Cost-effectiveness analysis and policy choices: investing in health systems.

Cost-effectiveness analysis and policy choices: investing in health systems C.J.L. Murray,1 J. Kreuser,2 & W. Whang3 The role of health systems infrastructure in studies of cost-effectiveness analysis and health resource allocation is discussed, and previous health sector cost-effectiveness analyses are cited. Two sub- stantial difficulties concerning the nature of health system costs and the policy choices are presented. First, the issue of health system infrastructure can be addressed by use of computer models such as the Health Resource Allocation Model (HRAM) developed at Harvard, which integrates cost-effective- ness and burden of disease data. It was found that a model which allows for expansion in health infra- structure yields nearly 40% more total DALYs for a hypothetical sub-Saharan African country than a model which neglects infrastructure expansion. Widespread use of cost-effectiveness databases for resource allocations in the health sector will require that cost-effectiveness analyses shift from reporting costs to reporting production functions. Second, three distinct policy questions can be treated using these tools, each necessitating its own inputs and constraints: allocations when given a fixed budget and health infrastructure, or when given resources for marginal expansion, or when given a politically constrained situation of expanding resources. Confusion concerning which question is being addressed must be avoided through development of a consistent and rigorous approach to using cost-effective- ness data for informing resource allocations. Introduction Cost-effectiveness analysis of health sector interven- tions was first applied in the 1960s based on meth- ods developed to analyse military investments (1). Since 1970, the number of published studies using cost-effectiveness analysis has been steadily rising, reflecting a growing concern for the appropriate use of scarce health sector resources (2). Initially, most cost-effectiveness studies reported results using indi- cators such as the cost per case diagnosed and treat- ed of a particular disease or the cost per fully immu- nized child. These studies using outcome or benefit measures that are very disease or context specific have been gradually replaced by studies using more general measures of health outcome. With more widespread reporting of results in terms of costs per quality-adjusted life year (QALY) or other general health measure, comparisons of the cost-effectiveness of interventions targeting different health problems I Associate Professor of International Health Economics, Harvard Center for Population and Development Studies, 9 Bow Street, Cambridge, MA 02138, USA. Requests for reprints should be sent to this author. 2 Information Officer, Information Engineering Unit, Organization and Business Practices Department, The World Bank, Washington DC, USA. 3 Columbia University College of Physicians and Surgeons, New York NY, USA. Reprint No. 5513 have become possible. League tables of the cost- effectiveness of different interventions are a natural consequence (3-8). Two landmark policy analyses have provided an impetus to using cost-effectiveness to compare a wide range of health interventions. These exercises provide enough information so that cost-effective- ness analysis for the first time can be used to inform resource allocations across the entire health sector. First, the Oregon Health Services Commission (9-18) examined 714 condition-treatment pairs (called inter- ventions in the rest of the following discussion) and calculated the cost per QALY. The valuation of out- comes from medical intervention and the rankings from cost-effectiveness analysis were then subject to extensive public review through a series of town meetings. The rank list of interventions from this process can then be used for selecting the interven- tions that Medicaid will finance in the State, which plans to fund (in order of the rank list) each interven- tion maximally until the budget runs out. This sector- al application of cost-effectiveness is now being implemented (18). The second major policy review was the Health Sector Priorities Review undertaken by the World Bank from 1987 to 1993 (7). Twenty- six major health problems of developing countries were reviewed by teams of economists, public health specialists and epidemiologists. The cost-effectiveness of more than 50 specific health interventions were evaluated using a standard methodology for costs Bulletin of the World Health Organization, 1994, 72 (4): 663-674 © World Health Organization 1994 663 C.J.L. Murray et al. and benefits.a These databases provide useful infor- mation on cost-effectiveness which will help deter- mine resource allocations across the entire health sector. Building largely on the Health Sector Priorities Review, the World Bank has promoted in the World development report 1993: investing in health (WDR) the concept of using cost-effectiveness of health sec- tor interventions and the burden of disease of health problems to develop essential packages of clinical and preventive care (23). The WDR also proposes that cost-effectiveness analysis be used to determine the package of services covered by insurance schemes and to inform health research priorities. In this issue of the Bulletin, Bobadilla et al. (24) pro- vide details on the method and rationale for selec- tion and of interventions and their quantities in the proposed package. In brief, estimates of the cur- rent burden of disease are combined with a cost- effectiveness rank list of interventions, to derive packages of services that, for a given budget, will purchase the largest improvement in health as measured by DALYs (disability-adjusted life years). Given the considerable attention gamered by the WDR, it is important to examine carefully the impli- cations of this new and more extensive application of cost-effectiveness analysis. Limitations of sectoral cost-effectiveness In recent years, the theoretical basis for using cost- effectiveness analysis to guide health sector resource allocations has been discussed: the validity of DALY or QALY maximization as a goal for the health sec- tor (25-30), the nature of individual preferences for health states and how these preferences are incorpo- rated into QALYs (31-35), the importance of margi- nal costs that change as a function of output (36, 37), the effect of intervention-specific fixed costs (36, 38), and the sensitivity of conclusions to abstract concepts such as discounting (39-51). These techni- cal issues are important and likely to be vigorously debated for many years but probably do not have a profound effect on the sectoral application of cost- effectiveness to policy choice, although further research may indicate important modifications and refinements are needed in the methods. Two more general and potentially important criticisms are concemed with the focus of cost-effec- tiveness analysis. First, cost-effectiveness analysis of health interventions, which are more often than not a To make the results of the Health Sector Priorities Review consistent with the Global Burden of Disease, benefits were measured using disability-adjusted life years. See Murray et al. (19-22) for details of the method of calculating benefits. disease specific, tends to neglect the role of the health system in delivering these interventions. There are no explicit analyses of the cost-effective- ness of improving the physical or human infrastruc- ture of the health system, which provide for direct comparisons between investing in the delivery system and purchasing more specific interventions delivered by the health system. Some may be con- cemed that the intervention focus of cost-effective- ness analysis may shift the focus of policy debate from who delivers health services to satisfying spe- cific targets or goals for particular activities. In the extreme, some accuse cost-effectiveness analysis of fostering a vertical approach to disease control as opposed to the horizontal approach embodied in the primary health care movement. Second, there is a potential for considerable confusion, including in the WDR, on the policy choice that should be informed by cost-effectiveness analysis. For example, should cost-effectiveness analysis be used to suggest the reallocation of resources between programmes that will lead to the greatest improvement in health or should it only be used to suggest how marginal increases in health sector resources could best be allocated to improve health? In this paper, we present in brief a proposed method by which the cost-effectiveness of investing in the physical and human infrastructure of the health system can be evaluated. A resource alloca- tion model, the technical details of which are described elsewhere (36), is illustrated with an appli- cation to sub-Saharan Africa. The model is then used to address the second issue of the range of policy questions that can be addressed with cost-effective- ness analysis. Finally, some implications for future cost-effectiveness studies are highlighted. Cost-effectiveness of investing in the health system The accepted standard for reporting the results of cost-effectiveness studies in the literature and un- published reports is to provide information on the average cost per unit of health output (such as a DALY) at one level of production. Average cost equals the sum of general or infrastructure fixed costs, programme-specific fixed costs, and vari- able costs divided by total output. Arbitrary rules are promulgated to allocate the general infrastructure fixed costs, such as the costs of hospitals and health centres to specific interventions undertaken in those facilities. These arbitrary divisions of joint produc- tion costs are usually based on some proxy measure of activity, such as staff hours, bed-days, or square- feet occupied. The treatment of the costs of maintain- 664 WHO Bulletin OMS. Vol 72 1994 Cost-effectiveness analysis and health sector policy choices ing the physical and human infrastructure of the health system in this arbitrary manner leads to two major problems with the cost-effectiveness approach when applied to sectoral decisions. First, the average cost-list approach used in the WDR (23) ignores existing infrastructure and implic- itly assumes that hospitals and health centres can be built in infinitely divisible quantities.b The average cost of an intervention includes a component due to the general fixed costs divided by the volume of out- put at the time of the assessment. Allocating resourc- es according to average unit costs implies that frac- tions of facilities, e.g., 2% of a health centre, can be built as required. Concomitantly, existing facilities can be used in shares less than one while the costs for the rest of the facility are not incurred. A resource allocation based on an average cost list may include only the costs of running 45% of district hos- pitals, ignoring the fact that hospitals and health cen- tres come in indivisible units. For example, the pack- age of essential clinical services proposed by the World Bank for low-income countries does not include all the costs of maintaining and operating the existing referral and district hospitals. Even the frac- tional costs of facilities depend on operating each new fraction at the same level of output as was included in the analysis. Otherwise the general fixed costs divided by output, which figures in average cost, would be different. Second, even if shares of facilities could be built or closed at will, the joint costing rules artifactually penalize interventions that are more technically effi- cient. Fig. 1 shows a production function for a health centre that undertakes only two activities: the expand- ed programme of immunization (EPI) and prenatal care. The area within the curve shows all possible combinations of the two activities, given the current staffing levels and operating budget for the health centre. The production possibilities frontier which is the curved line shows what could be achieved with maximal technical efficiency for both activities. Many health centres operate far from the production possibilities frontier. Consider a health centre at point A in Fig. 1; joint costing rules would allocate equal shares of the health centre's overhead costs to EPI and prenatal care. Imagine a new regional manager who works to increase the efficien- cy of EPI such that at no extra cost to the health centre it now operates at point B. Joint costing rules would now attribute a much higher share of the over- b While the cost-effectiveness analysis methods used in devel- oping the packages of care for the WDR do not explicitly address the health system, the WDR devotes the whole of Chapter 6 to the need for developing health systems. Fig. 1. Production function for hypothetical health cen- tre undertaking only two activities-prenatal care and expanded programme on immunization; misallocation of overhead costs through use of joint-costing rules. Point A represents a typical level of output and point B the increased level achieved through improved management. co 0 Expanded programme on immunization head costs to EPI than before. Clearly, fixed costs have not increased; only productivity has increased. The joint costing approach to calculating average costs entails a very real risk of penalizing with high- er estimated unit costs those programmes that are more efficient. To examine investments in human and physical health infrastructure in a cost-effectiveness frame- work, a more sophisticated approach to resource allocation questions is required. Correa (52) and Tor- rance et al. (38) developed hypothetical planning models to choose health maximizing mixes of inter- ventions under various constraints. Torrance et al. (38) discussed the possibility of designing a resource allocation model that would directly incorporate the limits on service delivery imposed by the current health system infrastructure and the possibility of improving the health system. At least four optimiza- tion models for the health sector applied to specific interventions that maximize a measure of health stat- us given a budget constraint and a variety of possible interventions have been developed (37, 53-55). None of these applications, however, attempted to incorporate the health system into the modelling exercise. Health Resources Allocation Model (HRAM) In order to deal with these problems, we have devel- oped at Harvard an optimization model for the health sector based on the burden of disease, the cost-effec- tiveness of available health interventions, and the available health system infrastructure. Our model has been developed in the General Algebraic Modeling System (GAMS), a computer system which facili- WHO Bulletin OMS. Vol 72 1994 665 C.J.L. Murray et al. tates the development of algebraic models in days, which previously took months (56). GAMS has been extensively used in other fields such as agriculture, education and industry to deal with complex non- linear optimization problems. Our model, HRAM, has also been designed to address technical problems related to intervention fixed costs, rising marginal costs, and regional heterogeneity. The details on the latter and the technical specifications of the model are provided elsewhere (36) and are not discussed in this paper in detail. The following discussion de- scribes the general strategy used to incorporate the health system into a cost-effectiveness framework. In order to put the appraisal of infrastructure in cost-effectiveness terms, we have defined several budget constraints that include interchangeable dol- lars and a series of constraints reflecting the current capacity of the health system to deliver various types of services. While there is flexibility in the design of the model to specify various types of budget con- straints, we have so far included constraints for ser- vices delivered at referral hospitals, district hospitals and health centres. Given current facilities and staff- ing levels, the Ministry of Health begins with a con- straint on the volume of services it can provide through referral hospitals, district hospitals and health centres. For referral and district hospitals, we have used bed-days as the unit of service delivery, and for health centres we have used patient-contact equivalents. Each intervention or activity may consume referral hospital bed-days, district hospital bed-days or health centre contacts in addition to interchange- able dollars. In other words, the use of the general health system infrastructure is captured in terms of units of service rather than using arbitrary joint cost- ing rules. Table 1 provides examples of production functions for several health interventions. In choos- ing an optimal allocation of health resources across activities, when the available budget of district hos- pital bed-days is exhausted, no further interventions using district hospitals can be bought. The same lim- itation would apply to referral hospital bed-days and health centre contacts. However, the govemment may choose to build new referral hospitals, district hospitals or health centres in order to relax the ser- vice constraint. In addition to the set of interventions included in the model, three more are added: con- struction and staffing of a referral hospital, district hospital or health centre. For health centres, we have also included a geographical access constraint. It is not sufficient to have an adequate total number of health centre contacts for the population; health cen- tres must be positioned close enough to the commu- nity so that they can use them. In the simulations for sub-Saharan Africa, expanding geographical access to health centres, particularly in remote areas, is a major force driving the expansion of infrastructure in an optimal resource allocation. While not included so far, geographical access constraints could also be added for district hospitals. At each budget level, the resource allocation model searches to see if the total output of the system in terms of DALYs avoided could be in- creased by using some resources to expand the health system rather than spending it on particular activities delivered with the current health infrastruc- ture. In other words, the ability of computers to undertake repetitive calculations at high speed is used to test if the total output of the health sector in terms of DALYs would be higher or lower by improving the health system. Improvements in the health system can be undertaken in this model by building, staffing and operating new referral hospi- tals, district hospitals or clinics.c In this framework, infrastructure investments can be evaluated in terms of the increase in the number of DALYs or equiva- lent measure of health status. As one purchases an intervention such as meas- les immunization at the point where all children are immunized, the marginal cost per DALY reaches infinity because no more health benefits are gained by expanding coverage beyond 100%. While techni- cally correct in micro-economic jargon, it is a cum- bersome approach to capturing the practical limits of each intervention. More intuitive is to constrain the purchase of each intervention by the total amount of DALYs that can be addressed with a particular inter- vention in a particular community. The link between the burden of disease or the total number of DALYs lost due to a particular health problem and cost- effectiveness is thus established. The example of the model, which is described below, makes use of the Global Burden of Disease study results (19) for sub- Saharan Africa. The estimates of the current burden of disease had to be modified to remove the impact of currently financed health interventions on the measured burden of disease. To explore the use of such an optimization model, we have used the World Bank's Health Sec- tor Priorities Review database on the cost-effective- ness of some 50 interventions (7), the same database utilized by Bobadilla et al. (24). Each estimate of c In this version of the model, we are able to build new infra- structure and use it in the same time period. The costs of open- ing and operating new units of infrastructure are the annual operating fixed costs plus the equivalent annual capital cost. As the model is not a multi-period model, we do not take into account the necessary time lag between the decision to improve the physical or human infrastructure of the health system and its implementation. 666 WHO Bulletin OMS. Vol 72 1994 Cost-effectiveness analysis and health sector policy choices Table 1: Data for the Health Resource Allocation Model for five interventions Per DALY Programme- Marginal Referral District Health specific fixed cost func- hospital hospital centre Intervention Segmenta costs (US $) tion (US $) (bed-days) (bed-days) (contacts) ARI screening 0 40 000 24.23 0 0.20 4.00 1 27.26 2 30.29 3 33.32 4 36.35 Poliomyelitis immunization 0 60 000 9.17 0 0 5.52 1 14.66 2 18.33 3 36.66 4 183.30 School-based 0 100 000 4.79 0 0 0 anti-helminthic 1 4.79 chemoprophylaxis 2 5.99 3 7.19 4 7.19 Short-course chemotherapy 0 53 000 1.71 0 3.85 0 for sputum smear-positive 1 2.74 tuberculosis 2 3.42 3 6.84 4 34.20 Tetanus referral 0 200 000 24.08 2.29 0 0 1 32.10 2 40.13 3 48.16 4 56.18 a To approximate the increasing marginal cost, the nonlinear marginal cost curve for each intervention within a region is broken up into five linear segments numbered from 0 to 4. cost-effectiveness was reviewed and modified to increase the comparability across interventions. Despite our attempts to unearth details, frequently only average cost results are reported in the literature or reports on cost-effectiveness. Where necessary, expert judgement was used to develop the intervention production functions and form of the marginal cost curve, examples of which are provided in Table 1.d The model was run for a hypothetical sub-Saha- ran African country with a population of 10 million and a GDP per capita of $340, using the regional GBD results adjusted to the total population to deter- mine the DALY limits for each disease. A digression on the burden of disease is necessary. The results of the Global Burden of Disease study provide an esti- mate of the current burden of disease. Current or d Table 1 provides a stepped marginal cost function for five interventions. For convenience, we divided non-linear marginal cost functions into five linear steps or segments which are sum- marized in the Table. measured burden incorporates the impact of current- ly financed health interventions; for example, if mea- sles immunization coverage is 70%, then a signifi- cant share of the burden of measles has already been avoided. There are three levels of the burden of dis- ease relevant to this discussion of resource alloca- tion: first, the current burden of disease; second, the burden of disease that would be present if currently financed health interventions were stopped; and third, the lowest achievable burden given a technical and allocative efficiency within a budget constraint. The resource allocation model used as an input esti- mated the burden in the absence of currently fi- nanced health interventions in order to calculate the lowest achievable burden of disease for a given budget. Fig. 2 shows the expansion path for the optimal allocation of health resources to maximize DALYs averted at each budget level. For reference, current expenditure in sub-Saharan Africa excluding South Africa is US$ 14 per capita. The equivalent annual capital cost of the existing health infrastructure and the fixed operating costs of the health system are WHO Bulletin OMS. Vol 72 1994 667 C.J.L. Murray et al. Fig. 2. DALY retrieval expansion path for sub-Saharan Africa. 3000 o 2500 'as 2000 1500 1. 75 2.5 3.25 4 4.75 Spending (% of GDP) 5.5 6.25 US$ 3.27 per capita. This health sector production function indicates that with increasing expenditure the marginal cost of each DALY purchased increases rapidly. For all budget ranges included, the marginal cost per DALY is higher than average cost. Table 2 shows the number of new referral hospitals, district hospitals and health centres bought at three budget levels. Even at current levels of health expenditure, nearly 25% of the budget should be spent on expand- ing the health system. The remainder should be spent on the set of interventions listed. The Table also pro- vides the utilization rates of the three types of facil- ities modelled at each budget level. Referral hospital bed-occupancy is less than 1%. The implication is that there is excess referral hospital capacity but a shortage of district hospital capacity. If closure or down-sizing of referral hospitals were politically fea- sible, this desirable option could be added as an additional intervention in the model. Table 3 shows the allocation to specific inter- ventions at three budget levels. Some highlights are worth discussing. Comparison with the WDR's $12 per capita package for low-income countries is diffi- cult, as their package is based on a marginal increase of $12 per capita, given current expenditures. At cur- rent budget levels, the most important interventions by expenditure are screening and treatment of acute respiratory infections (ARI), malaria control, tuber- culosis chemotherapy, measles immunization, oral rehydration therapy, breast-feeding, tetanus immu- nization, and hygiene promotion. With increases over current budget levels, the major gainers are chemotherapy for sputum smear-negative tuberculo- sis cases, oral rehydration therapy, malaria control and hygiene promotion. Fig. 2 shows two expansion paths. The top line is the expansion path for the complete model. The second line is the expansion path when the options of adding infrastructure are removed from the model. Simple inspection shows that expanding the infra- structure is a tremendously important component of health improvement. At current expenditure levels in sub-Saharan Africa, expanding the health system produces nearly 40% more total DALYs. Policy choice and sectoral cost- effectiveness Having illustrated a model that incorporates health system investment choice into a cost-effectiveness framework, we can return to the nature of policy questions that can be treated with these analytical tools. Three distinct policy questions using burden of disease and cost-effectiveness results can be framed. (1) Ground-zero. Given a fixed budget and health infrastructure, how can non-fixed resources be spent so as to maximally reduce the burden of disease? (2) Marginal expansion. Given an existing health infrastructure and a set of currently financed activ- ities, none of which can be changed, how best can marginal increases in the health sector resources be spent so as to maximally reduce the burden of dis- ease? (3) Politically constrained ground-zero. Given an existing health infrastructure, for political or other reasons there may be a set of services or activities that are deemed to be 'protected' from changes in budget and a set of other services or activities that could be expanded or contracted. For a fixed health sector budget, how can health resources be reallocat- ed to maximally reduce the burden of disease with- out reducing the resources allocated to 'protected' activities? Table 2: Infrastructure expansion and rising budget levels At 3% of GDP At 4% of GDP At 5% of GDP Additional Utilization Additional Utilization Additional Utilization Facility type facilities rate (%) facilities rate (%) facilities rate (%) Referral hospital 0 0 0 0 0 0.61 District hospital 41 99 47 100 53 98 Clinic 411 56 53 64 578 74 WHO Bulletin OMS. Vol 72 1994 350 HRAM -.1 -- I -- __ -without infrastructure expansion 668 Cost-effectiveness analysis and health sector policy choices Table 3: Allocations to specific interventions at varying health budget levels for a hypothetical sub-Saharan African countrya Intervention Fixed infrastructureb ARI screening and referral Oral rehydration therapy BCG added to DPT Hepatitis B immunization lodination of salt or water Measles immunization Poliomyelitis immunization Semiannual vitamin-A dose for children 0-5 years Tetanus immunization Breast-feeding promotion w/education or hospital routine for diarrhoeal diseases Improved weaning practices from education Oral iron supplementation during pregnancy Chlamydia treatment w/antibiotics Gonorrhoea treatment w/antibiotics Syphilis treatment w/antibiotics HIV blood screening Annual breast examinations Antibiotics for rheumatic heart disease Cataract surgery CVD preventive programme Improved domestic and personal hygiene Injected insulin and health education for IDDMc Leprosy multidrug clinic Low-cost management of acute Mlc Pap smear at 5-year intervals Pneumococcal vaccine Schizophrenia School-based anti-helminthic chemoprophylaxis Short-course chemotherapy for sputum- negative patients Short-course chemotherapy for sputum- positive patients Sugar or salt fortified with iron Tetanus referral case management Vector control for malaria Added infrastructure At 3.0% of GDP Spending DALYS ('000 $) ('000 $) 32 697 6 879 233 789 12 920 71 391 8 249 32 4 986 272 907 30 577 38 1 651 213 2 564 74 1 526 46 70 1 107 111 147 911 0 0 860 0 0 0 537 0 246 884 248 597 6 688 6 9 156 39 0 0 9 0 0 0 6 0 1 16 3 12 372 3 498 453 124 0 12 123 20 713 19 0 304 At 4.0% of GDP Spending DALYS ('000 $) ('000 $) 32 699 11 107 277 4 831 53 1 783 80 505 9 249 32 7 686 298 1 232 33 881 41 2 042 222 2 755 77 1 526 46 70 1 107 111 147 962 0 388 935 0 5 306 0 541 0 227 2 543 331 597 10 208 6 9 156 40 0 3 10 0 47 0 7 0 2 32 3 12 415 At 5.0% of GDP Spending DALYS ('000 $) ('000 $) 32 698 11 107 277 15 661 123 2 291 84 505 9 249 32 12 545 324 1 503 34 881 41 2 042 222 2 755 1 526 70 107 111 147 962 312 722 935 287 9 658 231 541 780 422 3206 331 597 14 120 77 46 1 6 9 156 40 5 10 2 72 0 7 3 2 36 3 12 443 5 018 484 5 218 487 124 0 16 587 24 502 19 0 348 124 1 176 18 942 27 238 19 3 364 Total costs ('000 $) Total cost per capita ($) 102 000 10.20 2 435 136 000 13.60 2 762 170 000 17.00 2 950 a Population is assumed to be 10 million and GDP per capita $340. b Fixed infrastructure reports the costs of construction, maintenance and staffing of the clinics, district hospitals and referral hospitals, which are assumed to have been constructed previously. Assumptions are based on infrastructure data for sub-Saharan African countries. c IDDM: insulin-dependent diabetes mellitus. Ml: myocardial infarction. All three and combinations of (2) and (3) can be addressed using the burden of disease and cost- effectiveness information, as described above for sub-Saharan Africa. The inputs to the process, how- ever, to answer each of these questions will be dif- ferent. Table 4 shows that the budget constraint on the purchase of interventions is fixed at the current level to answer questions (1) and (3), while for the second question there is a marginal increase in the budget available to buy further interventions. WHO Bulletin OMS. Vol 72 1994 669 C.J.L. Murray et al. Table 4: Setting health sector priorities: different Inputs to answer different policy questions Inputs to optimization or packaging Budget Burden Health system 1. Ground-zero Current budget minus fixed cost Burden in absence of currently Total available capacity of operating current health financed health interventions infrastructure 2. Marginal expansion Marginal increase in budget Current burden Unused capacity 3. Politically constrained Current budget minus fixed cost For protected, current burden Total capacity, less capacity ground-zero of operating health infrastructure For remainder, burden in used for protected services and cost of protected activity absence of currently financed health interventions The burden of disease estimates that should be used either in HRAM or in the World Bank packag- ing exercise will be different for the three questions. The first question, which can be labelled the ground- zero exercise, was the one addressed by the HRAM applied to sub-Saharan Africa. The burden of disease in the absence of currently financed interventions is the required input. To allocate marginal increases in resources, maintaining currently financed activities, the currently observed burden of disease is the appropriate input. Finally, to answer the third ques- tion, we would want to use the current burden of dis- ease for those conditions affected by currently financed and protected activities and the burden of disease in the absence of currently financed activities for those activities that are not protected. Finally, the approach to the infrastructure con- straints would also be different for the three ques- tions. The ground-zero exercise would use total available capacity at each level of the health system as the constraint on service delivery with the option for building new infrastructure. The input to the mar- ginal budget exercise would be the unused capacity at each level with the option of building new infra- structure. The politically constrained exercise would use total capacity at each level minus the capacity used to deliver 'protected' services. In the WDR, the World Bank proposes a pack- age of essential public health and clinical services that would cost $12 in a low-income country. This package is meant to be a marginal package of expen- diture and health gain on top of currently financed activities. Some confusion is generated when the World Bank states, "In fact, in the poorest countries total current public spending of $6 per person is about $6 short of the cost of the package. Total per capita spending, including private spending, is about $14, about the same as the proposed package." (23, page 67). The package, however, has been described at several junctures as addressing the marginal bur- den of disease and has been calculated using the cur- rent burden, not the burden in the absence of current- ly financed activities.e In other words, by the nature of its calculation, the World Bank package is a mar- ginal package on top of current expenditure. Paying for the package in a low-income country is not a question of resource reallocation but a question of increasing health expenditure by $12 per capita or a doubling of total health sector expenditure in a low- income country; if the increase was to come entirely from the public sector it would entail a tripling of publicly financed health expenditure. Potential con- fusion around the policy question that is being asked and the appropriate method of calculation by the World Bank highlights the importance of developing a consistent and internally rigorous approach to using cost-effectiveness for informing sectoral resource allocation questions. Implications The cost-effectiveness of interventions, the burden of disease, and information on the human and physical infrastructure in a health system can be combined to answer a host of resource allocation questions including variations of protected expenditures and marginal budget increases. Using a computer pro- gram like the one illustrated here, investments in the health system can be directly compared with ex- panding resources for particular interventions for a given level of the health system. The preliminary work presented on such models can easily be developed to incorporate other investments in health system quality or coverage. Investments in health infor- mation systems or training can be included as long e The method used to calculate the package is different for dif- ferent interventions. Some of the package is based on a cost per person receiving a service and an estimate of the desired coverage of the service so that this is closer to the ground-zero analysis. For others, including most of the clinical services, the package is estimated, based on the current burden of disease and the cost per DALY averted. 670 WHO Bulletin OMS. Vol 72 1994 Cost-effectiveness analysis and health sector policy choices as the chain of causation between these investments and improvement in health through the delivery of specific health interventions can be traced. More sophisticated versions of such a computer program could take into account the delay between the deci- sion to improve the health system and the comple- tion of new construction or training. A multi-period model would also allow for incorporating expected changes in the burden of disease due to demographic and epidemiological changes (57, 58). If more widespread use of cost-effectiveness databases to inform health sector resource allocation is intended, then it will be important to alter the stan- dards of reporting cost-effectiveness studies in the literature. Frequently studies report only an average cost per unit of service delivered or health benefit such as a DALY. Details on the component costs are often not provided. In order to examine the cost- effectiveness of investing in the health system, we must shift to reporting the different resources used in providing a health intervention rather than costs. Table 1 illustrates crude forms of such resource use profiles where the component inputs such as bed- days, clinic contacts, or outreach workers are denom- inated. More detailed resource use profiles could be provided outlining specific inputs and the necessary quality of the inputs such as nursing or surgeons' time, etc. There is an urgent need to develop a sim- ple but useful categorization of the inputs to health service production that provides sufficient detail for sectoral analysis. Another major benefit from a shift to reporting production functions would be to increase the trans- ferability of cost-effectiveness results from one envi- ronment to another. In the WDR, studies on the cost- effectiveness of a programme in the United Republic of Tanzania are directly compared with results of studies in Brazil where the same input such as nurs- ing time can be ten times more expensive in dollar terms; nevertheless, the World Bank is well aware of these limitations and the urgent need to refine meth- ods to transfer cost-effectiveness results from one context to another. Such comparisons obscure the real use of resources for health programmes which would be transparent if resource use is reported directly. We hope that the World Bank and the World Health Organization will take the lead in developing a standard approach to reporting health intervention resource use profiles and ultimately pro- duction functions. In this paper, we have argued that cost-effective- ness results, information on the burden of disease, and details on the available health system resources can be combined to provide useful insights into a wide range of questions on allocation of health sec- tor resources. All these questions, however, are com- plicated and require at present the assistance of sophisticated computer algorithms to define prefer- able patterns of resource allocation. Given the early stage of development of this sectoral application of cost-effectiveness, it appears that computer pro- grams, such as HRAM, will remain an essential adjunct to policy analysis. Ultimately, as these meth- ods are tested in a range of countries, simpler deci- sion rules may be developed that will allow for more rapid application of the cost-effectiveness and disease burden results to questions of resource allocation. Despite the challenges raised in this paper, the method proposed by the WDR and others remains a much better alternative to current practice. We should not let the perfect be the enemy of the good. The World Bank's Health Sector Priorities Review and the 1993 WDR have advanced technical analysis of health policy choices in developing countries by years if not decades. On the other hand, we must always remain cognizant of the fact that technical analysis of health sector priorities using the burden of disease, the cost-effectiveness of interventions, and the available resources is only one input to the policy process and is not intended to be a rigid pre- scription for all health system ailments. Acknowledgements We gratefully acknowledge the extensive efforts of John Kim and Robert Ashley. Comments and suggestions from Julio Frenk, Dean Jamison, Jose-Luis Bobadilla, Philip Musgrove, and Peter Berman have been very helpful. Resume Investigation du secteur de sant6: analyse coOt-efficacite et choix politiques Les 6tudes actuelles sur l'affectation des res- sources en fonction du rapport cout-efficacit6 - notamment l'Oregon State plan et le Rapport sur le d6veloppement dans le monde 1993 de la Banque mondiale - pretent le flanc a deux cri- tiques importantes. Tout d'abord, les analyses cout-efficacite tendent a negliger le r6le des infra- structures de sant6. Ensuite, se pose le probleme des choix politiques qui devraient etre documen- t6s par une analyse cout-efficacit6 de I'affectation des ressources. En premier lieu, les 6tudes qui negligent le r6le de l'infrastructure dans I'affectation des res- sources d'apres leur cout-efficacit6 supposent implicitement que l'infrastructure physique est infi- niment divisible. La m6thode utilisant la liste des coOts moyens (comme celle utilis6e dans le rap- WHO Bulletin OMS. Vol 72 1994 671 C.J.L. Murray et al. port de la Banque mondiale) suppose que des fractions des installations, 2% d'un h6pital de dis- trict par exemple, peuvent etre construites confor- m6ment aux normes. Une nouvelle difficulte inter- vient avec les effets indesirables de 1'etablissement conjoint des coOts. Cet effet infrastructure peut etre corrige en utili- sant un modele informatise tel que le Harvard Health Resources Allocation Model (HHRAM) qui, comme dans le rapport de la Banque mondiale, integre le rapport cout-efficacit6 et le poids de la morbidit6. Le HHRAM a e applique a un pays d'Afrique subsaharienne fictif, ayant une population de 10 millions d'habitants et un PIB par habitant de US$ 340; ce modble, qui tient compte de 1'expan- sion des infrastructures de sante dans I'affectation des ressources, donne un nombre total de DALY superieur de 40% a ce que donne un modele negli- geant l'infrastructure. Au niveau des budgets actuels - qui pour I'Afrique subsaharienne, a 1'exclusion de I'Afrique du Sud, est de US$ 14 par habitant - les interventions les plus importantes compte tenu des depenses sont le d6pistage et le traitement des infections respiratoires aigues, la lutte antipaludique, la chimioth6rapie antitubercu- leuse, la vaccination antirougeoleuse, la rehydrata- tion orale, I'allaitement au sein, la vaccination anti- tetanique et l'am6lioration de l'hygiene. Le second point est que I'analyse coOt- efficacite de I'affectation des ressources permet de traiter trois questions de politique distinctes, chacune avec ses propres contraintes budgetaires et infrastructurelles et ses propres estimations du poids de la morbidite. 1) Affectation a partir du niveau z6ro: etant donnes un budget fixe et une infrastructure de sant6, comment des ressources non fix6es peuvent-elles etre d6pens6es de facon a diminuer au maximum le poids de la morbidit6? 2) Affectation de ressources a 1'expansion margi- nale: etant donn6s une infrastructure de sant6 et un ensemble d'activit6s actuellement financ6es, dont aucune ne peut etre modifi6e, comment les augmentations marginales des ressources du sec- teur de sante peuvent-elles etre depens6es de facon a diminuer au maximum le poids de la mor- bidite ? 3) Affectation au niveau zero politique- ment limit6e: le budget du secteur de sant6 etant fixe, comment les ressources pour la sant6 peu- vent-elles etre r6affectees pour diminuer au maxi- mum le poids de la morbidite sans diminuer les ressources affect6es aux activit6s ",prot6g6es,,. Le probleme de savoir quelle est la question trai- tee par une etude doit etre evite en elaborant une methode coherente et rigoureuse d'utilisation du rapport coOt-efficacite pour documenter l'affecta- tion de ressources. Une consequence de I'analyse ci-dessus est que pour examiner le rapport cout-efficacite de l'investigation des systemes de sante, il faut ren- dre compte des fonctions de production des inter- ventions de sant6 plut6t que des couts. Nous sou- haitons que la Banque mondiale et l'Organisation mondiale de la Sante donnent l'exemple en met- tant au point une methode codifi6e pour l'evalua- tion des fonctions de production des interventions de sant6. Malgr6 les problemes soulev6s dans cet article, la methode propos6e par le rapport de la Banque mondiale et divers auteurs reste la meil- leure alternative a la pratique actuelle. Toutefois, ne pas oublier que I'analyse technique des priori- t6s du secteur de sante ne pr6tend pas etre une prescription rigide destinee a traiter tous les maux des systemes de sant6. References 1. Klarman HE, Francis JOS, Rosenthal G. Cost- effectiveness analysis applied to the treatment of chronic renal disease. Medical care, 1968, 6: 48-54. 2. Elixhauser A et al. Health care CBA/CEA: an update on the growth and composition of the litera- ture. Medical care, 1993, 31: jsl-jsl 1. 3. Williams AH. Economics of coronary artery bypass grafting. British medical journal, 1985, 291: 326-329. 4. Torrance GW, Zipursky A. Cost-effectiveness of antepartum prevention of Rh immunization. Clinics in perinatology, 1984, 11: 267-281. 5. Schulman KA et al. Cost-effectiveness of low-dose zidovudine therapy for asymptomatic patients with human immunodeficiency virus (HIV) infection. Annals of internal medicine, 1991, 114: 798-802. 6. Drummond M, Torrance G, Mason J. Cost-effec- tiveness league tables: more harm than good? Social science and medicine, 1993, 37: 33-40. 7. Jamison DT et al., eds. Disease control priorities in developing countries. New York, Oxford University Press, 1993. 8. Allen D, Lee RH, Lowson K. The use of QALYs in health service planning. International journal of health planning and management, 1989, 4: 261-273. 9. Oregon Health Services Commission. Oregon Medicaid priority setting project. Portland, 1991. 10. Brown LD. The national politics of Oregon's ration- ing plan. Health affairs, 1991, 10(2): 28-51. 11. Dixon J, Welch HG. Priority setting: lessons from Oregon. Lancet, 1991, 337: 891-894. 12. Fox DM, Leichter HM. Rationing care in Oregon: the new accountability. Health affairs, 1991, 10(2): 7-27. 13. Garland MJ, Hasnain R. Health care in common: setting priorities in Oregon. Hastings Center report, September-October 1990: 16-18. 14. Hadorn DC. Setting health care priorities in Ore- gon-cost-effectiveness meets the rule of rescue. Journal of the American Medical Association, 1991, 265: 2218-2225. 672 WHO Bulletin OMS. Vol 72 1994 Cost-effectiveness analysis and health sector policy choices 15. KIevit HD et al. Prioritization of health care ser- vices-a progress report by the Oregon Health Ser- vices Commission. Archives of internal medicine, 1991, 151: 912-916. 16. 0stbye T, Speechley M. The Oregon formula: a better method of allocating health care resources. Nordisk medicin, 1992, 107(3): 92-95. 17. Eddy DM. Oregon's methods-did cost-effectiveness analysis fail? Journal of the American Medical Asso- ciation, 1991, 266: 2135-2141. 18. Kitzhaber JA. Prioritising health services in an era of limits: the Oregon experience. British medical journal, 1993, 307: 373-377. 19. Murray CJL, Lopez AD, Jamison DT. The global burden of disease in 1990: summary results, sensi- tivity analysis and future directions. Bulletin of the World Health Organization, 1994, 72: 495-509. 20. Murray CJL. Quantifying the burden of disease: the technical basis for disability-adjusted life years. Bulle- tin of the World Health Organization, 1994, 72: 429-445. 21. Murray CJL, Lopez AD. Quantifying disability: data, methods, and results. Bulletin of the World Health Organization, 1994, 72: 481-494. 22. Murray CJL, Lopez AD. Global and regional cause- of-death patterns in 1990. Bulletin of the World Health Organization, 1994, 72: 447-480. 23. World Bank. World development report 1993: investing in health. New York, Oxford University Press, 1993. 24. Bobadilla J-L et al. Design, content and financing of an essential national package of health services. Bulletin of the World Health Organization, 1994, 72: 653-662. 25. Garber AM, Phelps CE. Economic foundations of cost-effectiveness analysis. Cambridge MA, National Board of Economic Research, 1992 (NEBR working paper 4164). 26. Phelps CE, Mushlin Al. On the (near) equivalence of cost-effectiveness and cost-benefit analyses. International journal of technology assessment in health care, 1991, 7(1): 12-21. 27. Loomes G, McKenzie L. The use of QALYs in health care decision-making. Social science and medicine, 1989, 28(4): 299-308. 28. Birch S, Gafni A. Cost-effectiveness/utility analy- ses-do current decision rules lead us where we want to be? Journal of health economics, 1992, 11: 279-296. 29. Johannesson M, J6nsson B. Economic evaluation in health care: is there a role for cost-benefit analy- sis? Health policy, 1991, 17: 1-23. 30. Hammer JS. Prices and protocols in public health care. Washington, World Bank, 1993 (Population Health and Nutrition Working Paper Series). 31. Mehrez A, Gafni A. Quality-adjusted life years, util- ity theory, and healthy-years equivalents. Medical decision-making, 1989, 9:142-149. 32. Miyamoto JM, Eraker SA. Parameter estimates for a QALY utility model. Medical decision-making, 1985, 5(2): 191-213. 33. Sackett DL, Torrance GW. The utility of different health states as perceived by the general public. Journal of chronic disease, 1978, 31: 697-704. 34. Pliskin JS, Shepard DS, Weinstein MC. Utility functions for life years and health status. Operations research, 1980, 28(1): 206-224. 35. Torrance GW. Measurement of health state utilities for economic appraisal: a review. Journal of health economics, 1986, 5:1-30. 36. Murray CJL, Kreuser J, Whang W. A cost-effec- tiveness model for allocating health sector resourc- es. Cambridge MA, Harvard Center for Population and Development Studies 1994 (Health Transition Working Paper). 37. Forgy L. Cost-effectiveness in child health: a mini- mum information approach to planning and forecast- ing. In: Hill K, ed., Child health priorities for the 1990s. Baltimore, Johns Hopkins University Institute for International Programs, 1992: 309-335. 38. Torrance GW, Thomas WH, Sackett DL. A utility maximization model for evaluation of health care programs. Health services research, Summer 1972: 118-133. 39. Daniels N. Just health care. New York, Cambridge University Press, 1985. 40. Lind R. Discounting for time and risk in energy poli- cy. Baltimore, Johns Hopkins University Press, 1982. 41. Little I, Mirrlees S. Project appraisal and planning for developing countries, London, Heinemann, 1974. 42. Hartman RW. One thousand points of light seeking a number: a case study of CBO's search for a dis- count rate policy. Journal of environmental econom- ics and management, 1990, 18(suppl.):s3-s7. 43. Martens LLM, van Doorslaer EKA. Dealing with discounting. International journal of technology assessment in health care, 1990, 6: 139-145. 44. Hammit J. Discounting health increments. Journal of health economics, 1993, 12: 117-120. 45. Krahn M, Gafni A. Discounting in the economic evaluation of health care interventions. Medical care, 1993, 31: 403-418. 46. Olsen J. On what basis should health be discount- ed? Journal of health economics, 1993, 12: 39-53. 47. Viscusi K, Moore M. Rates of time preference and valuations of the durations of life. Journal of public economics, 1989, 38: 297-317. 48. Johanneson M. On the discounting of gained life- years in cost-effectiveness analysis. International journal of technology assessment in health care, 1992, 8(2): 359-364. 49. Parsonage M, Neuberger H. Discounting and health benefits. Health economics, 1992, 1: 71-76. 50. Messing SD. Discounting health: the issue of sub- sistence and care in an underdeveloped country. Social science and medicine, 1973, 7: 911-916. 51. Keeler E, Cretin S. Discounting of life-saving and other nonmonetary effects. Management science, 1983, 29: 300-306. 52. Correa H. Health planning. Kyklos international review for social science, 1967, 20(4): 909-923. 53. Feldstein MS, Piot MA, Sundaresan TK. Resource allocation model for public health planning: a case study of tuberculosis control. Geneva, 1973 (supple- ment to vol. 48 of the Bulletin of the World Health Organization). 54. Chen MM, Bush JW. Maximizing health system output with political and administrative constraints using math- WHO Bulletin OMS. Vol 72 1994 673 C.J.L. Murray et al. ematical programming. Inquiry, 1976, 13: 215-227. 55. Barnum H. A resource allocation model for child survival. Cambridge MA, Oelgeschlager, Gunn and Hain, 1980. 56. Brooke A, Kendrick D, Meeraus A. GAMS - a user's guide release 2.25. South San Francisco, The Scientific Press, 1992. 57. Omran AR. The epidemiologic transition theory: a preliminary update. Journal of tropical pediatrics, 1983, 29: 305-316. 58. Bobadilla JL et al. The epidemiologic transition and health priorities. In: Jamison DT et al., eds. Disease control priorities in developing countries. New York, Oxford University Press, 1993: 673-702. 674 WHO Bulletin OMS. Vol 72 1994