Allocating Healthcare Budgets to General Practices
Peter C. Smith on behalf of PBRA team
Imperial College Business School & Centre for Health Policy
http://www.nuffieldtrust.org.uk/projects/index.aspx?id=338
The Person-based resource allocation (PBRA) project
• Led by Jennifer Dixon (Nuffield Trust) from 2007
• Initial purpose was to develop budgets for practice based commissioning based on individual patient data
• Coverage: secondary care, prescribing, community health services
3
Year Name Allocations to
Approximate population
size
Years applied
1976 RAWP 14 RHAs 3m 77/78 – 90/91
1980 RoR 14 RHAs 3m 91/92 – 94/95
1993 University of York
14 RHAs192 DHAs
3m250,000
95/96 – 01/02
2001 AREA 303 PCTs 175,000 02/03 – 06/07
2006 CARAN 152 PCTs 350,000 07/08 –
Reviews of resource allocation in English NHS Hospital and Community Health Services , 1976- today
Drawn from Bevan, and Bevan and Van der Ven Note: RAWP = Resource Allocation Working Party
RoR = Review of RAWPAREA = Allocation of Resources to English AreasCARAN = Combining Age Related Additional Needs (9)
PBRA modelling principles
• Use of individual-level data on both users and non-users of health care services (entire English population)
• Use of data from past NHS encounters to measure morbidity directly (via ICD chapters)
• Predict future expenditure at an individual level.• Developed on samples of 5 million patients
registered within GP practices – models validated on separate sample of 5 million patients.
• Models further assessed by performance of predictions at practice level
5
Linking the data sets for analysis
6
2005/06 2006/07 2007/08
Explanatory variables
Modelling principles
Prediction variable
Samples drawn from patients registered 1 April 2007
Modelling
• Hospital-based expenditure excluding maternity and mental illness
• Modelled hospital expenditure in year t as a function of:– Age and sex (36)– Diagnostic categories from hospital utilization in years t-1
and t-2 (152)– Attributed GP and small area needs characteristics (135)– Attributed small area supply characteristics (63)– PCT (152)
• Note: did not consider variables with potentially adverse incentive effects, eg number of encounters
Summary results of a set of five models, predicting costs for 2007/08 using data from 2005/06 & 2006/07
MODEL R2 individual R2 practice
Model 1: age and gender 0.0366 0.3444
Model 2 - ADD:
152 morbidity markers 0.1223 0.6084
Model 3 - ADD:
152 PCT dummies 0.1227 0.7437
Model 4 - ADD:
135 attributed needs & 63 supply 0.1230 0.7851
Model 5 - REDUCE TO:
7 attributed needs & 3 supply 0.1229 0.7735
Type of variable
Variable name
Individual • Age and gender
• 157 ICD-10 groups
Attributed needs
• Persons in social rented housing
• All disability allowance claimants
• Persons aged 16-74 with no qualifications (age standardised)
• Mature city professionals
• Proportion of students in the population
• Whether the person had a privately funded inpatient episode of care provided by the NHS in previous two years
• Asthma prevalence rate
Attributed supply
• Quality of stroke care (primary and secondary care), by weighted population
• Accessibility to MRI scanner
• Catchment population of the hospital trust that supplied the practice with the largest number of inpatient admissions
Using the formula to allocate to practices
• ‘Freeze’ supply variables at national levels• For each individual, calculate predicted NHS
hospital costs• For each practice calculate average costs in each
age/sex category• Assign age/sex specific averages to all individuals
in practice– To address data lags and changes in registration
• Share out PCT budget according to practices’ total predicted expenditure
0.5
11.
52
DFT
inde
x: re
lativ
e to
Eng
land
mea
n
0 10000 20000 30000 40000practice size: number of patients
Excludes the 16 practices with a DFT index > 2.
for the new model and practices with more than 500 patientsDistance from target and practice size
Distance from target
12
Percentage of practices more than x% away from target
> +/- 5% > +/- 10% > +/- 20%DFT relative to PCT mean 61.1 34.6 14.0
DFT relative to national mean 72.5 48.9 20.9
Phase III Objectives: in progress
• Refresh existing PBRA model using more recent data (for allocations 2011/12)
• Develop improved PBRA model (for allocations 2012/13)
• Model a variety of risk sharing arrangements (to inform shadow GP Consortia and NHS Commissioning Board)
• Develop a final PBRA formula (for allocations 2013/14)
Explanatory variables Prediction variable
2007/08 2009/102008/09
Basic model
2007/08 2008/09 2009/10 2010/11 2012/132011/12
Data lag
GP budgets and risk:we’ve been here before
• GP fundholding c.1991• Total fundholding c.1995• ‘Primary Care Groups’ c.1998• Practice based commissioning c.2002
Martin, S., Rice, N. and Smith, P. (1998), “Risk and the general practitioner budget holder”, Social Science and Medicine, 47(10), 1547-1554.
Smith, P. (1999), “Setting budgets for general practice in the New NHS”, British Medical Journal, 318, 776-779.
Fundholding
• Relatively generous budgets
• Limited set of elective conditions plus prescribing covered
• Per patient limit £6000
• Overspends largely borne by Health Authority
• Underspends kept by practice for patient services
• A very ‘soft’ budget
Decomposing the variation in practice expenditure
• The formula captures average clinical responses to measured patient and area characteristics. Therefore any variation from the formula will be due to:– Variations in clinical practice;– Variations in the prices of treatments used by the
practice;– Imperfections in the formula caused by known patient
characteristics that are not captured in the formula;– Random (chance) variations in levels of sickness
within the practice population.
High cost cases
Percentage of cases over £20K per person per year
Num
ber o
f pra
ctic
es
19
Sampled from patients (10m) within a 20% random sample of all patients100 replications for each consortium size
Consortium size increased in units of 10,000
-40
-20
020
40C
onso
rtium
risk
per
cap
ita(£
)
0 100000 200000 300000 400000 500000Consortium list size
Average risk Lower CIUpper CI
Simulations from all dataRisk smoothed over time - predicted versus actual expenditure
Consortia risk profile
14
-13.5
Upper 95% C.I.
Lower 95% C.I.
Average risk
0.2
.4.6
0.2
.4.6
0.2
.4.6
0.2
.4.6
0 2 4 6 8 10 0 2 4 6 8 10
0 2 4 6 8 10 0 2 4 6 8 10
Consortium size10000
Consortium size100000
Consortium size300000
Consortium size500000
Pro
babi
lity
Percentage VariationSimulations from all dataProbability of more than an X percent variation from annual budget
Consortia risk profile
Acknowledgement: Nigel Rice and Hugh Gravelle
0.2
.4.6
0.2
.4.6
0.2
.4.6
0.2
.4.6
0 2 4 6 8 10 0 2 4 6 8 10
0 2 4 6 8 10 0 2 4 6 8 10
Consortium size10000
Consortium size50000
Consortium size100000
Consortium size150000
Omit £100k Omit £150k
Pro
babi
lity
Percentage Variation
Probability of more than an X percent variation from annual budgetSimulations omitting high cost patients from practice lists
Consortia risk profile
Acknowledgement: Nigel Rice and Hugh Gravelle
Some possible consequences of ‘hard’ budget constraints
• Practices that perceive that their expenditure will fall below their budget may “spend up” in order to protect their budgetary position in future years;
• Practices that perceive that their expenditure will exceed their budget may be thrown into crisis as they seek to conform to the budget;
• Patients may be treated inequitably. Different practices will be under different budgetary pressures, and so may adopt different treatment practices.
• Within a practice, choice of treatment may vary over the course of a year if the practice’s perception of its budgetary position changes.
• General practices may adopt a variety of defensive stratagems, such as cream skimming patients they perceive to be healthier than implied by their capitation payment.
Some budgetary risk management strategies
• Pooling practices• Pooling years• Excluding predictably expensive patients• ‘Carving out’ certain procedures or services• Analysis of reasons for variations from budgets• Allowing some reinsurance of risk
– Limiting liability on individual episode– Limiting liability on individual patient– Risk sharing– Retention of a contingency fund– Etc
• Making sanctions and rewards proportionate