patient day and nursing hours per patient day provided were less than in NSW. Nursing workload in ACT is influenced by the number of different AR-DRGs per nursing unit. There is a wide degree of variability ranging from DRGs per ward, from a possible range of 613. It cannot be expected that nurses are equally skilled or comfortable caring for a wide range of patient types, each with its treatments, procedures, protocols, medications and physician teams. Smaller hospitals, such as found in ACT, cannot create the number of specialty units found in larger hospitals, a fact that managers need to appreciate. The nursing workload will always feel heavier in UNIVERSITY OF TECHNOLOGY, SYDNEY 113

114 wards with a large number of different AR-DRGs. Still, the role of casemix in staffing has been little identified nor studied. As we have found in previous research, there was considerable variation in nursing unit staffing and skill mix over only a two year period, variation that was neither seasonal nor predictable. There should be no expectation that every nursing unit has the same ratio of nursing hours per patient day nor the same skill mix for different mixes of cases. However, such variation itself increases nursing workload and may contribute to job dissatisfaction. Indeed, the cross-sectional results showed that adequacy of nursing resources was one of the stronger predictors of nursing job satisfaction. Decisions about how to titrate nursing resources to patient types should be made consciously, not simply allowed to vary with the ability of the nurse manager to advocate for resources or the constraints imposed by a tight labour market. Indeed, our analysis suggested that parity in nursing staffing could be achieved with modest increases in resources. Analysis also showed that increased skill mix was associated with decreased length of stay, although the relationship was not strong in this sample. It has been observed that physicians admit patients to hospital but nurses get them out. Yet skill mix has rarely been considered in itself an efficiency investment. When patient outcomes as Outcomes Potentially Sensitive to Nursing (OPSN) were examined, it was found that increasing RN hours by 10% could produce decreases in the adverse event rates studied from 11% to 45%. While we did not attempt cost analyses in this study, it is known that adverse outcomes such as hospital-acquired decubiti, infections etc. increase length of stay and cost. It should be in hospitals interest to invest in the resource(nurses) to lower such rates, not only for financial reasons but more importantly, to minimise harm to patients. The cross-sectional data amplified these findings. Comparisons were made where appropriate with our New South Wales study. This is a new area of inquiry, however, so the NSW findings cannot be taken as the gold standard they are simply descriptive of the situation as the data revealed it in the prior study. It was not possible to determine the impact that medicaton endorsed ENs might have on medication errors. 114 SUMMARY AND DISCUSSION

115 The nursing work environment in ACT was rated as somewhat better by the ACT nurses than NSW nurses rated theirs, and, largely because there were only two hospitals in the ACT study, we did not find the enormous variations in nursing units that we had found in NSW. Still, with a sample of only 16 units, there was a striking amount of variation in nearly every measure. Nursing supply/demand analysis showed that only 25% of the units were in balance, with the rest showing a deficit of nursing for patient requirements. When nurses reported numbers of tasks delayed or not done, these figures were related to a perception of resource adequacy staffing, support services etc. That is, where there were adequate resources, fewer tasks were reported undone or delayed. It was interesting to note, as it had been in NSW, that nurses on wards with larger proportions of patients from care facilities and wards with a higher proportion of agency staff and overtime reported more work undone at the end of shift. These are wards that are stressed; the necessity for involving agency staff is a signal to managers that something is not right on the ward with respect to staffing. The finding about patients from care facilities might signal a systemic problem of coordination of care across institutions or perhaps an issue of quality of facility care. A higher proportion of nurses in ACT reported experiencing a threat of violence or physical violence than did nurses in NSW but less emotional abuse. The perpetrators were most often patients or families. This is an under-appreciated aspect of nursing workload. In terms of nurse outcomes, 71.5% of nurses were satisfied with their current job and this was related to having adequate resources to do their job and a higher overall amount of nursing hours. More than three quarters (79.5%) were satisfied with nursing and again this was related to having adequate resources to do the job. While workload is an important factor in job satisfaction and satisfaction with nursing, there is evidence that nurses were more satisfied when they were busier (measured as higher patient turnover per bed). In terms of workforce planning, 74% of nurses had no intention of leaving their current job in the next 12 months and as resource adequacy improves, the intention to leave the current job declines. Overall, the study of ACT hospitals reveals hitherto unknown patterns in nursing staffing, the work environment and patient outcomes. The study suggests that to UNIVERSITY OF TECHNOLOGY, SYDNEY 115

116 successfully manage a hospital system requires an understanding of the nature of the work and a commitment to matching resources to workload. The workload/staffing software used in this study was developed from the NSW study and its test here shows interesting possibilities. 116 SUMMARY AND DISCUSSION

117 7. References Access Economics. (2004a). Employment Demand in Nursing Occupations. Canberra: Australian Government Department of Health and Ageing. Access Economics. (2004b). Employment Demand in Nursing Occupations Canberra: Dept. Health & Ageing. ACT Health / ANF. (2007). A.C.T. Public Sector Nursing and Midwifery Staff Union Collective Agreement Canberra: ACT Health & Australian Nursing Federation. Adams, A., & Duffield, C. (1991). The value of drills in developing and maintaining numeracy skills in an undergraduate nursing programme. Nurse Education Today, 11(3), AHWAC. (2002a). The Critical Care Nurse Workforce in Australia (No ). Sydney: Australian Health Workforce Advisory Committee. AHWAC. (2002b). The Midwifery Workforce in Australia (No ). Sydney: Australian Health Workforce Advisory Committee. AHWAC. (2004). The Australian Nursing Workforce - An Overview of Workforce Planning (No ). Sydney: Australian Health Workforce Advisory Committee. AIHW. (2005). Nursing and midwifery labour force 2003 (No. HWL 31). Canberra: Australian Institute of Health and Welfare. AIHW. (2006). Nursing and midwifery labour force 2004 (No. HWL 37). Canberra: Australian Institute of Health and Welfare. AIHW. (2007). Australian hospital statistics (No. HSE 50). Canberra: Australian Institute of Health and Welfare. Aiken, L. H., Clarke, S. P., & Sloane, D. M. (2002). Hospital staffing, organization, and quality of care: cross-national findings. International Journal for Quality in Health Care, 14(1), 5. Aiken, L. H., Clarke, S. P., Sloane, D. M., Sochalski, J. A., Busse, R., Clarke, H., et al. (2001). Nurses' reports on hospital care in five countries: the ways in which nurses' work is structured have left nurses among the least satisfied workers, and the problem is getting worse. Health Affairs, 20(3), Aiken, L. H., Lake, T. E., Sochalski, J., & Sloane, D. M. (1997). Design of an Outcomes Study of the Organization of Hospital AIDS Care. Research in the Sociology of Health Care, 14, Aiken, L. H., & Patrician, P. A. (2000). Measuring organizational traits of hospitals: the Revised Nursing Work Index. Nursing Research, 49(3), Aiken, L. H., & Sloane, D. M. (1997). Effects of organizational innovations in AIDS care on burnout among urban hospital nurses. Work and Occupations, 24(4), Aiken, L. H., Smith, H. L., & Lake, E. T. (1994). Lower Medicare mortality among a set of hospitals known for good nursing care. Medical Care, 32(8), 771. Aiken, L. H., Sochalski, J., & Anderson, G. F. (1996). Downsizing the hospital nursing workforce. Health Affairs, 15(4), American Nurses' Association. (1997). Implementing Nursing's Report Card: A Study of RN Staffing, Length of Stay and Patient Outcomes. Washington: ANA. ARHRC. (2005). Submission: Productivity Commission Health Workforce Study. Retrieved July 2005, from Australian Bureau of Statistics. (2007 -a). Australian Capital Territory at a Glance [Electronic Version], from C7AD/$File/13148_2007.pdf Australian Bureau of Statistics. (2007 -b). Australian Capitol Territory in Focus [Electronic Version], from C7AD/$File/13148_2007.pdf Baumann, A. O., Giovannetti, P., O'Brien-Pallas, L. L., Mallette, C., Deber, R., Blythe, J., et al. (2001). Healthcare restructuring: the impact of job change. Canadian Journal of Nursing Leadership, 14(1), UNIVERSITY OF TECHNOLOGY, SYDNEY 117

118 Baumann, A. O., O'Brien-Pallas, L., Armstrong-Stassen, M., Blyth, J., Bourbonnai, R., Cameron, S., et al. (2001). Commitment and care: The benefits of a health workplace for nurses, their patients and the system. Ottawa: Canadian Health Services Research Foundation and The Change Foundation. Birch, S., O'Brien-Pallas, L., Alksnis, C., Murphy, G. T., & Thomson, D. (2003). Beyond demographic change in human resources planning: an extended framework and application to nursing. Journal of Health Services Research and Policy, 8(4), Boyle, S. M. (2004). Nursing Unit Characteristics And Patient Outcomes. Nursing Economic$, 22(3), 111. Buerhaus, P. I. (1997). What is the harm in imposing mandatory hospital nurse staffing regulations? Nursing Economic$, 15(2), Buerhaus, P. I. (1999). Lucian Leape on the causes and prevention of errors and adverse events in health care. Image the Journal of Nursing Scholarship, 31(3), Chagnon, M., Audette, L. M., Lebrun, L., & Tilquin, C. (1978). Validation of a patient classification through evaluation of the nursing staff degree of occupation. Medical Care, 16(6), 465. Cho, S. H., Ketefian, S., Barkauskas, V. H., & Smith, D. G. (2003). The effects of nurse staffing on adverse events, morbidity, mortality, and medical costs. Nursing Research, 52(2), Clarke, S. P., & Aiken, L. H. (2003). Failure to rescue. American Journal of Nursing, 103(1), Clarke, S. P., & Aiken, L. H. (2006). More nursing, fewer deaths. Quality & Safety in Health Care, 15(1), 2-3. Czaplinski, C., & Diers, D. (1998). The Effect of Staff Nursing on Length of Stay and Mortality. Medical Care, 12, Daft, R. L. (1995). Organizational theory and design (5 ed.). St Paul: MN:West Publishing. Department of Employment and Workplace Relations. (2006). Do you have an occupation in demand? Retrieved 23/05/2007, from Diers, D., Bozzo, J., Blatt, L., & Roussel, M. (1998). Understanding nursing resources in intensive care: a case study. American Journal of Critical Care, 7(2), Diers, D., & Potter, J. (1997). Understanding the unmanageable nursing unit with casemix data: a case study. Journal of Nursing Administration, 27(11), Duffield, C., O'Brien-Pallas, L. L., & Aitken, L. M. (2004). Nurses who work outside nursing. Journal of Advanced Nursing, 47(6), Duffield, C., Roche, M., O'Brien-Pallas L. L., Diers, D. K., Aisbett, C., King, M., et al. (2007). Glueing it together: nurses, their work environment and patient safety. Sydney: Centre for Health Services Management, UTS. Estabrooks, C. A., Midodzi, W. K., Cummings, G. G., Ricker, K. L., & Giovannetti, P. (2005). The impact of hospital nursing characteristics on 30-day mortality. Nurs Res, 54(2), Estabrooks, C. A., Tourangeau, A. E., Humphrey, C. K., Hesketh, K. L., Giovannetti, P., Thomson, D., et al. (2002). Measuring the hospital practice environment: a Canadian context... revised Nursing Work Index (NWI-R). Research in Nursing & Health., 25(4), Fabb, W. E., Chao, D. V., & Chan, C. S. (1997). The trouble with family medicine. Fam. Pract., 14(1), Falco, S. M., & Lobo, M. L. (1990). Martha E. Rogers. In J. George (Ed.), Nursing Theories: The Base for Professional Nursing Practice (pp ): Prentice-Hall. Finn, C. P. (2001). Autonomy: an important component for nurses' job satisfaction. International Journal of Nursing Studies, 38(3), Freeman, T. (2005). Assessing the role of formal and informal caregivers in the current tertiary healthcare system: Factors influencing care roles and satisfaction with care. Unpublished Doctoral Dissertation, University of Toronto, Toronto. Goldstein, D. E. (2003). Digital e-care technology delivers quality healthcare. Internet Healthcare Strategies, 5(2), REFERENCES

119 Goldstein, H. (2003). Multilevel Statistical Models (2nd ed.). New York: John Wiley & Sons. Goodwin, M., & Hawkins, A. (1990). PAIS dependency system: a validation. Australian Journal of Advanced Nursing., 7(3), Graf, C. M., Millar, S., Feilteau, C., Coakley, P. J., & Erickson, J. I. (2003). Patients' needs for nursing care: beyond staffing ratios. Journal of Nursing Administration, 33(2), Grillo-Peck, A. M., & Risner, P. B. (1995). The effect of a partnership model on quality and length of stay. Nursing Economic$, 13(6), , 374. Hovenga, E. J. S. (1996). Patient Assessment and Information System (PAIS).Unpublished manuscript, Rockhampton, Australia. Ingersoll, G. L. (1998). Organizational Redesign: Changing Educational Needs of Midlevel Nurse Administrators. Journal of Nursing Administration, 28(4), Institute of Medicine. (2004). Keeping patients safe: transforming the work environment of nurses. Washington: The Institute of Medicine of the National Academies. IOM. (1999). To Err is Human: Building A Safer Health System. Washington: Institute of Medicine. IOM. (2001). Crossing the Quality Chasm: A New Health System for the 21st Century. Washington: Institute of Medicine. Jelinek, R. C. (1967). A structural model for the patient care operation. Health Serv Res., 2(3), Jelinek, R. C. (1969). An operational analysis of the patient care function. Inquiry, 6, Jiang, H. J., Stocks, C., & Wong, C. J. (2006). Disparities between two common data sources on hospital nurse staffing. Journal of Nursing Scholarship, 38(2), Kane, R. L., Shamliyan, T., Mueller, C., Duval, S., & Wilt, T. (2007). Nursing Staffing and Quality of Patient Care. Evidence Report/Technology Assessment No. 151 (Prepared by the Minnesota Evidence based Practice Center under Contract No ). Rockville, MD: Agency for Healthcare Research and Quality. Karmel, T., & Li, J. (2002). The Nursing Workforce Canberra: National Review of Nursing Education. Kovner, C., & Gergen, P. (1998). Nurse Staffing and Adverse Events Following Surgery in US Hospitals. Image the Journal of Nursing Scholarship, 30(4), Kramer, M., & Hafner, L. P. (1989). Shared values: impact on staff nurse job satisfaction and perceived productivity. Nursing Research, 38(3), Kramer, M., & Schmalenberg, C. (1991). Job satisfaction and retention insight for the 90's: Part I. Nursing, 21(3), Laschinger, H. K. S., Finegan, J., Shamian, J., & Wilk, P. (2004). A longitudinal analysis of the impact of workplace empowerment on work satisfaction. Journal of Organizational Behavior, 25(4), 527. Laschinger, H. K. S., & Leiter, M. P. (2006). The impact of nursing work environments on patient safety outcomes. Journal of Nursing Administration, 36(5), Laszlo, E. (1975). The systems view of the world: The natural philosophy of the new developments in the sciences. Oxford: Blackwell. McCloskey, B. A., & Diers, D. (2005). Effects of New Zealand's health reengineering on nursing and patient outcomes. Medical Care, 43(11), ,. McGillis-Hall, L. (1997). Staff mix models: complementary or substitution roles for nurses. Nursing Administration Quarterly, 21(2), McKenna, H. P., Hasson, F., & Keeney, S. (2004). Patient safety and quality of care: the role of the health care assistant. Journal of Nursing Management, 12(6), Microsoft Corporation. (2003). Microsoft Office Access 2003 (Version SP1). Redmond: Microsoft Corporation. Nadler, D. A., & Tushman, M. L. (1980). A Model for Diagnosing Organizational Behavior. Organizational Dynamics, 9, Needleman, J., Buerhaus, P., Mattke, S., Stewart, M., & Zelevinsky, K. (2002). Nursestaffing levels and the quality of care in hospitals. New England Journal of Medicine, 346(22), Needleman, J., Buerhaus, P. I., Mattke, S., Stewart, M., & Zelevinsky, K. (2001). Nurse staffing and patient outcomes in hospitals. Boston: Harvard School of Public Health. UNIVERSITY OF TECHNOLOGY, SYDNEY 119

120 O'Brien-Pallas, L. L., Baumann, A. O., Donner, G., Tomblin-Murphy, G., Lochhaas-Gerlach, J., & Luba, M. (2001). Forecasting models for human resources in health care. Journal of Advanced Nursing, 33(1), O'Brien-Pallas, L. L., Doran, D. I., Murray, M., Cockerill, R., Sidani, S., Laurie-Shaw, B., et al. (2001). Evaluation of a Client Care Delivery Model, Part 1: Variability in Nursing Utilization in Community Home Nursing. Nursing Economic$, 19(6), 267. O'Brien-Pallas, L. L., Doran, D. I., Murray, M., Cockerill, R., Sidani, S., Laurie-Shaw, B., et al. (2002). Evaluation of a Client Care Delivery Model, Part 2: Variability in Client Outcomes in Community Home Nursing. Nursing Economic$, 20(1), 13. O'Brien-Pallas, L. L., Irvine, D., Peereboom, E., & Murray, M. (1997). Measuring nursing workload: understanding the variability. Nursing Economic$, 15(4), O'Brien-Pallas, L. L., Thomson, D., Alksnis, C., & Bruce, S. (2001). The economic impact of nurse staffing decisions: time to turn down another road? Hospital Quarterly, 4, O'Brien-Pallas, L. L., Thomson, D., McGillis-Hall, L., Pink, G. H., Kerr, M., Wang, S., et al. (2004). Evidence-based Standards for Measuring Nurse Staffing and Performance: Canadian Health Services Research Foundation. Orne, R. M., Garland, D., O'Hara, M., Perfetto, L., & Stielau, J. (1998). Caught in the cross fire of change: nurses' experience with unlicensed assistive personnel. Applied Nursing Research, 11(3), Person, S. D., Allison, J. J., Kiefe, C. I., Weaver, M. T., Williams, O. D., Centor, R. M., et al. (2004). Nurse Staffing and Mortality for Medicare Patients with Acute Myocardial Infarction. Medical Care, 42(1), Plummer, V. (2005). Analysis of patient dependency data utilising the Trendcare system. Unpublished Unpublished PhD Thesis, Monash University, Frankston, Victoria. Putt, A. M. (1978). General systems theory applied to nursing. Boston: Little, Brown. Rafferty, A. M., Ball, J., & Aiken, L. H. (2001). Are teamwork and professional autonomy compatible, and do they result in improved hospital care? Qual Saf Health Care, 10(90002), 32ii-37. Schofield, D., & Beard, J. (2005). Baby boomer doctors and nurses: demographic change and transitions to retirement. Medical Journal of Australia, 183, Sheshkin, D. J. (2000). Handbook of parametric and nonparametric statistical procedures (2nd ed.). Boca Raton: Chapman & Hall/CRC. Shullanberger, G. (2000). Nurse staffing decisions: An integrative review of the literature. Nursing Economic$, 18(3), Silber, J. H., Williams, S. V., Krakauer, H., & Schwartz, J. S. (1992). Hospital and patient characteristics associated with death after surgery. A study of adverse occurrence and failure to rescue. Medical Care, 30(7), SPSS Inc. (2005). SPSS for Windows (Version ). Chicago: SPSS Inc. SPSS Inc. (2006). SPSS for Windows (Version ). Chicago: SPSS Inc. Stamps, P. L., & Piedmont, E. B. (1986). Nurses and Work SatisfactionAn Index for Measurement Tillman, H. J., Salyer, J., Corley, M. C., & Mark, B. A. (1997). Environmental turbulence staff nurse perspectives. Journal of Nursing Administration, 27(11), Tourangeau, A. E. (2002). Nursing skill mix and experience reduce patient mortality. Hospital Quarterly, 5(3), 19. Tourangeau, A. E., Doran, D., Pringle, D., O'Brien-Pallas, L., McGillis-Hall, L., Tu, J., et al. (2006). Nurse Staffing and Work Environments: Relationships with hospital level outcomes. Toronto: Canadian Health Services Research Foundation. Trend Care Systems Pty Ltd. (2004). Product Overview. Retrieved 15th June 2004, 2004, from Unruh, L. Y., & Fottler, M. D. (2006). Patient turnover and nursing staff adequacy. Health Services Research, 41(2), , 2006 Apr. VDHS. (1999). Nurse labourforce projections: Victoria Melbourne: Public Health and Development Division, Victorian Government Department of Human Services REFERENCES

121 8. Appendices Appendix 1 Theoretical Foundations Appendix 2 Format for Admitted Patient Care Data Appendix 3 Format for Ward Episode Data Appendix 4 Matching Wards Appendix 5 OPSN Analysis Appendix 6 Staffing of the Study Wards Appendix 7 Instruments for Cross-sectional Component UNIVERSITY OF TECHNOLOGY, SYDNEY 121

122 Appendix 1 Theoretical Foundations (Linda O Brien-Pallas) A theoretical framework guides this study. At the meso and micro level, the Patient Care System Model (Figure 11) developed by O Brien-Pallas and colleagues (2001; 2001; 2004) is used to guide the analysis of the relationship among the variables studied at the nursing subunit level and the hospital level. FIGURE 11 PATIENT CARE DELIVERY MODEL INPUTS Patient Characteristics Demographics Medical diagnoses Admission type Pre-operative clinic Nurse Characteristics Demographics Professional status Employment status Education Experience System Characteristics Geographic location Hospital size Unit size, type, patient mix Occupancy System Behaviours Workload Nurse-to-patient ratios Proportion of RN worked hours Continuity of care/shift change Unit instability Overtime Use of agency & relief staff # of units nurse works on Non-nursing tasks Patient Care Delivery Model (O Brien-Pallas et al., 2004) Interventions Patient Care Delivery System THROUGHPUTS Perceived Work Environment INTERMEDIATE OUTPUTS Worked hours Utilization Environmental Complexity Factors Resequencing of work in response to others Unanticipated delays due to changes in patient acuity Characteristics & composition of caregiving team Feedback OUTPUTS Patient Outcomes Medical consequences, including mortality status Resource intensity weight Nurse Outcomes Autonomy & control Job satisfaction Relationships with MDs Violence at work System Outcomes Length of stay Cost per resource intensity weight Quality of patient care Quality of nursing care Interventions delayed Interventions not done Absenteeism Intent to leave The framework considers aspects of patient, nurse, hospital and unit specific inputs (resources), which influence throughputs within the complexity of the environment. These independent variables combine to influence nurse patient and system outcomes. Consistent with General Systems Theory (GST) the patient, nurse and system outputs 122 APPENDICES

123 serve as dependent variables to the system as a whole (O'Brien-Pallas et al, 2001) but can also serve as independent variables for other analysis of the system. The underlying assumptions of the GST are as follows: GST is a general science of wholeness, concerned with the problems of organisation and dynamic interactions manifested in the difference of the behaviour of the parts when isolates (Falco & Lobo, 1990; Freeman, 2005; Putt, 1978). GST believes an organisation must be open and continually change, adapt and interact to meet the challenges posed by both the internal and external environment, in order to meet the needs of their clients and stakeholders (Shortell et al, 1991; Daft, 1995; Freeman, 2005). An open system interacts with the environment, taking input from the environment, subjects it to some form of transformation process and then produces an output (Nadler & Tushman, 1980). The holistic view that GST provides, allows a comprehensive and specific view of the system or individual under investigation, never as the mechanistic accumulation of parts in segregated causal relationships (Laszlo, 1975). A system is characterised by a number of constraining but interacting factors, each fulfilling a function not accomplished by the others which connect through communication and feedback mechanisms (Fabb, Chao, & Chan, 1997). Basic concepts of GST are those of: 1) nonsummativity, 2) input, throughput and output, 3) entropy, 4) equifinality/ multifinality, 5) equilibrium, 6) feedback and 7) control (Fabb et al., 1997; Freeman, 2005; Putt, 1978). GST concepts can be represented in the following propositions: 1. A system is a set of interacting and interrelated parts. A system is more than a sum of its parts; its characteristics derive from the association among the parts and from the system s connection with the environment (Fabb et al., 1997; Freeman, 2005). In this study unit characteristics including patient characteristics, staff characteristics, system characteristics and behaviours influence throughput including environmental complexity, interventions and perceived work environment. These in turn influence intermediate outcomes including workload and staff utilisation and these in turn influence patient, nurse and system outputs. 2. Open systems have permeable boundaries that continually engage in the input, throughput and output of matter, energy and information (Fabb et al., UNIVERSITY OF TECHNOLOGY, SYDNEY 123

124 1997; Freeman, 2005). In this study the system is conceptualised as interconnecting parts including nursing, patient and system variables and their relationship workload and staff utilisation recognising these in turn influence patient, nurse and system outputs. 3. Systems are capable of negative entropy, that is, systems can survive and grow rather than decay and die, if they are able to work out mutually beneficial relationships with their environment (negentropic) (Fabb et al., 1997). The process of entropy is universal, existing in both closed and open systems (Putt, 1978). In this study the system will be explored to identify factors that influence workload and patient nurse and system outputs. Through this study areas for improvement within work systems will be identified and positive change maybe recommended. 4. When acting on a system of interrelated parts, the effects cannot be gauged on knowledge of inputs alone but must include the entire system. The overall pattern must be considered, in order to determine the results of specific stimulus/ stimuli. In other words, the results of equifinality and multifinality must be taken into account (Freeman, 2005). In this study the nursing, patient and system inputs will be viewed within the broader scope of the unit throughputs and the nurse, patient and system outputs. 5. Systems tend to maintain steady states of dynamic equilibrium, in which conflicting pressures are balanced. Such steady states have the property of evolution; the more the system is threatened with disequilibrium, the more resources it will deploy to maintain or restore balance (Fabb et al., 1997; Freeman, 2005). In this study the factors that threaten nurses workload and patient nurse and system outputs will be explored. Further, the current practice and overall system will not change unless this research is conducted. 6. To maintain a steady state, open systems need adaptive processes such as feedback loops and control. This allows the system to detect applicable changes in the internal and external environment and adjust appropriately (Fabb et al., 1997; Freeman, 2005). In this study a feedback loop will be utilised to link the outputs to the inputs and throughputs to demonstrate the openness of the system. Consistent with systems theory (Jelinek, 1967), these dependent variables feed back into the system and, in turn, affect future inputs. This model allows the researcher 124 APPENDICES

125 to gain comprehension of the nursing system unit and the broader components of the patient care system. It permits the management of complex interdependent relationships that exist in the patient care system. Jelinek (1969), described the patient care systems model comprising inputs and outputs that can be affected by workload, the environment, and organisation factors. Inputs are postulated to refer to resources, both personnel and physical, involved in patient care. Organisational factors capture the form of organisation used in delivering patient care and include rules and policies. Workload factors explore the workload the patient imposes on the input resources. Environmental factors include factors that may affect patient care such as services a hospital offers. Output describes patient outcomes in terms of the quality and quantity of patient care delivered (O'Brien-Pallas et al, 2004). UNIVERSITY OF TECHNOLOGY, SYDNEY 125

126 Appendix 2 Format for Admitted Patient Care Data Name Size Label Bus.Rules addttime Date17 Date and Time of admisison admdate & admtime ageyrs N3 Age in years dateborn, addttime agedays N3 Age in days of infants aged under 1 year drg51 S4 DRG5.1 grouped by IMS epis N8 Episode number from hospital dateborn, addttime According to 3M Grouper Casemix Expert for Windows Version Code/ Library_Table hospid N2 Hospital Identification Y pin N8 Patient ID from hospital sex N1 Sex of patient Y spdttime Date17 Date and Time of separation sepdate & septime spyrmth S7 Financial Year and Month of separation spdttime pdx S7 Primary diagnosis ICD-10 code dx2 S7 Additional diagnosis - 2 dx3 S7 Additional diagnosis - 3 dx4 S7 Additional diagnosis - 4 dx5 S7 Additional diagnosis - 5 dx6 S7 Additional diagnosis - 6 dx7 S7 Additional diagnosis - 7 dx8 S7 Additional diagnosis - 8 dx9 S7 Additional diagnosis - 9 dx10 S7 Additional diagnosis - 10 dx11 S7 Additional diagnosis - 11 dx12 S7 Additional diagnosis - 12 dx13 S7 Additional diagnosis - 13 dx14 S7 Additional diagnosis - 14 dx15 S7 Additional diagnosis - 15 dx16 S7 Additional diagnosis - 16 dx17 S7 Additional diagnosis - 17 dx18 S7 Additional diagnosis - 18 dx19 S7 Additional diagnosis - 19 dx20 S7 Additional diagnosis - 20 dx21 S7 Additional diagnosis - 21 dx22 S7 Additional diagnosis - 22 dx23 S7 Additional diagnosis - 23 dx24 S7 Additional diagnosis - 24 dx25 S7 Additional diagnosis - 25 dx26 S7 Additional diagnosis - 26 dx27 S7 Additional diagnosis APPENDICES

127 Name Size Label Bus.Rules dx28 S7 Additional diagnosis - 28 dx29 S7 Additional diagnosis - 29 dx30 S7 Additional diagnosis - 30 dx31 S7 Additional diagnosis - 31 p1 S8 Procedure 1 ICD-10 code p2 S8 Procedure 2 p3 S8 Procedure 3 p4 S8 Procedure 4 p5 S8 Procedure 5 p6 S8 Procedure 6 p7 S8 Procedure 7 p8 S8 Procedure 8 p9 S8 Procedure 9 p10 S8 Procedure 10 p11 S8 Procedure 11 p12 S8 Procedure 12 p13 S8 Procedure 13 p14 S8 Procedure 14 p15 S8 Procedure 15 p16 S8 Procedure 16 p17 S8 Procedure 17 p18 S8 Procedure 18 p19 S8 Procedure 19 p20 S8 Procedure 20 p21 S8 Procedure 21 p22 S8 Procedure 22 p23 S8 Procedure 23 p24 S8 Procedure 24 p25 S8 Procedure 25 p26 S8 Procedure 26 p27 S8 Procedure 27 p28 S8 Procedure 28 p29 S8 Procedure 29 p30 S8 Procedure 30 p31 S8 Procedure 31 Code/ Library_Table UNIVERSITY OF TECHNOLOGY, SYDNEY 127

128 Appendix 3 Format for Ward Episode Data Name Size Label Code/Library_Table epis N8 Episode no. from admitted patient care dataset hospid N2 Hospital Identification Y pin N8 Patient ID from hospital wardid S3 Ward identifier Y wdindt S8 Date patient entered ward wdintm S4 Time patient entered ward trtype S1 Type of ward transfer Y finyr S4 Financial year of ward transfer 128 APPENDICES

129 Appendix 4 Matching Wards ( Ward Data Transfer Items ) Hospital ID WardID Wardcode WardName 82 10A Gastrointestinal Unit 82 11B Orthopaedic 82 12B Rehab and Rheumatology 82 14B Oncology 82 4HD Paediatrics High Dependancy 82 5HD Paediatrics High Dependency 82 5P2 Paediatrics - Isolation 82 5PD Paediatric Day Care on Level AX Holding Overflow Ward 82 7SU Stroke Unit 82 A/N Ante Natal 82 ACU Aged Care Unit 82 BC Birthing Centre 82 BMT Bone Marrow Transplant 82 CAR Coronary Care subacute 82 CAS Emergency 82 CCU Coronary Care (Acute) Unit 82 CLD Cardiac Lab 82 DEL Delivery Suite 82 DIA Dialysis 82 DSU Day Surgery Unit 82 EDS Extended Day Surgery Unit 82 EMU Emergency Medicine Unit 82 END Endocrinology Day Ward 82 GAS Gastro Procedure Unit 82 GAU Gynaecology Assessment Unit 82 HOC Hospital In The Home - Oncology 82 HOM Hospital In The Home 82 ICU Intensive Care Unit 82 ILU Independent Living Unit 82 L4B Paediatrics 82 L5A Adolescent 82 L6A Endocrinology, Respiratory, Cardiology 82 L6B Cardiac Surgery 82 L7A Infectious Diseases & Toxicology 82 L8A Previously renal medicine 82 L8B Renal Medicine 82 L9A Urology, Vascular Surgery 82 L9B Neurology and Neurosurgery 82 NA Post Natal Nursery A 82 NCP NCPH on ICU bed 82 NIC Neonatal Intensive Care 82 NNN Neonatal Nursery 82 ONC Oncology / Chemotherapy day bed 82 PDU Peritoneal Dialysis Unit on L8 82 PNA Post Natal A UNIVERSITY OF TECHNOLOGY, SYDNEY 129

130 Hospital ID WardID Wardcode WardName 82 PSA Psychiatry 82 PSD Psychiatry Day Ward 82 PSU Psychiatry 82 ROC Radiation Oncology Day Ward 82 SAT Satellite Dialysis Unit 82 SCN Special Care Nursery 82 NRS Northside Satellite Dialysis Unit 83 2A 23 hour recovery 83 2N Mental Health 83 3S Maternity 83 4E Surgical 83 4W Orthopaedic 83 5E Medical 83 5W Medical 83 CAB Aged Care Assessment Unit (ED) 83 CCU Coronary Care 83 CDU Clinical Decision Unit (ED) 83 CVL ACT Convalescent Unit 83 DC Day Care Unit 83 DS Delivery Suite 83 EDA Emergency Department Admission Ward 83 EDO Emergency Observation Ward 83 HH Hospital in the Home 83 HP Hospice 83 ICU Intensive Care & Intensive Care stepdown 83 NQ Special Care Nursery 83 NU Neonates on the post-natal ward 83 PEN Endoscopy Unit 83 TW Temp Ward (Public Patients admit to private hosp) 83 VAW Veterans Affairs Ward (within 5E) 83 ZM Oncology Ward 130 APPENDICES

131 Appendix 5 OPSN Analysis Steps for selecting Denominator 1. Combine All years ( ) data selecting fields: hospital, stay number, Age, LOS, MDC, separation mode, AR-DRG, same-day field. 2. Link to study hospitals and select only hospitals with nursedata = Run delete query (denomselect.sql) to exclude cases which: a. Have MDC = 14,15,19 or 20 b. Are paediatrics (ie age <18) c. Have LOS < 1 day d. Have LOS > 90 days e. Have DRG = Inappropriate diagnosis (ie. 961Z, 962Z, 963Z) 4. Add Med/Surg field 5. Update Med/Surg field where second character of AR-DRG: a. 6 and above = medical b. below 6 = surgical. 6. Get final denominators by running query finaldenom groupby year and med/surg, and count. This is in Adverse.mdb database 7. Final denominators have been calculated and added (overwritten old previous denominators) to the AdverseResults.xls should be cases 4,in DENOMSALL should be cases in DENOMS Study Hospitals Mini notes: Make DENOMSALL then copy for numerator, then delete irrelevant fields from denoms all.. then make a copy for Denoms Study hosps. Steps for selecting Numerators 1. Use table NurseworkAdverse (ie.denomintors) 2. Run queries (ngroup1 ngroup11, and Failure to Rescue) to select initial adverse and flag Group for type of adverse (1 or 0) UNIVERSITY OF TECHNOLOGY, SYDNEY 131

132 3. Do failure to rescue now before any conditions/restrictions are made. 4. Run conditional queries to un-mark any adverse events which meet conditions (ngroup1conditions ngroup11conditions) 5. Add AdEps Field and sum groups (incl failure to rescue) to get total adverse count for the patient record. 6. Delete any records with AdEps=0 7. This leaves the table of only adverse events, now called Adversework (has cases) 8. Produce final results by group-by FinYear, Med/Surg, and summing over Group1-11, failure to rescue and AdEps. 9. Final results are found in tables in excel sheet UpdatedDataAdverse.xls Notes: There are about 1000 SD fields not marked up. Change after adverse work is complete. Denominator Criteria: 1. From NSWV51 (years ) data: 2. Exclude cases which: a. Have MDC = 14,15,19 or 20 b. Are paediatrics (ie age <18) c. Have LOS < 1 day d. Have LOS > 90 days e. Have DRG = Inappropriate diagnosis (ie. 961Z) Failure to Rescue (numerator) Criteria: Patients who died (sepmode = 08) AND had either sepsis (Group #7), pneumonia (Group #3), GI bleeding (Group #5), or Shock (Group #8). Notes: Select from denominators all records with sepmode = 08 and make table called failure to rescue. Then re-run queries for select Group 7, Group 3, Group 5 and Group 8. Then delete those not involved and count per year and overall. 132 APPENDICES

133 Appendix 6 Staffing of the Study Wards FIGURE 12 WARD 1AA FIGURE 13 WARD 1AB UNIVERSITY OF TECHNOLOGY, SYDNEY 133

134 FIGURE 14 WARD 1AD FIGURE 15 WARDS 1AF & 1AI* * Note that these data were combined from 2 wards in order to retain reasonable stability in the time series, so should be viewed with caution. 134 APPENDICES

135 FIGURE 16 WARD 1AG FIGURE 17 WARD 1AH UNIVERSITY OF TECHNOLOGY, SYDNEY 135

136 FIGURE 18 WARD 1AK FIGURE 19 WARD 1AL 136 APPENDICES

137 FIGURE 20 WARD 1AM FIGURE 21 WARD 1AO UNIVERSITY OF TECHNOLOGY, SYDNEY 137

138 FIGURE 22 WARD 2AC FIGURE 23 WARD 2AE 138 APPENDICES