HEALTH POLICY AND PLANNING; 13(4): 359^370

Oxford University Press 1998

Zimbabwe's hospital referral system: does it work?

D SANDERS,1 J KRAVITZ,2 S LEWIN3 AND M MCKEE4

Health Programme, University of the Western Cape, Bellville, South Africa, 2Dept of Public Health and Preventive Medicine, Oregon Health Sciences University, Portland, Oregon, USA, 3Centre for Epidemiological Research in Southern Africa, Medical Research Council of South Africa, Tygerberg, South Africa, and 4Health Services Research Unit, Dept of Public Health and Policy, London School of Hygiene and Tropical Medicine, London, UK
Background: Anecdotal evidence has suggested inefficiency in the pyramidal health care referral system established in Zimbabwe in 1980, as part of its primary health care (PHC) model. Aim: To assess the functioning of the pyramidal referral system in two rural districts surrounding Harare, Zimbabwe, with regard to two common indicator conditions: pneumonia in children and malaria in adults. Methods: For a three-month period, all complete inpatient records with discharge diagnoses of pneumonia or malaria from three hospitals representing different levels of care were analyzed (n=227). Data were collected on demographic and patient care variables. The appropriateness of admissions and referrals was determined by an assessment of the severity of illness and `intensiveness' of care required. Data were analyzed for differences among the three hospitals and between the two indicator conditions. Per night inpatient bed costs for each hospital were also calculated. Results: For pneumonia in children, 56.8 % of patients admitted at the secondary level, 53.8 % of patients at the tertiary level and 57.8 % of patients at the quaternary level were of mild severity. For malaria in adults, 74.0 % of patients seen at the secondary level, 81.5 % of patients at the tertiary level and 54.3 % at the quaternary level were of mild severity. For pneumonia, there were no differences in severity between the three hospitals whereas for malaria significant case-mix differences among the hospitals were found. Most patients attending the highest level referral facility were inappropriate admissions who could have been treated at a lower level of care. The majority of patients at all the hospitals studied had used that hospital as their first or second point of contact with the health services. There were large variations in the inpatient per night bed costs between the three hospitals. Conclusions: Using the indicator diseases of pneumonia in children and malaria in adults, this study concluded that this network did not meet design expectations as the central level referral hospital cared for a similar case-mix of patients as the district level, but at six times the cost. The appropriateness of admissions and referrals could be improved by developing or strengthening intermediate level facilities, by changing mechanisms of access to specialist facilities and by training health professionals in community settings.

1Public

Downloaded from heapol.oxfordjournals.org by guest on June 2, 2011

Introduction
Prior to independence, millions of people on the geographic or social periphery of African countries received either marginal health care or none at all. Colonial health systems channelled resources to urban dwellers and white settlers at the expense of the predominantly black rural populations who had the greatest need (Aidoo 1982; Tudor-Hart 1971). When Zimbabwe gained independence in 1980, 44 % of public funds for health services went to urban central hospitals serving 15 % of the popula-

tion, while only 24 % of funds were dedicated to rural areas where 77 % of the population lived. The promise of `health services for all' mobilized political support for post-colonial governments (Walt 1990). Less developed countries such as Zimbabwe instituted primary health care (PHC) systems, and developed a pyramidal referral model to support the primary care level. Clinics and district hospitals were intended to provide local services for uncomplicated cases, referring patients with more serious conditions to regional/provincial and central hospitals.

360

D Sanders et al. 4. To evaluate and compare the influences of demographic and patient care variables (distance travelled to hospital, case-mix) on the utilization and costs of inpatient services at different hospital levels. 5. To assess the implications of these findings for referral policies in Zimbabwe and internationally. Although the data reported here were gathered in 1988, both the methods and findings still have important implications for service planning and development. Setting of the study More than one million people live in rural Mashonaland West Province, located in the north central plateau region of Zimbabwe (Figure 1). Karoyi district, population 250 000, and Chinhoyi district, population 280 000, are located here. Harare, the capital, population 1 000 000, is located 175 and 100 kilometres (km) from Karoyi and Chinhoyi towns, respectively (Zim Epi Bul 1988). Karoyi District Hospital (KDH) had 150 beds and at the time of the study was staffed by two medical doctors. Chinhoyi Provincial Hospital (CPH) had 250

Ingrained political and economic structures have, however, retarded full implementation of the pyramidal referral system in many countries (Sanders and Davies 1988; Loewenson et al. 1991; Loewenson 1993). As a result, peripheral facilities are often seen as providing an inadequate standard of service, while central hospitals are frequently overloaded and inefficient (WHO 1992). More seriously, significant segments of the population, especially in rural areas and peri-urban informal settlements, still remain at the periphery of the health services umbrella (Bloom 1985; Loewenson et al. 1991; Hirschowitz et al. 1995; Project for Statistics on Living Standards and Development 1994). In Zimbabwe's hospital referral plan, district (secondary level) hospitals were to provide general inpatient services, accepting referrals from urban and rural health centres (RHCs) and clinics (primary level). Provincial (tertiary level) hospitals were to receive patients referred from district (secondary level) hospitals and provide general specialist services (C.S.O. 1984). Quaternary level hospitals in the major urban centres were to serve as national referral facilities and provide specialist and subspecialist services. Pragmatic direct primary level referrals of less seriously ill patients were to be phased out (Planning 1983). Anecdotal evidence, however, has suggested inequity and inefficiency in the pyramidal health care model. Goal and objectives The goal of this study was to assess the functioning of the pyramidal referral system in two rural districts of Mashonaland West Province and in Harare, Zimbabwe, and to measure the extent to which the referral system operated as planned. The objectives of the study were: 1. To develop a demographic profile (age, gender, place of residence and distance to the health facility) of patients utilizing hospitals at different levels of care for two common indicator conditions: pneumonia in children and malaria in adults. 2. To ascertain the appropriateness of hospital utilization at different levels of care for these two indicator conditions. 3. To identify differences in case-mix, length of stay, referral patterns, use of laboratory and diagnostic tests, and clinical outcomes for the indicator conditions at different hospital levels.

Downloaded from heapol.oxfordjournals.org by guest on June 2, 2011

Figure 1. Zimbabwe: The study area and geographic relationship to neighbouring countries

Zimbabwe's hospital referral system beds and was staffed by five medical doctors. Parirenyatwa Central Hospital (PARI) in Harare, with 900 beds, functions as a teaching hospital and is seen as the referral centre for patients from secondary and tertiary hospitals in the region. It offers specialist and general outpatient and inpatient services. At the time of this study, Harare did not have a hospital intermediate between the primary level and PARI to which patients could have been admitted. Indicator diseases Paediatric pneumonia or acute lower respiratory infection (ALRI) and adult malaria were used as indicators because of their high incidence (Stansfield 1987; Zim Epi Bul 1988; Tembo 1988; Harare Health Dept. Bul 1987; Collings et al. 1985; Ruebush et al. 1986; Taylor et al. 1986) and because inpatient curative care is an effective form of management. The two diseases also fulfil the criteria for `tracer' conditions developed by Kessner et al. (1973). For the purposes of the study, they were therefore seen as useful indicators of health service utilization patterns.

361

Illness severity was derived from a severity index (SI). Both the ALRI and malaria severity index scores comprised two components. The first part recorded clinical findings on admission. The second part described clinical progress. The severity index therefore captured information on each patient's conditions both on admission and during treatment. Combining the two parts of the severity index gave a sum of 2 to 6 upon which illness severity was estimated: 2 (mild); 3^4 (moderate); and 5^6 (severe) (Tables 2 and 3). The appropriateness of the admission was derived from demographic and patient care variables. For utilization, a subjective assessment was made by the principal researcher regarding the appropriateness of each admission. For example, a malaria patient with a serious severity index, exhibiting chloroquineresistance and evidence of renal failure, who remained an inpatient for seven days, would have been ``appropriately'' treated at PARI. Utilization at any other level was ``inappropriate''. Other patients might have been most appropriately treated in a primary care setting. Referrals were also assessed as to whether low- or high-intensive care was required. ``Intensiveness'' referred to the level of diagnostic and management skills. ``Appropriate referral'' was judged by whether a patient requiring a particular ``intensiveness'' of care was referred to a hospital where this could best be given. Non-referrals were also assessed as to whether lowor high-intensive care was required. ``Appropriate non-referral'' implied that a decision to provide definitive care at the initial admission hospital was appropriate. The financial information available for the hospitals studied was very rudimentary. Recurrent costs and inpatient census data for the fiscal year 1985/86 were used to calculate per night bed costs in the hospitals studied. Ministry of Health (MOH) allocations did not separate inpatient and outpatient services, precluding disaggregation of costs. As outpatient services accounted for an estimated 10 % of total expenditures, this amount was subtracted from the gross annual recurrent costs to give the estimated annual recurrent inpatient expenditures. Additionally, expenses for country-wide services and training provided by PARI were not designated as separate allocations by the MOH. Costs of inpatient services

Downloaded from heapol.oxfordjournals.org by guest on June 2, 2011

Methods
Many hospital utilization study methods require the prospective collection of non-routine data. This is not always feasible logistically or financially in developing country settings. This study used retrospective examination of patient records to ascertain the patterns and appropriateness of hospital utilization at different levels of care. This study was conducted during August and September 1988. Inpatient records were scrutinized onsite at KDH, CPH, and PARI. Each hospital represented one level of the pyramidal design. All completed records with discharge diagnoses of ALRI or malaria between January and March 1986 were analyzed. Five hundred records were reviewed of which 227 were included in the study. The other 263 records were excluded on the basis of being incomplete or having a misclassified diagnosis. To avoid unpredictable variations in the extraction of medical information by multiple examiners (Cartwright 1987), one author collected all data. Data were collected on the variables outlined in Table 1. Information on income, education, and chronic illness was not available (Table 1).

Downloaded from heapol.oxfordjournals.org by guest on June 2, 2011

362

D Sanders et al.

Zimbabwe's hospital referral system and salaries at PARI were therefore reduced by an additional 10 %, reflecting the estimated financial burden associated with provision of outside services. Although the estimates derived in this way are crude, they provide a basic guide to the differential costs of beds across the hospitals studied. Analysis Chi-square analysis and the Fisher exact test were used to test for differences among the three hospitals for discrete variables. For continuous variables Fstatistics for analysis of variance were applied. Analysis of covariance was used to control for theoretically relevant variables. The statistical software program SPSSPC+ was used to analyze the data. Unless otherwise shown in tables or text, p-values shoul| d be assumed to be 40.05.

363

but there was no such association for ALRI. Most patients, with either condition and at any of the hospitals, were classified as mild and, for ALRI, the case-mix treated in each hospital was similar. For adult malaria, however, PARI did treat more severe patients. In contrast, patients with ALRI were kept in hospital for significantly longer at PARI and were investigated more intensively. The same pattern was seen for adult malaria although, as noted, the case-mix did differ. The longer stays in PARI have particularly important implications as, at the time of the study, the cost of an inpatient bed was $Z94 in PARI, $Z33 at provincial level, and Z$14 at district level (1Z$=$US0.59 at this time). For both conditions, patients were much more likely to have been referred and admitted inappropriately to PARI. For ALRI, the case fatality rate between the three hospitals was not statistically different For malaria there were statistically significant differences between the three hospitals regarding the number of patients discharged alive versus the number dying in hospital (p50.025, Fisher). Any inferences concerning mortality should, however, be judged cautiously because the numbers of patients and fatalities were small and because of differences in case-mix between the hospitals for malaria cases.

Results
For both malaria and ALRI, patients treated in each hospital were similar in terms of age and gender, except for patients treated at PARI where there was a male excess due to large numbers of soldiers referred from Mozambique. ALRI patients' ages ranged from two weeks to 14 years, with the majority of patients aged from seven months to five years. Patients admitted with malaria ranged in age from 15 to 70 years. Data on distance travelled are summarized in Table 4. Patients treated for malaria at CPH were less likely to have travelled more than 10km to hospital than those treated at either KDH or PARI (25 %, 73 % and 66 % respectively, p=0.0001). When patients are divided into commercial and non-commercial farm residents, it is clear that KDH and CPH draw significantly larger numbers of patients from surrounding commercial farming areas. Only 9 % of PARI patients came from a commercial farm. The distances travelled to hospital by these commercial farm residents were significantly greater than for non-commercial farm residents. The characteristics of patients treated in each hospital are shown in Tables 5 and 6. At each hospital, most patients self-referred and for 81 % of malaria and ALRI patients attending CPH or KDH, that hospital was their first contact with the health system. The corresponding figure for PARI, 58 %, was somewhat lower than that of the other two facilities, but still somewhat high for what is intended to be a quaternary referral centre. At each hospital, those patients treated for malaria who had travelled further were more likely to have had severe disease,

Downloaded from heapol.oxfordjournals.org by guest on June 2, 2011

Discussion
This study attempted to assess the effectiveness of the planned referral system for health care in areas north of Harare, Zimbabwe. It examined the functioning of the system, the influence of demographic and patient care variables on utilization, and identified differences in patterns of care and outcome at different hospital levels for two indicator conditions. The data demonstrate that the study hospitals, at different levels of care, are seeing similar mixes of patients. For children with ALRI, the admissions at all three hospital levels were similar with respect to illness severity, outcome, age, gender and distance travelled for inpatient care. Despite these similarities in ALRI admissions, significant differences were noted in measures of resource utilization, such as length of stay and use of diagnostic and laboratory tests. For malaria, the cases admitted at KDH and CPH were similar with regard to severity. At PARI a significantly larger group of patients (14 %) was rated as seriously ill when compared to the other

364

D Sanders et al.

Downloaded from heapol.oxfordjournals.org by guest on June 2, 2011

two hospitals studied, and required highly intensive treatment. Nevertheless, 60 % of PARI malaria patients were `inappropriate' users. KDH and PARI saw significantly larger proportions of patients who had travelled more than 10km to hospital than did CPH. This may indicate that CPH is not functioning as the tertiary referral unit for KDH patients as such referrals would have increased the number of `distance' patients at CPH. This assumption is supported by the finding that CPH admitted the largest proportion of mild malaria patients (82 %) of the three hospitals studied. Viewed overall, the results

seem to indicate that the referral system for malaria patients is not operating as intended. A number of factors were identified which may explain the malfunctioning of the referral system in the Harare area. Firstly, patients will usually utilize the hospital that is nearest to them, particularly if alternative, and more appropriate, sources of care are not available, as is the case in Harare and in commercial farming areas. For all referral hospitals studied, a large proportion (42 %) of inpatients came from within 10km of the hospital, with only one-

Zimbabwe's hospital referral system

365

Downloaded from heapol.oxfordjournals.org by guest on June 2, 2011

third residing more than 30km from the hospital attended. This inverse relationship between hospital utilization and distance, the ``distance decay curve'', is well described (Stock 1983). Studies of inpatients and referred and unreferred outpatients attending a superspecialist paediatric hospital in South Africa have also shown that only small numbers of patients came from outside the metropolitan area (Henley et al. 1991; Lachman 1988; Power et al. 1996). Although only a small proportion of patients had travelled long distances to the hospital, distance travelled was associated with illness severity for malaria, suggesting that very ill malaria patients are more likely, more able or have to travel long distances when seeking care. This was not the case for childhood pneumonia, possibly because the very ill children died before reaching a hospital, because the pneumonia was successfully treated at the primary level or because adults are given a higher `social weight' than children, with greater resources therefore being made available for them to seek distant health care. Another factor that might explain the malfunctioning of the referral system is that expensive quaternary services were used rather than lower levels of care.

For 58 % of PARI patients, the central hospital was the first point of contact with the health care system, with many seeking initial care at the PARI Casualty Department. Of the 38 % of patients referred directly from the primary level to PARI, 83 % came from local city health department clinics or private general practitioners. Of the referrals to PARI, more than half were inappropriate for both pneumonia and malaria, and substantial proportions of patients admitted should have been referred to a lower level of care. Most admissions did not, therefore, require the expensive, more intensive care provided at the central hospital but, as mentioned earlier, Harare did not have a hospital intermediate between the primary care and quaternary levels to which these patients could have been admitted. Appropriate accessibility, rather than accessibility per se was therefore an important issue, as has been demonstrated elsewhere (Henley et al. 1991; Criel et al. 1997). This situation will have to be addressed if the pattern of hospital admissions is to be changed (Barnum et al. 1993). Finally, limited resource availability was a major impediment to appropriate utilization and referral patterns. During the study it was noted that staffing

366

D Sanders et al. may be explained in part by the fact that PARI is a teaching hospital of the University of Zimbabwe School of Medicine. The teaching environment undoubtedly promotes greater use of diagnostic services (Schroeder et al. 1977). It is unlikely, however, that differences in length of stay between PARI and the two lower level hospitals can be explained by the retention of patients for teaching purposes, as most patients had mild forms of common illnesses. The increased length of stay at PARI may therefore indicate problems with discharge planning at that hospital. With a similar case-mix of ALRI patients at all three hospital levels, central care appears to have simply meant greater expense for illnesses that could have been treated at a lower level, where such care is available. The limitations of studying only two diseases are acknowledged. The utilization and referral patterns in the surgical specialities, for example, may have different characteristics from those in medicine and paediatrics. These results should, therefore, only be generalized to these other specialities with caution. Despite these limitations, retrospective record reviews combined with clinician assessments can be valuable and low-cost methods of ascertaining and monitoring the appropriateness of hospital utilization. The methodology could be readily applied by hospital managers and service planners to monitor the appropriateness of care on a routine basis. In summary, this study shows that the utilization and referral patterns of the pyramidal PHC system for ALRI and malaria in the districts surrounding Harare did not conform to design expectations in most cases. This reality of facilities simultaneously functioning as general, specialist and super-specialist hospitals is well described elsewhere. A study of the Hillbrow academic hospital in Johannesburg, South Africa, showed that it provided services at several different levels of care (Schneider et al. 1992). A similar study at an academic hospital in Cape Town showed that 54 % of patient-days could appropriately have been spent at a lower level of care (Bachmann et al. 1991). Other studies in the Southern African region have also shown that both in- and out-patients utilize specialist hospitals inappropriately (Zwarenstein et al. 1990; Henley et al. 1991; Rutkove et al. 1990; Power et al. 1997). The inappropriate utilization of referral facilities will remain a problem until quality accessible (and affordable) primary and secondary level care is available.

shortages, especially with respect to doctors, existed at district and provincial levels. In particular, the absence of specialist services at the provincial level rendered it less effective as a functional unit in the referral system and may have led to the perception that advanced care would only be provided centrally. This view is supported by the finding that only 25 % (n=7) of malaria patients attending CPH resided more than 10km from the hospital. This may indicate that referrals from KDH were sent directly to PARI, based on the perception that it offered services not available at the provincial referral hospital in Chinhoyi. The study has several limitations. Firstly, the Severity Index used may have led to misclassification of illness severity. Ascertainment bias of ALRI may also have resulted at all levels of care, especially at KDH and CPH, where CXR confirmation was 2 % and 35 %, respectively. The Severity Index for pneumonia was, however, based on respiratory rate which is generally accepted as a reliable indicator of ALRI (Shann et al. 1984), as sufficient reason to treat, and has been shown to be the best clinical sign predicting illness severity in ALRI (Cherian et al. 1988). Nevertheless, no single clinical finding may be a foolproof predictor of ALRI (Diehr et al. 1984). It should be noted that this study was performed before publication of the currently accepted World Health Organization classification of ALRI, which includes chest retractions. Although bias may have been introduced by only including completed patient charts (n=227), those patients with incomplete records (n=263) were unlikely to have been managed differently. The assessment of the appropriateness variables was also not blinded to the level of care at which the patient had been treated, potentially introducing bias into the assessment. Reliance on anaemia to categorize malaria morbidity should be tempered by an awareness of important confounding factors including malnutrition, parasitic infestation, and blood dyscrasias. The diagnosis of `clinical malaria' without laboratory confirmation was common in the hospitals under study and this may have resulted in some degree of over-diagnosis of mild patients and under-diagnosis of severe patients (Todd 1987), especially at KDH and CPH. Differences in the use of laboratory and diagnostic tests between PARI and the two lower level hospitals

Downloaded from heapol.oxfordjournals.org by guest on June 2, 2011

Zimbabwe's hospital referral system Notwithstanding the similar case-mix and comparable outcomes for patients managed at different levels, recurrent average costs incurred in their management are extremely different, as described in the international literature in the 1970s (Gish 1977) and more recently (Barnum et al. 1993). Justification for dedicating disproportionately high budget allocations to the central (quaternary) hospital level has in the past been based on the argument that these higher level facilities act as referral and training units, managing sicker patients and teaching students. By selecting two common and important indicator diseases and creating illness severity and `appropriateness' variables for them, this study has been able to demonstrate that, for the conditions studied and hospitals chosen, similar patients were being cared for at different levels. Significant cost savings and improved appropriateness of utilization might, therefore, be realized if patients in Harare had access to an acceptable intermediate facility with substantially lower overhead costs. The possibility of designating Harare Central Hospital ^ another `quaternary' facility within Harare ^ as an intermediate level hospital should be considered. Alternatively, a `functional split' could be created within PARI, such that different levels of care, resourced at different levels, could be delivered within one institution. The latter, while possible, is by no means easy and requires a clear distinction of the roles of the different care levels and the establishment of mechanisms for ensuring that the levels are appropriately resourced and that patients are efficiently referred between them. Creative ideas developed elsewhere to address the challenge of primary care patients attending referral hospitals should be examined for their applicability to the Zimbabwean setting (Dale et al. 1996). It should also be noted that the differences in the marginal costs of inpatient care at different levels are unlikely to be as large as the differences in average costs reported here. Cost savings from shifting patients to an appropriate level of care may therefore be less substantial than expected, but are still likely to be significant. In contrast, the costs both to the health services and to patients, of the maldistribution of resources across levels of care, are large.

367

(1977) has suggested that `Health service development must be based on the realistic possibilities for the referral of patients rather than merely the abstract desirability of such referral ... Lower level institutions must be strengthened to be more capable of providing adequate services in the absence of extensive referral possibilities until communications improve to justify a reassessment'. Other authors have noted that insufficient staffing and lack of supplies with resulting poor service quality may impede the efficient delivery of health care to patients (Annis 1981; Barnum et al. 1993; Bijlmakers et al. 1996). The essential message of these observations would appear to be valid for Zimbabwe at the time of this study and for many other developing countries today. Compared with the experience of many developing countries, Zimbabwe has made great strides in the development of a health care system based on the PHC model (Sanders and Davies 1988). However, an efficient and equitable health care system remains elusive, in part because of inadequate support of the primary and intermediate levels of the health care system (Bijlmakers et al. 1996). A number of recommendations emerge from this study. However, it is recognized that, since the study was conducted, some of the recommendations may have been partially or completely implemented. 1. Peripheral facilities need to be improved and supported in order to fulfil their role in the referral pyramid and to enhance rational referral. Others have also alluded to the relative lack of peripheral facilities, including district hospitals, in subSaharan Africa (Van Lerberghe et al. 1997) ^ Zimbabwe is no exception, particularly in commercial farming areas. Improvement in provincial level services in Zimbabwe may lead to a decline in the number of inappropriate central referrals from the district level, and reduce cost pressures at the central level. A significant reduction in transportation costs for patients and their families would also be realized. Where national referral centres exist, consideration should be given to developing good quality intermediate level facilities nearby or to creating `functional splits' within referral hospitals so as to formally allow different levels of care to be delivered within the same institution. This may go some way to reducing the number of inappropriate primary referrals to the specialist or subspecialist levels and to redistributing resources to underserved areas (WHO 1992).

Downloaded from heapol.oxfordjournals.org by guest on June 2, 2011

Conclusions
Pragmatism has been urged in the development of health services in less developed countries. Gish

368

D Sanders et al.
Bloom G. 1985. Two Models for Change in Health Services in Zimbabwe. Int J Health Services 15(3): 451^68. Cartwright A. 1987. Hospital Maternity Records. Community Med 9(3): 293. C.S.O. (Central Statistics Office). 1984. 1982 Population Census: A Preliminary Assessment. Harare, Zimbabwe. Cherian T, Simoes E, John T, Steinhoff M, John M. 1988. Evaluation of Simple Clinical Signs For the Diagnosis of Acute Lower Respiratory Tract Infection. Lancet 8603(2): 125^8. Collings D and Martin K. 1985. A Retrospective Analysis of Childhood Pneumonia in a District Hospital. Central African Journal of Medicine 31(8): 152^6. Criel B, Pariyo GW. 1997. Optimisation and Integration of Different Levels of Care in Hoima District (Western Uganda). Bamako Initiative Operations Research Programme. Research Paper No. 5. Unicef. New York. Dale J, Lang H, Roberts JA, Green J, Glucksman E. 1996. Cost effectiveness of treating primary care patients in accident and emergency: a comparison between general practitioners, senior house officers and registrars. BMJ 312(7042): 1340^4. Davenport RJ, Dennis MS, Warlow CP. 1996. Effect of correcting outcome data for case mix: an example from stroke medicine. BMJ 312: 1503^5. Diehr P, Wood R, Bushyhead J, Krueger L, Wolcott B, Tompkins R. 1984. Prediction of Pneumonia in Outpatients with Acute Cough ^ A Statistical Approach. Journal of Chronic Disease 37(3): 215^25. Gish O. 1977. Guidelines For Health Planners. The Planning and Management of Health Services in Developing Countries. London: Tri-Med Books; 39^40. Grimshaw JM, Russel IT. 1993. Effect of clinical guidelines on medical practice: a systematic review of rigorous evaluations. Lancet 342: 1317^22. Haddon B. Personal Communication. Provincial Health Services Administrator, Mashonaland West Province. Chinhoyi, Zimbabwe; 1988, August and September. Harare Health Department Bulletin. 1987. Zimbabwe Ministry of Health. Harare. May 26:2. Henley L, Smit M, Roux P, Zwarenstein M. 1991. Bed use in the medical wards of Red Cross War Memorial Children's Hospital, Cape Town. S Afr Med J 80: 487^90. Hirschowitz: R, Orkin M. 1995. A National Household Survey of Health Inequalities in South Africa. Overview Report. CASE Survey for the Henry J Kaiser Foundation. Kessner DM, Kalk CE. 1973. A strategy for evaluating health services. Contrasts in health status. Volume 2. Institute of Medicine. National Academy of Sciences. Washington DC. Lachman P. 1988. Referral Patterns to the Red Cross War Memorial Children's Hospital. Unpublished MMed dissertation. Department of Paediatrics and Child Health. University of Cape Town. South Africa. Loewenson R. 1988. Challenges to Health in Plantation Economies: Recent Trends. Health Policy and Planning 4(4): 334^42. Loewenson R. 1993. Structural Adjustment and Health Policy in Africa. Int J Health Services 23(4): 717^30. Loewenson R, Sanders D, Davies R. 1991. Challenges to Equity in Health and Health Care: A Zimbabwean Case Study. Soc Sci Med. 32(10): 1079^88. Paul V. 1993. Innovative programmes of medical education: II. Commentary and lessons for India. Indian Journal of Pediatrics 60(6): 769^76. Pearson C. 1995. The role of district hospitals and the action in international medicine network. Infectious Disease Clinics of North America 9(2): 391^405.

2. Attention needs to be given to changing mechanisms of access to specialist facilities. There is good evidence from both developed and less developed countries that restricting access to specialist units can increase the appropriateness of the mix of patients seen in these units and, in some instances, reduce specialist level patient loads (Davenport et al. 1996; Stoops 1996). Enforcing referral guidelines and educating health personnel in their use would encourage compliance with the pyramidal referral design. However, any system of guidelines must take into account the growing literature on how they can be implemented effectively (Grimshaw et al. 1993). Such referral policies are being implemented in South Africa and elsewhere in the region (Sutcliffe 1996), and it will be important to assess their effect on attendance patterns at different levels of care. Clearly, such policies, while potentially useful in the Zimbabwean context, need to be accompanied by strategies to provide adequate levels of capital and recurrent resources at each level of care. 3. The training of medical students and other health professionals in peripheral facilities, where the reality of resource limitations governs patient care, should be further encouraged in less developed countries (Paul 1993; Pearson 1995; Criel et al. 1997). This could lead to a greater appreciation of medical practice outside of a subspecialist academic environment. An additional benefit would be enhanced personnel to support understaffed peripheral hospitals. This training should highlight simple diagnostic and therapeutic approaches to case management.

Downloaded from heapol.oxfordjournals.org by guest on June 2, 2011

References
Aidoo T. 1982. Rural Health Under Colonialism and Neocolonialism: A Survey of the Ghanaian Experience. Int J Health Services 12(4): 637^57. Annis S. 1981. Physical Access and Utilization of Health Services in Rural Guatemala. Soc Sci Med (D) 15(4): 515^23. Bachmann OM, Zwarenstein MF, Benatar SR, Blignaut R. 1991. Levels of care needed by medical inpatients in a teaching hospital. S Afr Med J 80: 477^80. Barnum H, Kutzin J. 1993. Public Hospitals in Developing Countries; Resource Use, Cost, Financing. Published for the World Bank by the Johns Hopkins University Press. Baltimore. Bijlmakers L, Chihanga S. 1996. District Health Systems in Zimbabwe: Cost, Resource Adequacy, Efficiency and the Core Service Package. A comparison of Three Districts. Bamako Initiative Technical Report Series Number 36. Unicef. New York.

Zimbabwe's hospital referral system

Planning for Equity in Health. 1983. Government of Zimbabwe. Ministry of Health. Harare. Power M, Els R, Zwarenstein M, Lewin S, Vundule C, Mostert J. 1997. Most patients attending a ``walk-in'' clinic at the Red Cross Children's War Memorial Hospital could be safely managed at primary care level. S Afr Med J 87(1): 36^41. Ruebush T, Kaiser R, Breman J, Warren M. 1986. Selective Primary Health Care. Malaria. Reviews of Infectious Disease 8(3): 454^66. Rutkove SB, Abdool Karim SS, Loening WEK. 1990. Patterns of care in an overburdened tertiary hospital outpatient department. S Afr Med J 77: 476^8. Sanders D and Davies R. 1988. The Economy, The Health Sector, and Child Health in Zimbabwe Since Independence. Soc Sci Med 27(7): 723^31. Sanders D, Loewenson R, McKee M, Kravitz J. 1989, August. Utilization of different levels of health care and the operation of the referral system in Zimbabwe. Abstract. International Epidemiological Association. African Regional Conference; Harare; Zimbabwe. Schneider H, Broekmann R. 1992, August. A pilot study of levels of care provided at Hillbrow Hospital, Johannesburg. The Centre for Health Policy, University of the Witwatersrand, Johannesburg. Schroeder S, O'Leary D. 1977. Differences in laboratory use and length of stay between university and community hospitals. Journal of Medical Education 52: 418^20. Shann F, Hart K, Thomas D. 1984. Acute Lower Respiratory Infections in Children: Possible Criteria for Selection of Patients For Antibiotic Therapy and Hospital Admission. Bulletin of the World Health Organization 62: 749^53. Project for Statistics on Living Standards and Development. South Africans Rich and Poor: Baseline Household Statistics. 1994. SALDRU. School of Economics. University of Cape Town. Stansfield S. 1987. Acute Respiratory Infections in the Developing World: Strategies for Prevention, Treatment, and Control. Paediatric Infectious Disease Journal 6: 622^9. Stock R. 1983. Distance and the Utilization of Health Facilities in Rural Nigeria. Soc Sci Med 17(9): 563^70. Stoops N. Chief Professional Nurse. Health Information Department. Metropolitan region of Provincial Administration of the Western Cape, South Africa. Personal communication. November 1996. Sutcliffe TJ. 1996. Free Health Services. Letter to all Heads of Institutions from the Provincial Administration of the Western Cape Department of Health, 29 March. Targett G. 1991. Malaria: Where are we now? Health Policy and Planning 6(1): 100^4. Taylor P and Mutambu S. 1986. A Review of the Malaria Situation in Zimbabwe with Special Reference to the Period 1972^ 1981. Trans R Soc Trop Med Hyg 80: 12^19. Tembo T. 1988. Office of Statistics; Chinhoyi Provincial Hospital, Zimbabwe. Todd C. 1987. Mazowe District Health Services-Annual Report. Mashonaland Central Province, Zimbabwe. Tudor Hart J. 1971. The Inverse Care Law. Lancet 7696(1): 405^ 12. Van Lerberghe W, de Bethune X, De Brouwere V. 1997. Hospitals in sub-Saharan Africa: why we need more of what does not work as it should. Trop Med Int Health 2(8): 799^808. Walt G. 1990. Community Health Workers in national programmes: just another pair of hands? Oxford University Press. WHO Study Group on Primary Health Care in Urban Areas. 1992. The role of health centres in the development of urban

369

health systems: report of a WHO Study Group on Primary Health Care in Urban Areas. WHO Technical Report Series 827. Zwarenstein MF, Benatar SR, Shaw LR. 1990. The pattern of care in the medical wards of a teaching hospital. S Afr Med J 77: 460^3.

Acknowledgements
This study was undertaken through joint cooperation of the Department of Community Medicine, University of Zimbabwe Medical School, Harare, Zimbabwe; and the University of Washington, School of Public Health and Community Medicine, Seattle, Washington, USA. We would also like to acknowledge the support of the South African Medical Research Council in the writing, editing and review of this manuscript. The authors wish to express sincere appreciation to Dr Campbell Katito, Provincial Medical Director of Mashonaland West Province, for granting permission to undertake the study; to members of the Mashonaland West medical and hospital communities for their participation; to Mr Bryan Haddon, Dr Rene Loewenson, Ms Frances Chinemana, Dr Merrick Zwarenstein, Dr Arthur Heywood, Prof Anne Mills, Dr Steven Gloyd, Dr Debbie Bradshaw, Dr Jane Doherty, and Dr Richard Froese for their support and useful comments at various stages of the study. Special thanks to Dr Michael Dunn for his assistance in data analysis.

Downloaded from heapol.oxfordjournals.org by guest on June 2, 2011

Biographies
David Sanders qualified as a medical doctor in Zimbabwe and later specialised in Britain in Paediatrics and Tropical Public Health. In 1980 he returned to newly independent Zimbabwe as Coordinator of a rural health programme developed by OXFAM in association with the Zimbabwe Ministry of Health. He later joined the University of Zimbabwe Medical School, initially as a lecturer in Paediatrics and Child Health and later became Associate Professor and Chairperson of the Department of Community Medicine. In 1993 he was appointed as Professor and Director of a new Public Health Programme at the University of the Western Cape, Cape Town, South Africa, which provides practice-oriented education and training in public health and primary health care to a wide range of health and development workers. He is the author of two books: `The Struggle for Health: Medicine and the Politics of Underdevelopment' and `Questioning the Solution: the Politics of Primary Health Care and Child Survival' as well as several booklets and articles on the political economy of health, structural adjustment, child nutrition and health personnel education. Jay Kravitz earned a BA in Economics from Tulane University in 1968 and an MD in 1972. He earned an MPH from the University of Washington, Department of Health Services, International Health Program in 1989. He is board certified in Emergency Medicine. He has participated in refugee aid projects in Sa Keow, Thailand, 1979 and in Lalibella and the Ansokia Valley, Ethiopia, 1985. He recently served as Director of the Infectious Disease Survey and Assistant Project Manager of the Lesotho Highlands Health Survey, Maseru. Dr Kravitz is an assistant professor in Emergency Medicine and General Preventive Medicine and is Director of the Preventive Medicine Residency Program at Oregon Health Sciences University. Simon Lewin trained in medicine in Cape Town and has an MSc in Public Health for Developing Countries from the London

370

D Sanders et al.
worked at the London School of Hygiene and Tropical Medicine since 1987, where he is now Professor of European Public Health. He manages a large research programme addressing the challenges to health and health services in central and eastern Europe and the former Soviet Union. Correspondence: David Sanders, Public Health Programme, University of the Western Cape, Private Bag X17, Bellville 7535, South Africa.

School of Hygiene and Tropical Medicine, UK. He currently works as a researcher in the Centre for Epidemiological Research of the Medical Research Council of South Africa, and has also worked in the Public Health Programme, University of the Western Cape and the Health Policy Unit, London School of Hygiene and Tropical Medicine. Martin McKee qualified in medicine in Belfast and trained in internal medicine and subsequently public health. He has

Downloaded from heapol.oxfordjournals.org by guest on June 2, 2011