Title: Mortality associated with timing of admission to and discharge from ICU: A retrospective cohort study.

Transcription

1 Author's response to reviews Title: Mortality associated with timing of admission to and discharge from ICU: A retrospective cohort study. Authors: Kevin B Laupland (kevin.laupland@calgaryhealthregion.ca) Benoit Misset (BMisset@hpsj.fr) Bertrand Souweine (bsouweine@chu-clermontferrand.fr) Alexis Tabah (atabah@chu-grenoble.fr) Elie Azoulay (elie.azoulay@sls.aphp.fr) Dany Goldgran-Toledano (dany.toledano@ch-gonesse.fr) Anne-Sylvie Dumenil (anne-sylvie.dumenil@abc.aphp.fr) Aurelien Vesin (aurelien.vesin@bvra.ujf-grenoble.fr) Samir Jamali (sjamali@ch-dourdan.fr) Hatem Kallel (hatem.kallel@ch-cayenne.fr) Christophe Clec'h (christophe.clech@avc.aphp.fr) Michael Darmon (michael.darmon@chu-st-etienne.fr) Carole Schwebel (cschwebel@chu-grenoble.fr) Jean-Francois Timsit (jftimsit@chu-grenoble.fr) Version: 2 Date: 1 September 2011 Author's response to reviews: see over

2 Faculty of Medicine Departments of Medicine, Critical Care Medicine, Community Health Sciences, and Pathology and Laboratory Medicine June 23, 2011 Guian Paolo Declaro Journal Editorial Office BioMed Central Re: MS: Dear Professor/Dr. We are writing to submit our revised manuscript to BMC Health Services Research. We have revised the article according to editorial and reviewer comments. A point-by-point response is listed on the following pages. We hope that you agree that the manuscript is improved. We look forward to hearing from you. Sincerely Yours, Kevin B. Laupland MD MSc FRCPC Associate Professor, Faculty of Medicine University of Calgary

3 Editor's comment: This is an interesting piece of work that could contribute to the growing literature on the timing of discharge from ICU's. However, do note the major concerns of the reviewers and address them before the journal can make a decision on suitability for publication. In particular, please consider carefully the following points among others: 1) choice of time periods RESPONSE: As per the responses detailed to the reviewers, we have explored different definitions and have added post hoc analyses to our discussion section. Changing definitions do not appreciably change our conclusions. Given that the choice of definitions is variable among studies and influenced by personal opinions, we have elected to stay with our a priori defined time periods that have been used before in several published works. 2) further details of the study population (see comments of Reviewer 2) RESPONSE: We have added as much of this information that is possible and detail in a new limitations section in the discussion where it is not. 3) more extensive discussion on after hours discharge RESPONSE: We are limited in our ability to evaluate factors post-discharge because our database is ICU admission duration focused. Despite this we believe that our study is one of a small number that evaluate discharge timing (as opposed to just admission) and sets the stage for future studies to explore these determinants. 4) inclusion of information from more recent papers in the literature review (see comments of reviewer 2)." We have attempted to present the body of literature related to ICU admission and discharge. Critically ill patients are a distinct population from the hospital population at large and we therefore have not included non-icu references (such as one authored by the reviewer). Another reference that the reviewer suggested was overlooked as it was already referenced in our original manuscript. One reference related to ICU patients had been published only shortly before submission of our original manuscript and was missed. We have now added it as suggested. Reviewer: Antony Tobin Reviewer's report: This is a well-written article that seeks to answer questions that have important implications for ICU discharge and staffing practices. These questions have been the subject of a number of papers in the past, however doubts remain and I think asking the question in different environments is relevant and important. Major Compulsory Revision 1. I have a problem with the choice of time periods used in the analysis. The authors defined weekends as midnight Friday to midnight Sunday. This does not reflect usual work practices where weekends are generally 6pm (or there about)

4 Friday to 8am Monday. I think that it would make more sense to perform the analysis with weekends defined as such especially given the suggestion of increased mortality associated with Friday and when one of the a priori assumptions was that admission on the weekend may be associated with increased mortality. I would like to see the analysis repeated with the above definitions of weekend. RESPONSE: We agree with Dr. Tobin that choice of definitions for the after hours and weekends is challenging and is subject to differences in opinion. Indeed there is significant variability in the literature in application of these definitions and to our knowledge there is no defined overall consensus. Our definitions have been previously published (Scand J Infect Dis Dec;42(11-12):862-5, J Trauma Manag Outcomes Jul 27;3:8; J Crit Care Sep;23(3):317-24). It is important to note that we chose to present data on days of the week (ie Saturday and Sunday versus Monday to Friday) and as well as by hours (day/night) of the day separately. This is because we believe that there is a potentially clinically important difference in the days of the week and times of the day. Dr Tobin s suggestion is to compare weekends (ie Friday 1800 to Monday 0759) vs. weekday days and nights (Monday 0800 to Friday 1759), requires merging of days and times (like our definition of after hours ). We have undertaken such an analysis in order to explore a change of definitions. Using the Tobin definition the risk of ICU death is 919/5,139 (18%) with weekdays admission as compared to 469/2,141 (21%; p=0.002) for weekends; the data in our original analysis was 1035/5672 (18%) vs. 353/1708 (21%; p=0.026). Similarly, weekend discharge using the Tobin definition results in an in-hospital mortality rate of 75/1004 (7%) for weekends vs 280/4988 (6%; p=0.03). This compares to 60/903 (7%) vs. 295/5089 (6%;p=0.3) in the original analysis. Although the p-values and numbers of cases change somewhat, the results (proportions) are largely the same as with our original definition and conclusions unchanged. We have therefore kept our initial a priori defined, previously published definition and have added comments to the discussion to this end. Minor Essential 2. The results section of the abstract is wordy and confusing. The authors list a lot of univariable results prefaced by the phrase statistically significant only to discount most of these in the second last sentence reporting the logistic regression results. Then in the final sentence they do not give any statistical results for the association of afterhours discharge and death. This needs to be revised. RESPONSE: This has been revised as recommended. 3. The method appears valid and is similar to a number of papers published elsewhere on the subject apart from the comment in 1. The authors however do not state the number of cases that where not included. This study represents data from about a 5 year period from 10 ICUs yet there are only 7380 patients what proportion of all patients admitted over that period do these represent? RESPONSE: In some cases participants in the Outcomerea group have enrolled consecutive patients admitted to ICU and in others sampling has been performed where all consecutive admissions during a period of time during the year or to certain ICU beds are included. This methodological point has been added to the second paragraph of the methods section.

5 4. The discussion in relation to after-hours discharges is a bit limited. The authors offer no explanation other than reduced care on the wards as to why mortality might be higher other (eg lack of medical versus nursing care, home team not present or not contacted, no or poorer medical handover, reduced nurse ratios, discharge of patients who are expected to do poorly..) or exploration of the implications of their findings RESPONSE: Unfortunately our research was focussed in the ICU environment, and quite simply, we do not have detailed data on subsequent aspects of ward care. This is a limitation. We have added discussion to this effect in a new limitations paragraph. 5. There is no discussion of the studies limitations retrospective nature, missing data (see point 3), variation in care or practices between sites (where individual hospitals included I the multivariate analysis?) RESPONSE: We have added a new full paragraph to the discussion to review study limitations. Discretionary Revisions 6.I would have liked to see a table comparing the characteristics of patients that died with those that did not thus allowing the reader a better appreciation of the univariate statistics. RESPONSE: Whether to include tables of univariate statistics contributing to a multivariable model is an area of considerable debate among individuals and journals. We elected to not include tables of these univariate statistics in part because many of the risks for mortality in ICU patients are generally well described (severity of illness, age, diagnostic class etc). Crude risks for death associated with the primary variables of interest related to timing of admission and discharge are presented. We believe that for the sake of brevity that these are best not included, especially considering that this was only a discretionary recommendation. We would be pleased to provide these at further editorial and reviewer request. Reviewer: Massimo Gallerani Reviewer's report: - Major Compulsory Revisions abstract lin : Adults (#18 years) admitted to French ICUs participating in Outcomerea between January 2006 and November 2010 were included. And pag 6:..Outcomerea is a prospective observational study that includes detailed clinical and outcome data on patients admitted to participating French ICU.. RESPONSE: It is unclear as to the recommendation for revision or question asked by the reviewer. How many patients have been admitted in ICU directly from Emergency Room? How many have been accepted from other hospital wards? How many days (or hours) have elapsed since in-hospital admission or transfer

6 to tge ICU? RESPONSE: These data have now been added to the first paragraph of the results section. Which time of admission has been considered in the analysis of the in-hospital risk of mortality: the in-hospital or ICU admission? RESPONSE: These are detailed in the text of the manuscript where reported in each instance. The analysis of in-hospital mortality risk should consider all factors related to emergency programs organization. RESPONSE: This study was focussed on ICU admission and we do not have detailed variables for analysis on patients before and after ICU admission. Page 7 line 5: A weekend was a priori defined by the period from 00:00 Saturday to 23:59 Sunday,. Do the National holiday days of the Country have been considered additionally to the Sundays? Otherwise your could have underestimated the total number of festive days. RESPONSE: This is a limitation that has been added to the discussion limitations section. Pag 7 line 5: days as 08:00 to 17:59, and nights as 18:00 to 07:59 It could be set out the reasons that led you to make such distinction and not following the classical sub-groups of 12 hours (night 8:00 p.m. to 7:59, and day 8:00 a.m. to 7:59 p.m.). I can assume that this choice has been made on organizational issues (access to diagnostic services, operating rooms, etc...) of the hospital, but then all preliminary assessments should be related to it. Several studies (i.e., Bhonagiri D, Pilcher DV, Bailey MJ. Associated with Increased Mortality after-hours and weekend admission to the intensive care unit: a retrospective analysis. Med J Aust Mar 21; 194: ) considered the night admission after 6:00 p.m. to 05:59 a.m., which is more related to the rhythms of daily-life. RESPONSE: We agree with that the choice of definitions for the after hours and weekends is challenging and is subject to differences in opinion. Indeed there is significant variability in the literature in application of these definitions and to our knowledge there is no defined overall consensus. Our definitions have been previously published (Scand J Infect Dis Dec;42(11-12):862-5, J Trauma Manag Outcomes Jul 27;3:8; J Crit Care Sep;23(3):317-24). Pag 8 line 12:.. The crude risk for in-hospital death associated with time of ICU admission was highest in the morning between :59 as shown in Figure 1. In the figure 1 some bars of mortality rate after discharge in the night hours are missing: is there a reason for this? May be data have been collected in two hours subgroups: why?

7 RESPONSE: The data in Figure 1 has been reviewed and the figure is correct. Data were collected hourly. Where there are no bars in the figure, no deaths occurred (ie value=zero percent). Pag 8 line 13: The crude risk for in-hospital death associated with time of ICU admission was highest in the morning between :59 as shown in Figure 1. On the other hand, the crude risk for in-hospital death after ICU discharge was lowest during the daytime with rates increasing after 18:00 and highest in the early hours of the morning (Figure 1). Is not clear the reason why, for this analysis, it has been considered an interval of six hours instead of the declared night and day intervals ( and I prefer the usage of six-hours intervals. The distribution of mortality seems to really higher among the patients admitted in the morning, and on the other hand the mortality rate in the group of patients discharged in afternoon-night hour seems to the higher compared to those discharged on morning-afternoon hours. This claim should be associated with the results of the Chi-square statistical analysis. RESPONSE: We chose the intervals a priori based on periods used in prior publications so as not to risk overfitting of data (ie fitting the definitions to the analysed data). We report the hourly data for descriptive purposes for the reader with the six-hour period intended to be used descriptively only. We have revised this accordingly in the manuscript. Pag. 8 line 16 although no significant difference in mortality was associated with admission (878/3,855 (23%) vs. 865/3,525 (25%); p=0.079) during night (18:00-07:59) as compared to day (08:00-17:59) hours Are results similar even if we consider time periods of 6 hours? % in-hospital mortality Admission Discharge ,7 10, ,7 7, ,0 5, ,0 9,7 RESPONSE: This post hoc analysis has been completed and is detailed with p-values in the limitations section in the discussion. Page 8 last paragraph and the beginning of page 9:.. The crude in-hospital mortality rate varied significantly (p=0.045) according to the day of the week of ICU admission as shown in Figure 2. A statistically significant increased crude ICU-mortality (353/1708; 21% vs. 1035/5672 (18%); p=0.026) and overall hospital mortality (432/1708; 25% vs. 1311/5672; 22%; p=0.005) was observed with admission to ICU during weekends as compared to weekdays National holidays were considered as similar to Sunday? How is their contribution to this difference? It would be expect an higher rate of deaths on WE + national holidays compared to WD.

8 RESPONSE: We did not consider national holidays or other days or periods of the year when there may not be official or national holidays but staffing and usual activity may be significantly reduced (ie around the times of Christian, Jewish, and Muslim holidays) or seasons (such as summer holidays). This is a limitation that we have added to the limitations paragraph in the discussion. Pag 9 lines 6-11 and Table 1 e 2. The characteristics of patients admitted during weekdays were different from those admitted during nights and / or weekends (Table 1), and this difference is present also at the time of discharge (Table 2), but the overall length of ICU stay was not different between patients admitted on weekends or weekdays (p = 0.075) or in the afterhours (p = 0.11). This could contribute to the difference in the outcome of patients. RESPONSE: Patient characteristics were included in the development of multivariable model. Pag 9 line 16 and table 3: Multivariable logistic regression models were developed to assess factorsv associated with in-hospital death. In the first model (Table 4), neither admission on evenings or weekends to the ICU was associated with increased risk for in-hospital death.. What is the risk of in-hospital mortality of patients admitted in the WD compared with those admitted on the WE / holidays and / or night? RESPONSE: This information has been added to the third results section paragraph. Pag 9 line 22:.. As shown in Table 4, discharge from ICU during nights was independently associated with subsequent in-hospital mortality This table seems to show some relationship between specific diseases (i.e. cardiovascular, respiratory,...) and increased risk of mortality. This is consistent with the findings on Table 2, in particular patients discharged at night or in WE show major clinical severity than those discharged on WD. The increased risk of mortality appears related to medical categories. Could it be possible investigate in the database difference in mortality rate considering the different wards were the patients are delivered to? RESPONSE: Unfortunately limitations of our database outside of the ICU environment do not allow such a detailed analysis post discharge. Table 3 indicates that the pre-icu hospital stay was correlated with an increased risk of in-hospital mortality. What is the number of patients are you referring to? What is the average time of hospitalization in ICU-hospital before being admitted? It not clear what kind of wards are they coming from? May be internal? This justifies the increased risk of death related to kidney/ toxic / metabolic diseases.

9 RESPONSE: We have added the information in the first paragraph of the results section as to the location of the patients prior to admission in the ICU. The precision for this variable was measured in days. This information has also been added to the results as requested. Discussion The discussion and conclusions are not well balanced and adequately supported by the data, and the limits are not clearly discussed in the work. Page 10 first paragraph: In this study we found that while most admissions to ICU occur in the afterhours and that weekend admissions were associated with a higher crude case-fatality rate, the timing of admission was not associated with mortality once adjustment for confounding variables was performed (Table 4).. This sentence considers the admission of patients in ICU, but never take into account any kind of time past in other hospital services or emergency department. RESPONSE: We have added a new limitations paragraph to detail these issues. Page 11 line 17 Few clinicians would agree that discharges late at night would be considered to represent optimal care. In most of our participating ICUs weekend day discharges are discouraged and all night discharges are not standard practice. Many weekend and virtually all night discharges are considered premature discharges. These are nearly always due to limited bed capacity in the ICU and need to admit a more acutely or severely ill patient Standard practice in must hospitals is often different from the optimal way reported in literature, It would be interesting clear if in your ICU system has been carried out a group of protocols guiding the transfer processes. More over if do they exist, it would be useful clear if this is related whit reduction in mortality rate. RESPONSE: Such protocols are not presently in place but rather what we report is our general practice patterns and an informal survey of our ICU practices. Page 11 last paragraph: It is also notable that we observed a significant increase in risk for subsequent mortality following discharge on Fridays and an increate risk that was not statistically significantly associated with weekend day discharges. We speculate that this increased risk for death could be reflective of decreased intensity of care on wards on weekends.. Speculate on mortality rate on patients discharged on Friday (CI 95% p = 0.046) without any dip analysis between patients severity and ward discharged to is not useful. RESPONSE: This odds ratio is adjusted for a number of characteristics including severity of disease at discharge. We detail the limits of the post-discharge data in the limitations paragraph. Conclusions Pag12 riga 8..we found no association with either the time of admission during the day or

10 day of the week and subsequent outcome associated with ICU admission. Has noted before it is important to take into the right account not only the moment of the admission in ICU but also the one in any other hospital service. RESPONSE: We have reported different time intervals in the post hoc analysis now presented in the discussion. We have detailed our data limitations pre- and post-icu admission. References It is necessary select more recent studies. RESPONSE: We have attempted to present the body of literature related to ICU admission and discharge. Critically ill patients are a distinct population from the hospital population at large and we therefore have not included the hundreds of non-icu references. A recent reference related to ICU patients has now been added as suggested. - Minor Essential Revisions The aim of this study cited in the abstract ( The objective of this study was to assess effect of timing of admission to and discharge from ICUs and subsequent risk for death ) is not consistent with the title Hospital mortality is associated with ICU discharge time: A retrospective cohort study., and with the sentence in the background section (line 17 page 5) Page 7 line 5 and table 1 RESPONSE: The title has been revised as suggested. More precision needed in hours pattern: i.e. 0 (=24) = , ecc..23: RESPONSE: In the methods section and tables we refer to time points to the minute precision. Figures have used less precision to keep them as simple and readable as possible. We would be pleased to add a footnote regarding the precision in the figures at further editorial and reviewer request. Page 9 line 10: During the course of the ICU stay, 538 patients had a new do not resuscitate (DNR) order established. Statistical analysis about the difference in DNR in WE and WD is missing. RESPONSE: We reported DNR in the text but used the term decision to forego life sustaining therapy (DFLST) in the tables. We have revised this accordingly. The requested statistics are in the tables. Page 10 line 6: Since the initial reports identifying higher neonatal mortality rates associated with weekend deliveries more than 30 years ago there has been hundreds of subsequent publications evaluating potential after hours effects in wide ranges of patients and settings [14, 15]. Looking at patients older than 18 at would be better consider only literature related with adult patients.

11 REFERENCE: This paper was included to give due credit to the first investigators who initiated this line of research. We have now included all studies evaluating admission timing to adult ICUs conducted to our knowledge. I suggest to consider some recent articles such as: Kevat DA, Davies AR, Cameron PA, Rajaratnam SM. Increased mortality Associated with after-hours and weekend admission to the intensive care unit: a retrospective analysis. Med J Aust Jun 6;194(11):616. PubMed PMID: RESPONSE: This important study was published shortly prior to manuscript submission and was missed. It has now been added. Ricciardi R, Roberts PL, Read TE, Baxter NN, Marcello PW, Schoetz DJ. Mortality rate after nonelective hospital admission. Arch Surg May;146(5): PubMed PMID: RESPONSE: This study is not focussed on ICU patients. Gallerani M, Imberti D, Ageno W, Dentali F, Manfredini R. Higher mortality rate in patients hospitalised for acute pulmonary embolism during weekends. Thromb Haemost Jul 4;106(1):83-9. Epub 2011 May 5. PubMed PMID: RESPONSE: This study is not focussed on ICU patients. Kuijsten HA, Brinkman S, Meynaar IA, Spronk PE, van der Spoel JI, Bosman RJ, de Keizer NF, Abu-Hanna A, de Lange DW. Hospital mortality is associated with ICU admission time. Intensive Care Med Oct;36(10): Epub 2010 Jun 15. PubMed PMID: ; PubMed Central PMCID: PMC RESPONSE: This reference (reference 3) was included in our original manuscript and highlighted as a key reference in the first paragraph of the introduction.