Factors associated with variation in hospital use at the End of Life in England

Transcription

1 Factors associated with variation in hospital use at the End of Life in England Martin Bardsley,Theo Georghiou, John Billings Nuffield Trust

2 Aims Explore recent work undertaken by the Nuffield Trust 1. A study documenting relative costs of different health and social care services in the final 90 days of life. 2. A study of variations in hospital use in the last 12 months of life examining the relative importance of person level verses system level factors. 3. An examination of differences in hospital use at the end of life associated with ethnic group. Our work looks at service use and estimated cost specifically at end of life important for planning, managing resources

3 The Nuffield Trust Promote independent analysis and informed debate on healthcare policy across the UK Charitable organization founded in 1940 Formerly a grant-giving organization Since 2008 we have been conducting in-house research and policy analysis Significant interest in uses of predictive risk techniques William Morris 1st Viscount Nuffield ( )

4 Nuffield Trust Research team data linkage projects nuffield trust Combined predictive model nuffield trust Predicting social care costs Predictive risk modelling Resource allocation nuffield trust Risk sharing for CCGs nuffield trust Person based resource allocation nuffield trust Social care at end of life nuffield trust Cancer and social care Descriptive studies Evaluations nuffield trust Integrated care pilots nuffield trust Virtual Wards nuffield trust WSD nuffield trust Marie Curie Nursing Service

5 Exploiting person level data Linking data a. over time to look at what happens to people not just events b. across care providers to build broader picture Person level Capture services provided ->costs; quality Descriptions of health -> outcomes

6 Information flows

7 Pseudonymisation using encrypted NHS numbers

8 Health and social care event timeline for one person entering intensive social care.

9 Estimated costs per decedent, Hospital and social care: Change in costs per decedent during final year 2,500 2,000 1,500 1,000 Hospital costs Social care costs Month prior to death

10 Estimating costs across sectors of care at end of life

11 Emergency admissions for cases where nursing started 8-14 days before death

12 Impact of MCNS care on hospital costs Table 1 Post index date hospital costs for Marie Curie cases and matched controls Mean (sd) hospital costs per person Activity Type Marie Curie cases Matched controls Difference Emergency admissions 463 ( 1,758) 1,293 ( 2,531) 830 Elective admissions 106 ( 961) 350 ( 1,736) 244 Outpatient attendances 33 ( 212) 76 ( 340) 43 A&E attendances 9 ( 34) 31 ( 60) 22 All hospital activity 610 ( 2,172) 1,750 ( 3,377) 1,140 Significantly greater reduction in costs among those with no recent history of cancer Also cost reduction much greater for those who started receiving MCNS care earlier ( 2,200 for those >2 weeks before death)

13 Final 90 day costs multiple care services Reflection on data gathered for multiple uses over several years Summary of what we know from our own work Opportunistic - data often limited, not necessarily intended to be exhaustive Secondary aspect modelling impacts of changes in care patterns at the end of life

14 Approach to costing Costs are related to commissioner not necessarily the provider or related to local efficiency Uses unit costs (PSSRU mainly) related to a measure of activity eg GP visit, Community nurse contact, social care care days Hospital activity based on tariffs with some especially base estimates for non tariff costs Hospice costs based on advice MCCC and NCPC

15 Final 90 days Hospital Care 2009/ /12 England wide HES - ONS data Average cost per day, per person N = 1.3M All Care 4,580 Unplanned adms 3,465 Planned adms 822 Outpatients 211 AE 82

16 Final 90 days Social Care LA funded care only 7 areas Social care data linked to SUS, and GP register Average cost per month, per person N = 73,243 All Care 1,010 Res/Nurs home Home care Other (~two thirds none)

17 Final 90 days District nursing Single London borough Community nursing dataset linked to GP register Average minutes per day, per person N = 10,779 All Care 278 (Average minutes in final 90 days = 214; Two thirds - none)

18 Final 90 days GP consultations data Three London PCTs GP dataset Estimated consultations Estimated GP consultations per day N = 21,522 All Care 147 (Average number estimated consultations = 4.6; One third none)

19 Hospice stays No direct data ONS data to identify rates of people dying in hospices (5.5%) Estimated average LOS of stays in hospice (14 days) Used per diem costs averaging PSSRU, Help the Hospices and Marie Curie estimates to 400 per day Included stays that did not end in death (assume within 90 days) Ended up approx 10,000 per person who died in hospice. Equivalent to about 550 per person who died

20 Summary cost estimates

21 Guessing the breakdown of expenditure across the last 90 days of life 2% 8% 4% Emergency inpatient admissions Non-emergency inpatient admissions Outpatient attendances 15% AE visits 1% 3% 53% LA funded social care District nursing care 13% GP visits CAVEAT not the same populations

22 Observations Scale of costs outside of institutional (hospital/hospice ) care seemed low and not enough to offset potential savings from home based eol care Our approach was pragmatic scratching together what data we could and making some assumptions With more time we could have done this better and we think there needs to be a better understanding of costs for this critical phase of care

23 What factors are driving variation in hospital care in the last 12 months of life?

24 Variations in the percentage of people dying in hospital CCGs England

25 Aims To assess the variability in hospital use at the end of life and assess the influence of three sets of factors: a. The age/gender and socio economic status of patients b. The health problems presented by the patient, and the cause of death c. Features of the local health systems including common practices within individual hospital or primary care providers, and availability of alternatives to hospital care

26 Methods Based on all people who died over a 3 year period Link hospital records at person level to identify activity in last 12 months of life Generate descriptive variables Create models to explain Admissions in last 12 months of life Bed days in last 12 months of life Whether a person died in hospital

27 HES as a data source Positives Inpatient records from 2001 A&E attends and Outpatient from Comprehensive for NHS care across the country reasonably standardised Negatives Limited clinical information Varying quality of recording esp diagnoses Focuses on acute hospital care and misses community Common person level identifier Linked to ONS death data

28 Cohort studied No. excluded % excluded (of all deaths) No. remaining in cohort All death records 2009/10 to 2012/ ,385,103 No hospital record in final year 132, % 1,252,515 No gender information 8, % 1,243,808 No residential (LSOA) information % 1,243,034 Resided outside England 8, % 1,242,687 Subsequent (post death date) hospital activity 11, % 1,223,859 Final study cohort - - 1,223,859

29 Impact of age on hospital use

30 Changes cost per day

31 Changes in activity during last year of life (log scale)

32 Independent Variables Patient characteristics Age bands 5 Sex 2 Ethnicity 10 IMD (excluding health) C Population Density C Clinical/health characteristics Diagnostic Group 145 Cause of Death 117 Number of Long Term Conditions 10 Charlson Index 22 Local health system List Size GP C Number of GP FTE C Access to care home/hospice beds C Predomionant hospital 364

33 Model performance Model dependent variable: Admissions Bed Days Died in hospital Mean % Standard Deviation % 10 percentile % Median % 90 percentile % Unadjusted coefficient of variation Model performance R Squared Adjusted coefficient of Variance

34 Unadjusted differences by hospital type

35 Adjusted differences by hospital type

36 Ethnicity and hospital use at end of life

37 Variables that explain admission to hospital in last 12 months Admissions Variable group % 'explained' Age bands 11.1% Sex 0.4% Ethnicity 4.5% IMD (excluding health) 0.1% Population Density 0.0% Clinical/health characteristics Diagnostic Group 65.0% Cause of Death 9.4% Number of Long Term Conditions 0.4% Charlson Index 0.7% Local health system List Size GP 0.0% Number of GP FTE 0.0% Access to care home/hospice beds 0.0% Predominant Hospital 8.4%

38 Diagnostic codes associated with greatest bed day use Bed days Description Estimate Standard Error Pr > t F04-F09 Other organic including symptomatic mental disorders <.0001 L00-L14 L55-L99 Other infections and disorders of the skin <.0001 L40-L45 Papulosquamous disorders (including Psoriasis) <.0001 C81-C96 Malignant neoplasms of lymphoid, haematopoietic & rel. tiss <.0001 G20-G26 Extrapyramidal & movement disorders (incl. Parkinsonism) <.0001 L50-L54 Urticaria and erythems <.0001 N99 Other disorders of the genitourinary system <.0001 E10-E14 Diabetes Mellitus <.0001 G10-G13, G30-G32 Other degenerative diseases (incl. Alzheimer) <.0001 A15-A19 Tuberculosis <.0001 T80-T88 Complications of surgical & medical care nec <.0001 M95-M99 Other disorders of the musculoskeletal system & conn. tiss <.0001 F00-F03 Dementia <.0001 G35-G37 Demyelinating diseases (incl Multiple Sclerosis) of the CNS <.0001 A00-A09 Intestinal infectious diseases <.0001 C40-C41 Malignant neoplasm of bone and articular cartilage <.0001 F30-F39 Mood [affective] disorders <.0001 R69 Unknown & unspecified causes of morbidity <.0001 F80-F99 Other mental and behavioural disorders <.0001 F20-F29 Schizophrenia, schizotypal and delusional disorders <.0001 F70-F79 Mental retardation <.0001

39 Relationship between ACS admissions and IMD Figure 7.2 Age/Sex Adjusted ACS Admissions/1,000 by Local Authority and Area IMD

40 Admissions in last 12montsh of life by IMD Figure 7.3 Age/Sex Adjusted Admissions Last Year of Life by Local Authority and Area IMD

41 Overview Admissions Bed Days Died in Hospital Variable group % 'explained' % 'explained' % 'explained' Ethnicity 4.5% 3.4% 0.2% IMD (excluding health) 0.1% 0.2% 0.1% Clinical/health characteristics Diagnostic Group 65.0% 52.6% 57.0% Cause of Death 9.4% 13.3% 35.7% Number of Long Term Conditions 0.4% 0.1% 0.1% Charlson Index 0.7% 0.7% 2.2% Local health system Access to care home/hospice beds 0.0% 0.1% 0.0% Predominant Hospital 8.4% 28.8% 3.9%

42 Selected summary points Able to apply much more sophisticated adjustments for the range of factors influencing care use Provider seems to have a relatively small impact on admission and dying in hospital but much bigger impact on length of stay=> if you want to improve efficiency look at stay lengths and across more than just final episode of care Significant residual differences by ethnic group Variability in hospital use at end of life associated with deprivation much less than for ACS admissions

43 Sign-up for our newsletter Follow us on Twitter ( Thank You