At the time of this study, the disease pattern in Zimbabwe is heavily dominated by communicable, maternal, perinatal and nutritional conditions 19 similar to other countries in Sub-Saharan Africa although Zimbabwe is plagued with an unusually large disease burden due to HIV (Annex 1). The health of the nation has traditionally been a high priority and large investments in the public health care sector in the 1980s led to impressive improvements in key health indicators although the years following 1990 saw a reversal in most health indicators 20 – a development further exacerbated in more recent time due to decreasing GDP, dwindling health care funds and massive emigration of health sector personnel 21. The health care sector is a highly heterogeneous section of the economy. Provision of health care services is offered by government, church missions and other NGOs, industries and mines, private practitioners and traditional healers. Measured by the number of health facilities, government is the single most important provider 22. Private practitioners and hospitals are relatively abundant in larger cities where these providers are able to attract large proportions of the available health personnel. Government of Zimbabwe has succeeded in organising its own institutions as well as church mission facilities and some of the private sector facilities into a four-tiered system of health care service delivery. Health centres manned by qualified nurses are the first level followed at the next levels by district, provincial and central hospitals where hospital services of increasing complexity are offered requiring more specialised personnel and equipment. The head office of the Ministry of Health and Child Welfare constitutes the highest level of the public health care sector and it is the main actor in terms of health policy making and development. For instance, the head office is responsible for the allocation of all government health care funds among health facilities as well as steering important processes such as the Zimbabwe Essential Drugs Action Programme (ZEDAP) which results in a list specifying the most cost-effective drugs for a large number of health problems 23 that is used extensively by all health facilities in the country.

Curative interventions for the present study included the treatment of common health problems at hospital inpatient and outpatient departments as well as health centres. These interventions covered both single treatment episodes and more long term management of chronic conditions. Preventive interventions included five vertical activities: residual house spraying to prevent malaria, immunisation of children (measles, polio, tuberculosis, diphtheria, pertussis and tetanus), surveillance and targeted supplementary feeding of wasted children, HIV prevention through improved access to treatment of sexually transmitted infections (STIs) and health promotion of personal and domestic hygiene in order to decrease the incidence of diarrhoeal diseases.

In order to estimate the costs of individual curative and preventive health interventions, a number of public health providers were visited for the collection of the necessary cost data. Study sites were randomly chosen from all over the country. With respect to curative health interventions, six health centres out of a total of around 1200 were selected for the cost analysis. Health centres offered outpatient services and selected preventive activities such as immunisation. Five district level hospitals including two mission hospitals from a total of 130 hospitals were sampled for the costing of inpatient services, surgical procedures and outpatient services. Finally, two provincial hospitals (from a total of 8) were randomly selected and these offered similar services as district hospitals but the former hospitals were able also to provide more specialised services. The highest level, central hospitals, was excluded from the costing analysis. Preventive interventions were organised in a vertical fashion involving provincial health offices and district hospitals as well as services performed by health facilities (e.g. vaccinations at health centres and hospitals). Two provinces out of a total of eight were randomly chosen and two districts were randomly selected within each province (a total of 14 districts). Finally, the Ministry of Health Headquarters and two provincial health offices were visited to capture additional programme costs of curative and preventive interventions such as central purchasing of drugs and high level administrative personnel 10.

The costing perspective taken for this study was the health provider's view (Ministry of Health and Child Welfare) since the objective of the present cost-effectiveness analysis was to determine how the largest slice of the burden of disease could be cut using a given government budget 24.

Activities at each study site incorporated the identification, measurement and subsequent valuation of resources required to offer health services. Government accounting systems provided at each study site the level of actual, recurrent expenditure by category including for example salaries by type of personnel, stationery, electricity, maintenance and drugs. With respect to capital inputs at each study site, a quantity surveyor estimated the present day construction costs per square metre by type of office or department. Further, a list of available equipment and furniture was developed and subsequently valued using market prices. From these replacement costs of buildings, equipment and furniture, an annual equivalent was calculated using the annuitization method 2425 assuming a real discount rate of 3% and expected life spans of 30, 7 and 10 years for the mentioned capital inputs.

These costs by category were at each study site allocated to the health interventions selected for this study. This was done by applying the standard step-down costing methodology 2426 consisting initially of categorising activities (in practice wards and departments) in a study site into a hierarchical system with the final product (such as patient care) at the lowest level and with support and overhead activities at successively higher levels. Subsequently, the aggregate costs by category were allocated to final activities in a step-wise fashion using simultaneous equation techniques [24, Ch. 4] and the development of allocation criteria reflecting actual resource use. At the end of the standard step-down costing procedure, all costs of a study site had been distributed to the final service departments so that an average costs figure could be calculated by dividing the number of services provided by individual departments. Micro-costing techniques 27 were used to supplement the above information in order to achieve information on interventions against individual diseases. For instance, a review of a sample of inpatient notes was performed at hospitals in order to capture the treatment pattern of the most common health problems. With respect to the treatment of the less common health problems, official treatment guidelines were used 23.

Having finalised the study activities described above, unit costs of individual curative and preventive services were available for the study sites included.

Unit costs of individual health interventions estimated from data collected at the study sites were utilised for calculating the total costs of offering this service for a population group as a whole. This was done to take account of the fact that costs and effects measured in DALYs averted depended on age of onset of disease. The total costs of a specific curative health intervention were calculated for a hypothetical district population of 250,000 individuals in Zimbabwe with the same age and sex distribution and incidence of diseases as the country as a whole. The number of treatments for each disease was determined by incidence and the health seeking behaviour of the population. Information on incidence of diseases was drawn from a national study which provided estimates of new cases of disease by age and sex groups for the year 1997 18. In addition, the proportions of cases by disease likely to seek treatment were determined based on advice from clinical experts as well as the National Health Information System 19. Using these two types of information, the total number of treatments by age and sex could be estimated for each disease under study. Subsequently, the total costs of a curative health intervention were estimated by multiplying this number with the relevant unit costs:

C j = U j ∑ a ∑ s I a s j H a s j MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xI8qiVKYPFjYdHaVhbbf9v8qqaqFr0xc9vqFj0dXdbba91qpepeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGaem4qam0aaWbaaSqabeaacqWGQbGAaaGccqGH9aqpcqWGvbqvdaahaaWcbeqaaiabdQgaQbaakmaaqafabaWaaabuaeaacqWGjbqsdaqhaaWcbaGaemyyaeMaem4CamhabaGaemOAaOgaaOGaemisaG0aa0baaSqaaiabdggaHjabdohaZbqaaiabdQgaQbaaaeaacqWGZbWCaeqaniabggHiLdaaleaacqWGHbqyaeqaniabggHiLdaaaa@444E@

where C^j is the population level costs of intervention j, U^j indicates the unit costs of curative health intervention j. In addition, Iasj MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xH8viVGI8Gi=hEeeu0xXdbba9frFj0xb9qqpG0dXdb9aspeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGaemysaK0aa0baaSqaaiabdggaHjabdohaZbqaaiabdQgaQbaaaaa@3134@ is the absolute, annual number of incident cases of a health problem (which may be treated by intervention j) in population group of age a and sex s while Hasj MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xH8viVGI8Gi=hEeeu0xXdbba9frFj0xb9qqpG0dXdb9aspeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGaemisaG0aa0baaSqaaiabdggaHjabdohaZbqaaiabdQgaQbaaaaa@3132@ is the proportion of incident cases seeking treatment in the same population group. Outpatient services were offered both at health centres and hospitals. It was assumed that 80% of all cases were treated at health centres and 20% at district hospital outpatient departments corresponding to the actual health seeking behaviour 19. Some health problems required life long treatment like for instance insulin-dependent diabetes. In these cases, the specific cost figures estimated for a given length of time were recalculated to match the life expectancies at various ages of onset of the disease as indicated in the formula below:

C j = ∑ a ∑ s I a s j H a s j ∑ t = 1 T ( a s ) [ ( 1 + r ) − ( t − 1 ) A t j ] MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xI8qiVKYPFjYdHaVhbbf9v8qqaqFr0xc9vqFj0dXdbba91qpepeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=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@5E7C@

where Atj MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xH8viVGI8Gi=hEeeu0xXdbba9frFj0xb9qqpG0dXdb9aspeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGaemyqae0aa0baaSqaaiabdsha0bqaaiabdQgaQbaaaaa@2FDB@ is the annual costs at time t for health intervention j for a chronic condition while T(as) indicates the life expectancy of an individual belonging to population group of age a and sex s. Future costs were discounted using a real discount rate r of 3 percent.

The primary preventive interventions incurred costs at district and provincial health offices and typically also at the level of health providers such as health centres and hospitals. The pattern of cost components for preventive interventions therefore followed the general form:

C j = D j + P j + U j ∑ a ∑ s M a s j N a s j MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xI8qiVKYPFjYdHaVhbbf9v8qqaqFr0xc9vqFj0dXdbba91qpepeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGaem4qam0aaWbaaSqabeaacqWGQbGAaaGccqGH9aqpcqWGebardaahaaWcbeqaaiabdQgaQbaakiabgUcaRiabdcfaqnaaCaaaleqabaGaemOAaOgaaOGaey4kaSIaemyvau1aaWbaaSqabeaacqWGQbGAaaGcdaaeqbqaamaaqafabaGaemyta00aa0baaSqaaiabdggaHjabdohaZbqaaiabdQgaQbaakiabd6eaonaaDaaaleaacqWGHbqycqWGZbWCaeaacqWGQbGAaaaabaGaem4CamhabeqdcqGHris5aaWcbaGaemyyaegabeqdcqGHris5aaaa@4B88@

where D^j and P^j represent the overall costs related to preventive intervention j at the district and the particular district's share of the provincial office respectively. In addition, U^j denotes the unit costs of preventive activities such as vaccinations or STI treatments performed at health centres and hospital outpatient departments. Finally, Masj MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xH8viVGI8Gi=hEeeu0xXdbba9frFj0xb9qqpG0dXdb9aspeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGaemyta00aa0baaSqaaiabdggaHjabdohaZbqaaiabdQgaQbaaaaa@313C@ is the absolute number of individuals in population group of age a and sex s targeted for intervention j and with Nasj MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xH8viVGI8Gi=hEeeu0xXdbba9frFj0xb9qqpG0dXdb9aspeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGaemOta40aa0baaSqaaiabdggaHjabdohaZbqaaiabdQgaQbaaaaa@313E@ denoting the percentage actually covered. Information on the number of individuals in each age and sex group in the study population could be obtained from the most recent census 2829 and updating these figures using estimates of population growth 30. Coverage of the five preventive health interventions was established through discussions with the responsible staff in the four districts. For some activities such as immunisation, information on coverage was collected as part of a recent Demographic and Health Survey 31.

The benefits of an intervention were measured as the reduction in the burden of disease (DALYs averted) as a result of the intervention. Following the Global Burden of Disease methodology 323334, the burden of disease for an individual of sex s dying prematurely at age a, BOD_as, and with life expectancy T(as) (or suffering from a disease episode starting at age a with length T(as)) could be calculated from the formula:

B O D a s = ∫ t = a a + T ( a s ) W K t e − β t e − r ( t − a ) d t MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xI8qiVKYPFjYdHaVhbbf9v8qqaqFr0xc9vqFj0dXdbba91qpepeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGaemOqaiKaem4ta8Kaemiraq0aaSbaaSqaaiabdggaHjabdohaZbqabaGccqGH9aqpdaWdXbqaaiabdEfaxjabdUealbWcbaGaemiDaqNaeyypa0JaemyyaegabaGaemyyaeMaey4kaSIaemivaqLaeiikaGIaemyyaeMaem4CamNaeiykaKcaniabgUIiYdGccqWG0baDcqWGLbqzdaahaaWcbeqaaiabgkHiTiabek7aIjabdsha0baakiabdwgaLnaaCaaaleqabaGaeyOeI0IaemOCaiNaeiikaGIaemiDaqNaeyOeI0IaemyyaeMaeiykaKcaaOGaemizaqMaemiDaqhaaa@56AE@

where W is a quality adjustment factor (disability weight) representing different levels of health [3335: Annex 3]. The component Kte^-βt is an age weighting curve of an inverted u-shape so that the relative value of life years in young adulthood is higher than in other ages while e^-r(t-a) is the discount factor using discount rate r = 0.03. Finally, rather than using actual life expectancies of the population under study, the DALY methodology employs long life expectancies from a low mortality model life table (Coale-Demeny West Level 26 36). Life expectancies T(as) therefore depend on both age and sex. The benefit in terms of DALYs gained from a successful intervention j for a person of age a and sex s is calculated in the following way:

Δ B O D a s j = B O D a s − B O D a s j MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xI8qiVKYPFjYdHaVhbbf9v8qqaqFr0xc9vqFj0dXdbba91qpepeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGaeuiLdqKaemOqaiKaem4ta8Kaemiraq0aa0baaSqaaiabdggaHjabdohaZbqaaiabdQgaQbaakiabg2da9iabdkeacjabd+eapjabdseaenaaBaaaleaacqWGHbqycqWGZbWCaeqaaOGaeyOeI0IaemOqaiKaem4ta8Kaemiraq0aa0baaSqaaiabdggaHjabdohaZbqaaiabdQgaQbaaaaa@44CD@

where BODasj MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xH8viVGI8Gi=hEeeu0xXdbba9frFj0xb9qqpG0dXdb9aspeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGaemOqaiKaem4ta8Kaemiraq0aa0baaSqaaiabdggaHjabdohaZbqaaiabdQgaQbaaaaa@335E@ is the burden of disease after a successful intervention. For instance, the number of DALYs gained for an individual dying prematurely at age a₁ without treatment but postponing death until age a₂ (a₁ <a₂) following an intervention can be calculated using equation (5). A detailed explanation using worked examples of how to calculate DALYs for cost-effectiveness analysis has been presented by Fox-Rushby and Hanson 37.

Effectiveness of health interventions in a real world setting depend on a wide range of factors 1138. Four factors were identified for the present study as having an important influence on the effect of curative interventions: efficacy of individual drugs, diagnostic accuracy, appropriateness of the treatment prescribed and patient compliance.

With respect to efficacy, sources of information for this measure by type of drug were mainly a World Bank review 5, Cochrane systematic reviews (such as for instance trachoma 39) or articles identified through the Cochrane register of randomized controlled trials. Very little hard evidence from the Zimbabwean setting could be found on the other three factors so estimates of these aspects were determined for each health problem based on discussions with clinical experts. In a similar fashion as applied by Evans et al. 13, the effectiveness of a health intervention was estimated by reducing the efficacy of the relevant drug by a factor between 0 and 1. The benefits at population level in terms of DALYs averted of a specific curative health intervention j were subsequently calculated as:

D A L Y s j = E j B j F j G j ∑ a ∑ s I a s j H a s j Δ B O D a s j MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xI8qiVKYPFjYdHaVhbbf9v8qqaqFr0xc9vqFj0dXdbba91qpepeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGaemiraqKaemyqaeKaemitaWKaemywaKLaem4Cam3aaWbaaSqabeaacqWGQbGAaaGccqGH9aqpcqWGfbqrdaahaaWcbeqaaiabdQgaQbaakiabdkeacnaaCaaaleqabaGaemOAaOgaaOGaemOray0aaWbaaSqabeaacqWGQbGAaaGccqWGhbWrdaahaaWcbeqaaiabdQgaQbaakmaaqafabaWaaabuaeaacqWGjbqsdaqhaaWcbaGaemyyaeMaem4CamhabaGaemOAaOgaaOGaemisaG0aa0baaSqaaiabdggaHjabdohaZbqaaiabdQgaQbaakiabfs5aejabdkeacjabd+eapjabdseaenaaDaaaleaacqWGHbqycqWGZbWCaeaacqWGQbGAaaaabaGaem4CamhabeqdcqGHris5aaWcbaGaemyyaegabeqdcqGHris5aaaa@59F4@

where E^j, B^j, F^j and G^j are efficacy of the drug prescribed, diagnostic accuracy, correct treatment and patient compliance respectively for curative intervention j measured as percentages. Expressed in words, this equation estimates the number of individuals cured through treatment j by excluding ineffective services from the total number of individuals seeking treatment and translating the resulting health benefits into DALYs averted.

A similar procedure was applied to preventive interventions involving first determining the effect under ideal conditions followed by adjusting this to incorporate real world conditions. Efficacy of malaria spraying was derived from a study in South Africa which compared the prevalence of malaria infection in sprayed areas and non-sprayed areas 40. Similarly, efficacy estimates were derived for environmental health 414243, food supplementation 44, vaccines 4546 and STI syndromic management 4748. The number of DALYs averted at population level for a given preventive intervention j was calculated as:

D A L Y s j = E j R j ∑ a ∑ s I a s j Δ B O D a s j MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xI8qiVKYPFjYdHaVhbbf9v8qqaqFr0xc9vqFj0dXdbba91qpepeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGaemiraqKaemyqaeKaemitaWKaemywaKLaem4Cam3aaWbaaSqabeaacqWGQbGAaaGccqGH9aqpcqWGfbqrdaahaaWcbeqaaiabdQgaQbaakiabdkfasnaaCaaaleqabaGaemOAaOgaaOWaaabuaeaadaaeqbqaaiabdMeajnaaDaaaleaacqWGHbqycqWGZbWCaeaacqWGQbGAaaGccqqHuoarcqWGcbGqcqWGpbWtcqWGebardaqhaaWcbaGaemyyaeMaem4CamhabaGaemOAaOgaaaqaaiabdohaZbqab0GaeyyeIuoaaSqaaiabdggaHbqab0GaeyyeIuoaaaa@4F59@

where E^j is the efficacy of the intervention under ideal circumstances and R^j is any necessary downward adjustment (less than perfect coverage) of efficacy while Iasj MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xH8viVGI8Gi=hEeeu0xXdbba9frFj0xb9qqpG0dXdb9aspeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGaemysaK0aa0baaSqaaiabdggaHjabdohaZbqaaiabdQgaQbaaaaa@3134@ is the incidence of disease in different age- and sex groups. Coverage of the five preventive health interventions was established through discussions with the responsible staff in the four districts included or in the case of EPI utilising the Demographic and Health Survey 31.

Having estimated the total costs and effectiveness of various health interventions, the cost-effectiveness ratio for intervention j, CER^j, was found as:

C E R j = C j D A L Y s j MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xI8qiVKYPFjYdHaVhbbf9v8qqaqFr0xc9vqFj0dXdbba91qpepeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGaem4qamKaemyrauKaemOuai1aaWbaaSqabeaacqWGQbGAaaGccqGH9aqpjuaGdaWcaaqaaiabdoeadnaaCaaabeqaaiabdQgaQbaaaeaacqWGebarcqWGbbqqcqWGmbatcqWGzbqwcqWGZbWCdaahaaqabeaacqWGQbGAaaaaaaaa@3C9E@

where costs were estimated using equation (1), (2) or (3) and effects were estimated using (6) or (7).

The selection of health interventions for essential health packages may be done by applying different sets of principles. According to the World Bank principles for developing health packages 16, desirable health interventions are those with low cost-effectiveness ratios and at the same time address important health problems. Another possible set of principles is a pure cost-effectiveness criterion 49. This entails utilising a process consisting of selecting first the intervention with the lowest cost-effectiveness ratio and then calculating the total costs of averting this health problem. The subsequent step chooses the intervention with the second lowest cost-effectiveness ratio and also calculating the total costs of averting this health problem and so on until the budget is exhausted.

Assuming that the cost-effectiveness ratios estimated for the health interventions of this study complied with the assumptions of perfect divisibility and constant returns to scale 5051, the total costs and effects in terms of disease reduction of various sets of interventions could be estimated. Median cost-effectiveness ratios were utilised for each type of intervention. Estimates of the burden of disease by cause which must be addressed by the selected interventions were obtained from a previous national study 18 and reproduced in Annex 1. It was further assumed that 300 millions of Zimbabwe dollars were available for the essential packages. This corresponded to just below 10 percent of actual capital and recurrent expenditure at the national level in the financial year 1997/1998. Two additional restrictions were imposed. First, the majority of diseases could not be averted at a single level of the health system. For instance, if an intervention against pneumonia was selected to be part of the package, this health problem could not be fully avoided by offering treatment through health centres. It would be necessary to offer hospital treatment as well. Secondly, it was assumed that at most 30 percent of the HIV burden could potentially be averted through the preventive intervention included in the study (STI treatment) to avoid the budget being exhausted by this single intervention.

The sensitivity of the cost-effectiveness ratios was assessed by varying important parameters and assumptions. Instead of a 3% discount rate utilised for the baseline calculations of cost-effectiveness ratios, these were recalculated using discount rates of 6 and 10%. Estimated time preferences with respect to life years vary a lot 52 although a recent empirical study in Tanzania suggested a time preference of a similar size to the range chosen above 53. The size of the discount rate affected both the effects of health interventions through the DALY formula and the costs of health services. The long life expectancies from the chosen model life table were replaced by actual, period life expectancies as recommended in the recent World Bank health sector priorities review 12. Much shorter, actual Zimbabwean life expectancies were estimated based on the population size and the number of deaths by age and sex obtained from the census of 1997 30. Furthermore, rather than the inverted u-shape of the age weighting function suggested by the DALY methodology, an age weighting function with an equal value of 1 on each life year was used as also suggested by the World Bank health sector priorities review 12. Some of the health facilities visited operated considerably below full capacity during the study year thus pushing up costs of services. For the sensitivity analysis, cost-effectiveness ratios were recalculated under the assumption that all health facilities were moderately and significantly better utilised (e.g. 80 and 95% bed occupancy rates in inpatient departments respectively). Finally, assessing the robustness of the cost-effectiveness ratios to changes in the effectiveness of interventions was very important since this was an area with little hard evidence. Therefore, cost-effectiveness ratios were recalculated assuming a lower effectiveness of individual health interventions than in the baseline calculations. Calculations were performed utilising effectiveness estimates that were 90, 70 and 50% of the baseline estimates.

The above findings must be seen in the light of the limitations of the study. As a starting point, the cost function assumed for this study was simple and restrictive. The assumptions of constant returns to scale and perfect divisibility ensured that the average costs per unit of health of any given health intervention were the same irrespective of the level of production. These assumptions facilitated the identification of which interventions should go into an essential health package as well as the exact level of health services production necessary to eliminate the disease burden of a health problem. While there may be several reasons for deeming these assumptions unrealistic, they are frequently applied both for curative and preventive interventions 565758. Relieving the assumptions of constant returns to scale and perfect divisibility to develop cost functions with non-constant unit costs and limited possible production levels would require more complicated optimising techniques such as linear, non-linear or integer programming 5960.

Much of the information necessary for costing of health interventions was available in principle in the sense that the majority of health facilities and vertical programmes visited kept a good record of most of the resources used although only at an aggregate level (i.e. total resources for a whole hospital and not broken down by diagnosis). However, the costing exercise was nevertheless in some aspects not as detailed and extensive as could have been desired mainly as a result of limited research resources. First, a relatively limited number of health centres, hospitals and preventive activities in districts were included as study sites leading to a risk of not capturing a representative pattern of unit costs for the country as a whole. Second, the costing methodology also involved a number of simplifying assumptions in the data collection at study sites including, for instance, that the average personnel and other costs per inpatient day were the same irrespective of the diagnosis of a patient and that the treatment of most conditions followed the treatment guidelines 23 rather than the pattern found through the micro-costing data collection. Third, the number of health interventions included was relatively low due in particular to problems of identifying sufficient information to estimate effectiveness of interventions.

Perhaps the most serious limitation was the scarcity of evidence on effectiveness of health interventions and of health systems research in Zimbabwe. The observations made by several authors 146162 that we desperately lack data on these aspects are therefore also very relevant for the Zimbabwean situation. Research on factors influencing how health interventions work in the real world was extremely limited. One example of an extremely useful piece of research for the purposes of the present study was the finding of Vundule and Mharakurwa 63 that as many as 21 percent of villagers might refuse access of spraying teams to some rooms during residual house spraying. For the present study, the necessary information required to estimate effectiveness was for the majority of health interventions based on best guesses by clinical experts.

The present exercise replicated the methods underlying the 1993 World Development Report 16 and utilised also in more recent analyses 6. The Zimbabwean research proved very data intensive and most of the needed information to determine the burden of disease, costs and effects of health interventions was not immediately available which required very extensive data collection in the field. Results presented here as well as the burden of disease estimates published previously 18 represented what was feasible within the time and resources allocated. The net time of work for this research was approximately one and a half year with one researcher, a study administrator and eight research assistants working full time and a core team of ten researchers and civil servants from selected ministries working on a part-time basis (3–6 hours per individual per week on average). Despite this substantial resource input, it was deemed necessary to adopt a number of simplifying assumptions and less extensive data collection activities some of which have been discussed above.

Substantial efforts were invested in searching for published and unpublished studies on the effectiveness of health interventions as well as health systems research on how health interventions operated in reality in Zimbabwe. The lack of knowledge was a main obstacle to this study and came to some extent as a surprise to research team. If this situation had been anticipated, the data collection would have focused more on collecting additional information on the functioning of the health system which could subsequently have been utilised to adjust the findings from efficacy studies measuring impact under ideal conditions to arrive at estimates of effectiveness for the Zimbabwean situation. Efficacy studies from similar settings in Africa are more plentiful [i.e. 6465]. One example of a possible useful piece of health systems research would be a review of the health centres of this study to investigate the compliance of tuberculosis patients on directly observed treatment short course (DOTS).

The characteristics described above had as a consequence that it was possible to include only a limited number of interventions for the cost-effectiveness analysis. In other words, cost-effectiveness assessments were not performed for a whole range of relevant interventions. For instance, interventions directed at health problems with a significant burden of disease such as depression and anxiety disorders and road traffic accidents were not included. Preventive interventions were incorporated only in a limited number so that many common HIV preventive activities were left out of the analysis including condom promotion, voluntary testing and counselling, peer-based programmes to educate high risk groups and prevention of mother-to-child transmission but also preventive activities against non-communicable diseases and injuries. As a result of these shortcomings, it is not possible to conclude that we have now identified the best or the most efficient essential package of care which must be focused on for many years. There may be other more efficient packages since it is possible only able to make conclusions with regards to reallocations among the interventions investigated in this study. The present endeavour may more appropriately be seen as just one possible direction towards improving allocative efficiency in the health care sector.

Nevertheless, the present exercise was still extremely useful. It captured costs per DALY of a range of health interventions representing the situation in the health care sector at the time of study. Before this study, there was only very limited and scattered information on the costs and even less on the effects of health interventions 55666768. This study also demonstrated what was possible in a setting like Zimbabwe given a certain level of research resources but also that most of the limitations mentioned could in principle be overcome if more research resources were available. For instance, the cost data collection could have been extended and more health systems research could have been planned to inform the effectiveness component which could have enlarged the range of health interventions for the cost-effectiveness analysis. Finally, this study highlighted the most important gaps in the knowledge for priority setting i.e. the shortage of hard evidence on effectiveness of health services.

According to the health information system, acute respiratory infections, malaria and skin diseases are the most common health problems treated at outpatient department level whereas pulmonary tuberculosis (as much as 22% of all inpatient days in the country), malaria and pneumonia are the most frequent diagnoses in individuals admitted as inpatients at hospitals in 1998 19. Large proportions of pulmonary tuberculosis and pneumonia cases are probably caused by underlying HIV 18. Most of these interventions have been deemed relatively cost-effective in this study and are components of the packages suggested (Tables 2 and 3). However, the present study suggests that HIV prevention is more cost-effective than treatment (including TB treatment) which has been confirmed by other studies 69. Also malaria prevention in high incidence areas appeared to have lower costs per DALY compared to treatment. The central level of Ministry of Health controls the overall allocation of funds between hospitals, where the majority of interventions are curative, and preventive programmes. It is therefore possible to change priorities at the macro level by shifting the balance in favour of the preventive component and furthermore to focus on HIV and malaria prevention. Apart from deciding the funds available for individual hospitals, the central level of Ministry of Health has less influence on the priority setting at health centre and hospital level where the interventions offered to a large extent are determined by self-referral of patients. Rationing decisions for treating different patient groups will be done by health practitioners and may involve a variety of considerations and values (see i.e. 70) resulting in a focus which may differ from priority setting based on pure cost-effectiveness criteria. The present study pointed to a direction of focus among curative health interventions which may be difficult to enforce among health practitioners. Increased legitimacy and support among clinicians and other health sector personnel as well as patients may be secured if they are involved in the priority setting procedure through a consultative process where these groups are allowed to incorporate their own values. For instance, several authors have suggested that other criteria such as equity, severity of disease, age of patient groups and capacity to benefit may affect the rank order of health priorities in the opinion of health personnel 717273.