Introduction
In a recent White Paper the new British Conservative government emphasised the importance of clinical outcomes. It notes that, in future, success will be measured, not through the achievement of process targets, such as short waiting times, but against outcomes such as cancer and stroke survival rates.12 Although the NHS budget is ring-fenced against the ongoing public sector deficit reduction programme, its budget is still likely to be under considerable pressure, and attention is likely to focus on the extent to which any additional health-care expenditure yields genuine patient benefits in the form of improved health outcomes.
However, one of the most fundamental yet unresolved issues in health policy is the extent to which additional health-care expenditure yields patient benefits, in the form of improved health outcomes. The work of health technology agencies, such as NICE, has greatly improved our understanding at the micro-level of the costs and benefits of individual therapeutic technologies. However, there remains a dearth of evidence at the macro-level on the benefits of increased health system expenditure.
Recently a series of studies has taken advantage of the availability of two new data sets to examine the relationship between NHS expenditure and mortality rates for various disease categories.59,60,62,63One data set contains mortality rates for various disease categories at the level of geographically defined local health authorities, known as PCTs. The other data set presents NHS expenditure by PCT on 23 broad programmes of care. This data set embraces most items of publicly-funded expenditure, including inpatient, outpatient and community care, and pharmaceutical prescriptions.
Like previous studies, we employ a model that assumes that each PCT receives an annual financial lump sum budget from the national ministry and allocates its resources across the 23 programmes of care to maximize the health benefits associated with that expenditure. Estimation of this model using the expenditure and mortality data facilitates two related studies: first, a study of how changes in the NHS budget impact on expenditure in each care programme; and second, a study of the link between expenditure in a programme and the health outcomes achieved, notably in the form of disease-specific mortality rates. The latter study also permits the calculation of the cost of an additional life-year for individual programmes of expenditure.
DOI: 10.3310/hta19140 HEALTH TECHNOLOGY ASSESSMENT 2015 VOL. 19 NO. 14
© Queen’s Printer and Controller of HMSO 2015. This work was produced by Claxtonet al.under the terms of a commissioning contract issued by the Secretary of State for Health. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.
The work presented here draws heavily on previous studies. These were constrained in a number of ways and, in this analysis, we build on and improve these previous studies in four major ways:
¢ First, due to data limitations previous studies related expenditure in time periodtto mortality in periods t,t–1, andt–2 combined. In doing this, such studies assumed that PCTs had reached some sort of equilibrium in the expenditure choices they make and the outcomes they secure. This is probably not an unreasonable assumption given the relatively slow pace at which both types of variable change but, with more recent mortality data now available, here we relate expenditure in time periodtto mortality in periodst,t+ 1, andt+ 2 combined (seeModel estimation using 2006/7 expenditure data and mortality data for 2006/7/8: CARAN need and two market forces factors).
¢ Second, previous studies have tended to focus on a very limited number of care programmes
(e.g. cancer, circulatory disease, gastrointestinal problems and respiratory problems). Here we present plausible outcome models for a larger number of budgeting categories.
¢ Third, previous estimates of the cost of a life-year have been for individual programmes of care. Here we present estimates of the cost of a life-year for an enlarged number of programmes and, importantly, with the aid of assumptions about the productivity of programmes without a meaningful mortality-based outcome indicator, we extend our individual programme estimates to incorporate expenditure across all programmes of care.
¢ Finally, although previous results and our current models‘pass’the appropriate statistical tests, we subject our latest results to a substantial sensitivity analysis.
The structure of this report is as follows.Previous studiespresents a brief review of previous empirical studies in this domain, which have often yielded conflicting results. A straightforward theoretical model of the budgetary problem faced by a PCT manager seeking to allocate limited funds between competing programmes of care is presented inTheoretical model. The PB and health outcome (mortality) data are described inNHS programme budgeting in EnglandandHealth outcome and other datarespectively. Estimation issues and strategyoutlines our estimation methods and some of the issues surrounding them. InAnalysis of programme budgeting expenditure for 2005/6 and mortality data for 2002/3/4we commence our empirical work by estimating well-specified econometric models that outline (a) the budgetary expenditure choices and (b) the health outcomes achieved by PCTs using expenditure data for 2005/6 and mortality data for 2002/3/4.Analysis of programme budgeting expenditure for 2006/7presents results using expenditure data for 2006/7 and mortality data for 2004/5/6. It also presents results using the same expenditure data but updating the mortality data to 2006/7/8. Several pieces of sensitivity analysis are also included inAnalysis of programme budgeting expenditure for 2006/7, but the major piece of sensitivity analysis–examining the impact of relaxing the instrument validity restriction–is reported inThe sensitivity of the outcome elasticity to the validity of the instrument exclusion restrictions.
InAnalysis of programme budgeting expenditure for 2007/8 and mortality data for 2007/8/9we re-estimate our model using updated expenditure and mortality data. In particular, we use the PB expenditure for 2007/8 and mortality data for 2007/8/9 to re-estimate our outcome and expenditure equations. InAnalysis of programme budgeting expenditure for 2008/9 and mortality data for 2008/9/10 we update the data set again, and this time we employ PB expenditure data for 2008/9 and mortality data for 2008/9/10. We also compare the elasticities and cost of a life-year estimates that we have obtained using expenditure and mortality data for different years.
Finally,Summary and concluding remarkspresents a summary of our findings and some concluding remarks. Previous studies
There is a large body of literature on the determinants of international variations in health-care spending in which income levels often play a central role.140However, whether or not more expenditure generates better outcomes–for example, in terms of reduced mortality–remains a matter of debate. For example, Fisher and Welch141note various ways in which more health care might harm patients and they cite various
studies supporting their arguments. In a comprehensive review, Nolte and McKee67discuss many studies that examine the impact of health care and other explanatory variables on some measure of health-care outcome. Nolte and McKee67point out that researchers usually combine a production function approach with the application of regression analysis. For example, in an early cross-sectional study of 18 developed countries, Cochraneet al.64use regression analysis to examine the statistical relationship between mortality rates on the one hand and per capita GNP and per capita consumption of inputs such as health-care provision on the other. They find that the indicators of health-care provision were generally not associated with the outcomes in the form of mortality rates. Thereafter, the failure to identify strong and consistent relationships between health-care expenditure and health outcomes (after controlling for other factors) has become a consistent theme in the literature, while, in contrast, socioeconomic factors are often found to be good determinants of health outcomes.65–67
This failure to detect a significant positive relationship between expenditure and health outcome might reflect the difficulties associated with any such study rather than the absence of such a relationship. For example, Gravelle and Backhouse68examine some of the methodological difficulties associated with empirical investigation of the determinants of mortality rates. These include simultaneous equation bias and the associated endogeneity problem (that the level of health-care input might reflect the level of health outcome achieved in the past), and that a lag may occur between expenditure and outcomes (studies typically assume that expenditure has an immediate effect on mortality). To avoid the difficulties imposed by data heterogeneity inherent in international analyses, the study by Cremieuxet al.69examines the relationship between expenditure and outcomes across 10 Canadian provinces over the 15-year period 1978–92. They find that lower health-care spending is associated with a significant increase in infant mortality and a decrease in LE.
Although challenging the received empirical wisdom, one difficulty with the Cremieuxet al.69study is that the estimated regression equation consists of a mixture of potentially endogenous variables (such as the number of physicians, health spending, alcohol and tobacco consumption, and expenditure on meat and fat) and exogenous variables (such as income and population density). The authors’chosen estimation technique (GLS) does not allow for this endogeneity and consequently the coefficients on the endogenous variables may be biased.68Or
’s142study of the determinants of variations in mortality rates across 21 Organisation for Economic Co-operation and Development countries between 1970 and 1995 may suffer from the same weakness. She finds that the contribution of the number of doctors to reducing mortality in Organisation for Economic Co-operation and Development countries is substantial but her estimation technique assumes that the number of doctors is exogenous to the health system.
Nixon and Ulmann70provide a detailed review of 16 studies that have examined the relationship between health-care inputs and health outcomes, using macro-level data. They also undertake their own study using data for 15 EU countries over the period 1980–95. They employ three health outcome measures–LE at birth for males and females, and the infant mortality rate–and a dozen or more explanatory variables including per capita health expenditure; number of physicians (per 10,000 head of population); number of hospital beds (per 1000 head of population); the average length of stay in hospital; the inpatient admission rate; alcohol and tobacco consumption; nutritional characteristics; and environmental pollution indicators. Nixon and Ulmann70conclude that although health expenditure and the number of physicians have made a significant contribution to improvements in infant mortality,‘. . . health care expenditure has made a relatively marginal contribution to the improvements in LE in the EU countries over the period of the analysis’. Again, however, the study does not allow for the possibility that some of the explanatory variables may be endogenous.
Although loosely based on the notion of a health production function, the traditional empirical study described above has rarely been informed by an explicit theoretical model. This is understandable, as the processes giving rise to the observed health outcomes are likely to be very complex, and any theoretical model might become rather unwieldy. However, this absence of a model has usually led to a theoretical search for measures of health inputs demonstrating a statistically‘significant’association with health
DOI: 10.3310/hta19140 HEALTH TECHNOLOGY ASSESSMENT 2015 VOL. 19 NO. 14
© Queen’s Printer and Controller of HMSO 2015. This work was produced by Claxtonet al.under the terms of a commissioning contract issued by the Secretary of State for Health. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.
outcomes. In contrast, in this study we inform our empirical modelling with a theoretical framework. We believe that this may lead to a more convincing and better specified model of health outcomes than that used in many previous studies, and this model is outlined in the next section.
Theoretical model
Our modelling framework assumes that each PCTireceives an annual financial lump sum budgetyifrom the national ministry, and that annual total expenditure cannot exceed this amount. The PCT must then decide how to allocate its budget across theJprogrammes of care (J=23 in this case). For each programme of care there is a‘health production function’fi(.) that indicates the link between local spendingxijon programmejand health outcomes in that programmehij. Health outcomes might be measured in a variety of ways, but the most obvious is to consider some measure of improvement in LE, possibly adjusted for QoL, in the form of a QALY.
The nature of the specific health production function confronted by a PCT will depend on two types of local factors: the clinical needs of the local population relevant to the programme of care (which we denotenij); and broader local environmental factorszijrelevant to delivering the programme of care (such as input prices, geographical factors, or other uncontrollable influences on outcomes). Both clinical and environmental factors may be multidimensional in nature. Increased expenditure then yields improvements in health outcomes, as expressed, for example, in improved local mortality rates, but at a diminishing rate. That is:
hi j =fj(xi j,ni j,zi j);δfj=δx>0;δ2fj=δx2<0: (9)
We assume there is a PCT social welfare functionW(.) that embodies health outcomes across theJ
programmes of care. Assuming no interaction between programmes of care, each PCT allocates its budget so as to maximise total welfare, subject to the local budget constraint and the health production function for each programme of care:
max W(hi1,hi2,…,hi J) subject to ∑ j xi j≤yi hi j=fj(xi j,ni j,zi j); j=1,…J: (10)
It can of course quite plausibly be argued that decision-makers do not discriminate between health outcomes in different programmes of care, and thatW(.) is merely the sum of such outcomes. However, there is no need for that assumption in our formulation.
Each PCT allocates expenditure across the 23 programmes of care so that the marginal benefit of the last pound spent in each programme of care is the same. This is represented diagrammatically inFigure 12, which illustrates the trade-off between just two programmes of care. The top left-hand quadrant indicates the health production function for programme 1, whereas the bottom right-hand quadrant indicates the health production function for programme 2, albeit in transposed form. The bottom left-hand quadrant indicates the budget constraint; the expenditure choice must lie on the budget line. This means that for each feasible pair of expenditure choices (points on the budget constraint line), a pair of health outcomes in the two programmes emerges, which is traced out as the health production possibility frontier in the top-right quadrant. The PCT will choose the point on this frontier that maximizes welfare. In this example, we have indicated a simple health maximizing approach (the maximum health summing across the two programmes), leading to optimal health outcomes (H1*, H2*) and expenditure (X1*, X2*).
Solving the constrained maximisation problem yields the result that the optimal level of expenditure in each category,x*ij, is a function of the need for health care in each category (ni1,ni2,. . . , niJ),
environmental variables affecting the production of health outcomes in each category (zi1,zi2,. . ., ziJ), and PCT income (yi). Thus:
x
i j =gj(ni1,: : :ni J,zi1,: : :zi J,yi); j=1,: : :, J (11)
Thus, for each programme of care there exists an expenditure equation (seeEquation 11) explaining expenditure choice of PCTs and a health outcome equation (seeEquation 9) that models the associated health outcomes achieved.
Our model is static in the sense that the health production function (seeEquation 9) assumes that all health benefits occur contemporaneously with expenditure. We acknowledge that for some programmes of care benefits might occur≥1 year after expenditure has occurred. This is particularly likely to be the case for those programmes aimed at encouraging healthy lifestyles, where some benefits may occur decades after the actual programme expenditure. For other programmes, such as maternity/reproductive conditions and neonate conditions, benefits may be largely contemporaneous with expenditure.
Furthermore, we do not model the decision-maker’s time preferences.
For our empirical modelling, however, we are constrained by the data we have available, which are largely cross-sectional in nature. Owing to data limitations, previous studies have had to relate expenditure in periodtto mortality data in periodst,t–1, andt–2 combined so that the mortality data precedes the expenditure data. This is not ideal. Implicitly previous studies have had to assume that the data represent a quasi long-run equilibrium position, and that relative expenditure levels and health outcomes within each PCT have been reasonably stable over a period of time. As we shall see, this appears to be a reasonable assumption because we obtain similar results when we estimate our models using expenditure for periodt with either mortality data for periodst,t–1, andt–2 combined (seeModel estimation using 2006/7 expenditure data and mortality data for 2004/5/6: CARAN need and two market forces factors) or with mortality data for periodst,t+ 1, andt+ 2 combined (seeHealth outcome and other data).
Having outlined our model, in the next section we discuss the data sets used to estimate this model. NHS programme budgeting in England
The English NHS is the archetypal centrally-planned and publicly-funded health-care system. Its revenue derives almost entirely from national taxation, and access to the system is generally free to the patient.
H2
H1
X2
X1
The budget line Programme 1
Programme 2 (H1*,H2*)
(X1*,X2*)
FIGURE 12 Graph showing optimal trade-off between two programmes of care.
DOI: 10.3310/hta19140 HEALTH TECHNOLOGY ASSESSMENT 2015 VOL. 19 NO. 14
© Queen’s Printer and Controller of HMSO 2015. This work was produced by Claxtonet al.under the terms of a commissioning contract issued by the Secretary of State for Health. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.
Primary care is an important element of the system, and GPs act as gatekeepers to secondary care and