Hospital costs of high-burden diseases: malaria and pulmonary tuberculosis in a high HIV prevalence context in Zimbabwe
Micro-costing is a concept based on principles of industrial engineering science, which use input-based methods of measuring resource use. The approach involves direct enumeration and costing of inputs consumed in the treatment of a particular patient. Input costs are then summed to obtain a total cost, which can be converted, to desired units of interest—for example, cost per case or cost per day.
Eisenberg et al. (1984) define time and motion analysis as involving the observation of workers carrying out their usual activities and recording the time consumed during each step of the process.
The panel consisted of three public health physicians and two senior nursing officers with considerable experience in the health care delivery system in Zimbabwe, the Provincial Medical Director for Mashonaland East province, the Medical Officer of Health (Maternal and Child Health), the Medical Officer of Health (Disease Control and Epidemiology), the Provincial Nursing Officer, and the Provincial Community Nursing Officer.
Government Medical Stores (GMS) is the central procurement and distribution unit for drugs and medical supplies for public facilities. Private sector suppliers are only used for products out of stock at GMS. In fact, GMS approves private procurement.
The only possible source of differences between hospitals was in staff costs. In other words, the same test conducted by a laboratory technologist cost more compared to one carried out by a junior technician.
At this level of analysis (pooled), malaria severity was not independent of referral status (F test, P = 0.02).
Questions about the efficacy of malaria microscopy tests, and whether the patients actually had malaria or just parasitaemia are not relevant in this study. All cases that were clinically categorised by health staff as having malaria were considered as such and treated accordingly.
According to an Independent Samples t-test.
Hospital 2 had generally the least staff-to-patient day ratios: 37:1 for doctors, 2.7:1 for nurses and 55:1 for pharmacy technicians compared to the respect average of 43:1 for doctors, 3:1 for nurses and 186:1 for pharmacy technicians.
To reduce congestion in hospitals, and to contain hospital costs, TB patients are kept in hospital for a shorter period of time (1–2 weeks) during the intensive phase of treatment rather than the conventional 2 months in hospital. Once patients are stabilised, they are discharged for continued home treatment. What is important is that someone close to the patient (relative or Village Health Worker) is trained to directly observe the patient taking medication at prescribed times. Direct observation improves treatment compliance and treatment success rate (MOH 1998).
Such patients are classified as Category 1 patients according to the National TB Treatment Guideline. These are cases with pulmonary TB, admitted for the first time with TB and with no evidence of resistance to TB drugs (MoH 1994).
Principal reason for admission.
This paper explores the measurement of hospital costs and efficiency in a context where data is scarce, incomplete or of poor quality. It argues that there is scope for using tracers to examine and compare hospital cost structures and relative efficiency in such contexts. Two high-burden diseases, malaria and pulmonary tuberculosis, are used as tracers to calculate the average costs of inpatient care at selected tertiary hospitals. This study shows that it is feasible to prospectively collect cost data for specific diseases and explore in detail both patient cost distribution and susceptible areas for efficiency improvement. The present study found that the critical source of efficiency variation in public hospitals in Zimbabwe lies in the way hospital beds are used.
Analysis of hospital costs poses some challenges in contexts where data is scarce, aggregated or of poor quality. Many studies on hospital costs in such contexts have tended to use crude methods that do not provide insights into service efficiency. Case mix and case severity are rarely considered, which has major implications for the interpretation of costs and relative efficiency. Countries with a high HIV prevalence rate such as Zimbabwe (25% among adults) (UNAIDS 1999) have seen an upsurge in tuberculosis (TB) cases over the years with considerable impact on hospital case mix. By 1998, TB was one of the major causes of morbidity and mortality in the country, with 90% of the admitted TB cases estimated to be HIV positive (MoH 1998). Malaria continues to be one of the top causes of morbidity and mortality. Consideration of case mix and case severity is vital in understanding the hospital cost structure and relative cost differences.
Accounting methods are commonly used for hospital costing in low- and middle-income countries (Barnum & Kutzin 1993). However, there are intrinsic problems to these methods. First, the costing process assumes a linear underlying cost function, which is unlikely to pertain. Secondly, resource-use inefficiencies are hidden, because the implicit assumption is that apportioned aggregate costs are incurred appropriately. Unnecessary costs can be incurred through idle staff time, over-prescription of drugs, unnecessary diagnostic tests, inappropriate length of stay and other redundant activities. The “tracer” approach to costing has the potential of resolving some of these practical problems.
The overall aim of the study was an attempt to use tracer diagnoses to analyse and compare the cost structure in different hospitals. This paper addresses three objectives. First, to show the application of the tracer method in establishing hospital service costs. Secondly, to profile tracer costs by major input category, as a means of exploring resource-use patterns and potential areas of resource wastage or inefficiencies across hospitals. Thirdly, to draw relative efficiency implications based on the observed cost structures.
A case for the tracer method
The concept of tracers is not new to the medical field as it has a long application history in radiology and radiotherapy. The notion of using tracers to understand complex biological systems can be extended to understanding health systems or parts of it. Work by Kessner et al. (1973) on assessing quality of care using tracers provides an early attempt to use tracers to understand complex issues in health. The major challenge in employing this method lies in identifying tracers that are capable of exposing the workings of the system under investigation.
Kessner et al. (1973) define a tracer as ‘‘a specific health problem, that, when combined into sets, allows health care evaluators to pinpoint the strengths and weaknesses of a particular medical practice setting or entire health service network by examining the interaction between provider, patients and their environment’’. What is important is that tracers must be identifiable and distinct conditions capable of complementing each other in shedding light on the system under review (or its components). This method has been mostly used for assessing process and outcome aspects of quality of care (Mates & Sidel 1981; Payne et al. 1984).
The selection of tracers is crucial because it determines the validity of extrapolating from the analyses. Kessner et al. used the following selection criteria: (1) a tracer should reflect the activities of health professionals—it should be a condition with significant functional impact. That is, it should be likely to be treated and capable of causing considerable functional impairment; (2) the condition should be well defined and easy to diagnose; (3) the condition must be prevalent to allow for sufficient numbers; (4) the condition must be susceptible to the quality or quantity (or both) of the services received by the patient; (5) there must be consensus on minimal standards for managing the condition; and (6) the effects of non-medical factors (social, cultural, economic and behavioural) on the tracer and its epidemiology should be understood.
In practice it might not be possible to meet all the criteria because most conditions fail to meet at least one. For instance, understanding non-medical factors associated with a condition or disease may be complex. The absence of certainty of outcomes of many medical interventions means that reaching consensus on management protocols is difficult. Nevertheless, there are usable international and national guidelines on the management of most health problems. The criteria outlined by Kessner et al. (1973) are adaptable and serve as a workable framework for effective application of the method by ensuring that tracers selected reflect as much as possible the target health system (or its components).
Tracers can be used for a variety of evaluation purposes in health care. For example, exploration of macro-level health costs and economic efficiency (Arredondo 1997), institutional disease costing (Babson 1972; Kirigia et al. 1998; Sanyika et al. 1999), provider behaviour and quality of care (Mates & Sidel 1981; Payne et al. 1984) and differences in patient severity across hospitals (Lezzoni et al. 1996) are all implemented applications.
Tracers and hospital resource use
We argue that the tracer method used in combination with micro-costing1 provides a powerful tool in exploring hospital cost structures and relative efficiency for three reasons. First, the process of estimating and costing input consumption per defined case or patient brings us closer to the actual cost function of a given service which most accounting methods cannot do. Secondly, the use of clearly defined tracers or sets of tracers mitigates the confounding effect of case mix in comparing costs. Thirdly, operational inefficiencies are easily exposed by comparing actual costs against standard or normative costs derived from standard management protocols.
Micro-costing is generally acknowledged as the most effective method of costing (Babson 1972; Drummond et al. 1997), especially when used in conjunction with randomised clinical trials or observational research (Luce et al. 1996). Its application in a hospital setting requires a number of data collection strategies: time and motion surveys,2 medical chart reviews, patient diaries and periodic interviews with staff and patients. This approach can also be applied to retrospective data but it requires sophisticated information systems, and it is difficult to track some inputs after the fact (Luce et al. 1996). It is useful to use micro-costing when macro measures correspond poorly with resource use or when the cost of an input is integral to the analysis (Eisenberg et al. 1984). The main demerit of micro-costing is that it requires intensive research.
Frequently, micro-costing is used in combination with macro-costing in a single analysis (Hanson et al. 2000; Hongoro 2001). Macro-costing methods generate cost structures or unit costs that are based on aggregate cost patterns, which might incorporate production inefficiencies. By using common diseases across hospitals with clear case definitions, the problem of differences in case mix, and to some extent case severity, in comparing hospitals could be partially addressed.
The study was conducted in 1999 at six provincial hospitals (coded 1–6) in Zimbabwe. Ethical clearance was obtained from the Medical Research Council of Zimbabwe (MRCZ). The study design was prospective and used three tracer diseases: pulmonary TB, simple and severe malaria. The choice of tracers was made by a selected panel of experts3 using the criteria outlined by Kessner et al. as a guide.
Malaria and TB were selected because they are the major causes of hospital morbidity and mortality across provinces (MoH 1998). In one province for instance, 3965 TB cases were hospitalised (7% of all admissions) in 1997, and 438 died (20% of all hospital deaths). The MoH provided the appropriate national management guidelines for the study.
The case definitions used in the study are critical in the interpretation of costs. Malaria case definitions vary with age, levels of malaria transmission and acquired immunity (McGuinness et al. 1998), and controversies surrounding definitions are not unusual (WHO 1990). In this study, a patient was classified as having simple malaria if he or she was presented with a fever and either lived in or had travelled in a malarial area within the past 3 weeks (MoH 1998). Fever is obviously associated with many other diseases which may lead to over-diagnosis. In fact, low-density parasitaemia may require no therapeutic intervention (Redd et al. 1996, McGuinness et al. 1998). However in highly malarious areas, it is advised that any child presented with a fever, whether the other obvious causes are present or not, should receive malaria treatment (Redd et al. 1992). The government's policy on malaria is that of presumptive treatment, and laboratory confirmation is carried out at hospital level. A similar policy exists in Malawi (Redd et al. 1996), The Gambia (Müller et al. 1996) and many other countries. Although highly sensitive, such a policy is associated with the problem of false positives. Nonetheless the benefits of such a policy are argued to outweigh the costs.
A patient was considered to have severe/complicated malaria if he or she had one or any of these signs: impairment of consciousness, fits/convulsions, pallor, jaundice, ‘‘Coca-Cola’’ urine, breathlessness, prostration, low blood pressure, persistent vomiting or bleeding from the mucosa (WHO 1990; MoH 1998). Once again, WHO acknowledges the controversies associated with such a definition, and argues that for practical purposes clinical suspicion alone, cases previously exposed to P. falciparum should prompt a therapeutic trial of an effective antimalarial.
The case definition for TB was described according to sputum status and history. Using the former, sputum positive pulmonary TB refers to a patient with two consecutive positive sputum smear examinations or with one positive smear examination and a chest X-ray suggestive for TB. According to history, a new case is defined as a patient who never received a full course of TB treatment or received treatment for less than 1 month (MoH 1994a). The conventional treatment protocol used for such patients is Category 1 High priority: 2 months of intensive treatment followed by a continuation phase of 4 months. The drug regimens are given in the Essential Drug List for Zimbabwe (MoH 1994b). The introduction of Directly Observed Treatment (DOT) in some areas in the past few years might have altered this management regimen.
Case recruitment for malaria was carried out using principal clinical diagnosis, and for TB it was entirely on the basis of case definition criteria. The operationalisation of case definitions was understandably a function of health provider compliance with existing guidelines, and to some extent the knowledge and experience of research assistants used in the study. We did not seek to evaluate the national protocols but simply to use them as constructs of experts’ consensus irrespective of their shortcomings.
For practical purposes secondary diagnosis was not considered in the recruitment exercise. Co-morbidities were common confounders especially for TB cases. During the time of the study HIV/AIDS-related drugs were not available in public hospitals. We made an attempt to explore costs of secondary or opportunistic infections by separating TB drug costs from total drugs costs because the overall cost of treatment was likely to be affected by treatments to control HIV-related conditions (Nunn et al. 1993; Saunderson 1995), however homogeneity was retained given the similarities in the number of HIV–TB cases (about 90%) across hospitals.
A total of 207 malaria cases and 158 TB cases were recruited for the study between January and June 1999. No malaria and TB data were collected at hospitals 5 and 6, respectively, due to logistical constraints. Patients were recruited by experienced nurses at hospital outpatient or casualty departments. Any case that presented with signs and symptoms of any of the tracers, and was subsequently admitted was recruited into the cohort. We obtained informed consent from the patient or patient's guardian before including them into the cohort. Once recruited each patient was given an identifier code and followed up from the time of admission in hospital to discharge.
We used both micro (tracer)- and macro-costing methods and our information sources were patient daily records, the patient, laboratory and ward staff. Daily visits took place at varied time intervals to check what was done to and for the patient. We recorded number and type of laboratory tests carried out, number of chest X-rays taken, drugs administered (type, doses and period) including what was provided at discharge, medical supplies provided (type and quantity), number of days in hospital, frequency of meals per day, and attending staff type, numbers and grade mix.
We used unit prices obtained from the Government Medical Stores4 to value the input quantities used. We calculated staff costs using average nursing, doctors’ and general hands’ full time equivalents per day, and the daily average number of inpatients in the ward. Equivalent salary costs per hour per staff category and the length of stay were then used to calculate the total staff costs per case. The cost of bedding and linen, and laundry was computed as an average cost per day. The average cost per meal per day was calculated using total annual kitchen costs minus staff costs, and the number of inpatient days. Food costs per case were obtained by multiplying the average cost per meal per day by the number of meals per patient per day and by patient length of stay.
We costed, in detail, malaria parasite tests (MPs) and TB sputum microscopy tests in hospitals 4 and 6 using the ingredients approach. The results were similar to those found in a country-wide study (DANIDA 1999). Input quantities and purchase prices were essentially the same for public hospitals because of the central procurement system.5 We did a similar costing for chest X-rays. Overhead costs although important were purposefully omitted because we sought to understand costs that varied with patient care, their magnitude and the sources of this variation.
Cost and activity data was analysed using spss and excel computer software packages. Data was trimmed to correct for outliers for specific aspects of the analysis explain the differences in sample sizes shown in the results. After data trimming 197 malaria cases and 151 TB cases remained for further analysis.
Characteristics of malaria sample
Of the 197 enlisted malaria patients, 53% (104) were referred cases and 47% (91) were self-referred cases reflecting referral system problems6 (Table 1). Fifty-three percent (105) of the patients were females and 47% (92) were males and significant differences were observed across hospitals (P < 001). Patient ages ranged from 1 to 84 years, and the mean age was 26 years. The number of simple and severe malaria cases was 83 (42%), and 114 (58%), respectively. Hospitals 1 and 2 had high proportions of simple malaria cases; 73 and 59%, respectively. Interestingly, we found case severity to be independent of referral status (P > 0.05, χ2–Fisher's Exact test). The mean hospital case fatality rate was 17%. Validation of the principal clinical diagnosis at admission involved looking at the results of 154 recorded malaria microscopy (MPs) tests: 76% (117) of the cases were positive and 24% (37) were negative, and differences across hospitals were significant7. Altogether, there were significant differences in patient sex distribution, lengths of stay, patient referral status and case severity mix across the hospitals.
Table 1. Descriptive statistics of admitted malaria patients by hospital
|Mean patient age||22||25||34||22||30||NS|
| Female||31 (55%)||13 (38%)||23 (49%)||16 (43%)||22 (96%)|| |
| Male||25 (45%)||21 (62%)||24 (51%)||21 (57%)|| 1 (4%)||<0.0001‡|
|Type of malaria|
| Simple malaria||41 (73%)||20 (59%)|| 8 (17%)||10 (27%)|| 4 (17%)|| |
| Severe malaria||15 (27%)||14 (41%)||39 (83%)||27 (73%)||19 (83%)||<0.0001‡|
| Referred||28 (50%)||12 (35%)||20 (44%)||18 (49%)||21 (91%)|| |
| Self-referred||28 (50%)||22 (65%)||25 (56%)||19 (51%)|| 2 (9%)||<0.001‡|
|Length of stay (days)|| 3|| 4|| 5|| 6|| 5||<0.001¶|
| Mean LOS: simple|| 2.9|| 3.2|| 3.4|| 5.0|| 4.8||<0.02¶|
| Mean LOS: severe|| 3.4|| 5|| 5.1|| 6.6|| 4.6||<0.03¶|
| Case fatality rate|| 6 (11%)|| 6 (17%)|| 8 (17%)|| 6 (16%)|| 5 (25%)||<0.001‡|
| Positive||27||18 (55%)||45 (98%)|| 8 (32%)||19 (86%)|| |
| Negative||(48%)†||15 (45%)|| 1 (2%)||17 (68%)|| 3 (14%)||<0.0001‡|
Costs of malaria management
Analysis by the hospital provided an opportunity for an in-depth look at variations in unit costs and cost structure across hospitals. The mean cost per case varied across hospitals with hospitals 1 and 2 having relatively low costs per case (Table 2). Hospitals 3 and 5 had mean costs above the overall mean cost per case of ZW$1528.
Table 2. Cost per malaria case by hospital in ZW$ (US$1 = ZW$55)
Hospital ranking does not change when cost per case results are presented by case severity—that is, for simple and severe malaria (Table 3). Hospitals 1 and 2 maintain lower unit costs compared to the rest. This means that observed low costs are not wholly explained by differences in case severity. The cost per simple case of malaria is oddly high for hospital 5 because of a relatively small sample size. Hospital 4 remains the most costly per hospitalised malaria case. Excluding hospital 5, the cost per simple malaria case is significantly less (by at least 1.2 times) than that for severe malaria case. The observed differences in unit cost seem to be explained largely by significant differences in the mean length of stay by severity across the hospitals (P < 0.018).
Table 3. Mean cost per case for severe and simple malaria (ZW$) (1US$ = ZW$55)
| ||n = 41||n = 20||n = 8||n = 10||n = 4|
|Simple malaria (min/max)||693 (135–2725)||769 (238–1930)||1317 (327–2692)||1970 (371–3631)||2498 (1879–3087)†|
| ||n = 15||n = 14||n = 39||n = 27||n = 19|
|Severe malaria (min/max)||1336 (314–3517)||1524 (253–3666)||1936 (197–5580)||2501 (458–3962)||1716 (478–3630)|
A breakdown of costs per case by input category shows that hotel services constitute a large proportion of hospitalisation costs in most cases (Table 4), but hospital 2 was an exception with 50% of costs attributable to staff time and only 38% to hotel services. Low hotel services could be explained by poor quality of services or production efficiency. Quality assessment results reported elsewhere (Hongoro 2001) show no quality difference between these hospitals suggesting the latter as a possible explanation. The relatively high staff costs might be attributed to relatively higher staff numbers and grades9 because the hospital is near the capital city which is an attractive feature for the staff.
Table 4. Cost distribution, malaria (ZW$) (1US$ = ZW$55)
|Hotel services||608 (70%)||401 (38%)||1263 (68%)||1497 (63%)||1088 (59%)||<0.001|
|Staff||183 (21%)||532 (50%)||471 (25%)||700 (30%)||549 (30%)||NS|
|Diagnostic||29 (3%)||57 (5%)||46 (2%)||34 (1%)||103 (6%)||<0.001|
|Drug||30 (4%)||37 (4%)||48 (3%)||77 (3%)||74 (4%)||<0.01|
|Medical sup.||15 (2%)||27 (3%)||38 (2%)||54 (2%)||40 (2%)||NS|
|Total||865 (100%)||1055 (100%)||1866 (100%)||2363 (100%)||1852 (100%)|| |
Drugs and medical supplies constituted a small proportion of unit costs with a mean of 7%. We found significant differences in the proportions of hotel services costs (P < 0.001), drug costs (P < 0.01) and diagnostic costs (P < 0.001) across the hospitals implying that hospitals employed different resource-use strategies (or input mix) in the production of malaria services10. All but hospital 2 had similar proportions of staff costs. This is probably explained by similarities in public staffing patterns.
Overall, there seemed to be no significant difference across hospitals in the distribution of component costs per case between severe and simple malaria, and between total malaria and each of the severity categories. However, differences in cost levels can be seen across all cost categories between severe and simple malaria, and hospitals. Resource-use patterns appear to be fundamentally similar and overall cost differences seem to be driven by differences in length of stay.
Characteristics of tuberculosis cases
Background characteristics of TB patients were generally similar across the five hospitals (Table 5). There were significant differences in TB referral patterns across hospitals. Again, this probably reflects a weak referral system. The overall mean length of stay in hospital was 7 days. Patients could be admitted for 1 to 32 days. The length of stay was influenced by case severity, clinical practice at the hospital and whether or not there was a specialised infectious disease hospital nearby. Hospital 6 had the highest mean length of stay of 10 days. Such short hospitalisations might be explained by the implementation of DOTs, and home-based care programmes in the 1990s for HIV positive cases11. Hospital deaths caused by TB ranged from 8 to 19%, which is considerably higher than those for the other diseases.
Table 5. Background characteristics for tuberculosis patients by hospital (n = 151)
|Mean patient age||33||33||33||30||38|| |
| Female||17 (57%)||12 (33%)||11 (42%)||13 (42%)||11 (42%)||NS|
| Male||13 (43%)||24 (67%)||15 (58%)||18 (58%)||15 (58%)|| |
| Referred||16 (53%)||19 (51%)||17 (65%)||10 (31%)||19 (76%)||<0.01|
| Self-referred||14 (47%)||18 (49%)||9 (35%)||22 (69%)||6 (24%)|| |
|Length of stay (days)||5||6||4||9||10||<0.01|
|Case fatality rate||4 (13%)||3 (8%)||5 (19%)||5 (16%)||2 (8%)||<0.01|
The patient sex profile in the cohort was similar. However, there were significant differences in the average length of stay (P < 0.01), referral status of patients (P < 0.01) and hospital deaths due to TB (P < 0.01) across the hospitals. Hospital 3 had the highest death rate of 19% despite its strict referral policy to a specialist hospital.
Costs of tuberculosis management
TB patients were not classified by the level of illness severity, and the cost results presented are based on 151 WHO Category I12) cases. The mean cost per TB case includes direct costs incurred by patients admitted for the first time with smear-positive TB (or indicative X-ray results), for their intensive-phase treatment.
Table 6 shows the cost per case of TB by the hospitals considered. Hospital 6 had the highest cost per case with costs ranging from ZW$1315 (US$23.91) to ZW$6010 (US$109.27). The high costs can be linked to its relatively longer mean length of stay (10 days). Hospitals 1 and 2 had the least cost per case while hospitals 4 and 3 had intermediate costs. The wide ranges in cost per case for all the hospitals might be explained by differences in case severity and length of stay. However, median values show a similar cost pattern.
Table 6. Cost per tuberculosis case by hospital in ZW$ (US$1 = ZW$55)
A breakdown of the cost per TB case shows that the input cost pattern per case is similar across hospitals (Table 7). Drug costs are particularly low because most cases had short stays in hospitals and were at the beginning of the intensive-phase treatment. In interpreting these results it is important to note the difference between drug costs for an episode of TB, and drug costs during initial hospitalisation.
Table 7. Cost distribution per tuberculosis case in ZW$
|Staff|| 298 (19%)|| 638 (44%)|| 321 (24%)|| 632 (23%)||1520 (44%)||<0.001|
|Diagnostic|| 148 (10%)|| 159 (11%)|| 139 (10%)|| 93 (3%)|| 115 (3.4%)||<0.001|
|Drug|| 10 (1%)|| 10 (1%)|| 6 (0.5%)|| 13 (0.5%)|| 22 (0.6%)||<0.001|
|Medical sup.|| 0.12 (0%)|| 6 (0%)|| 0.6 (0%)|| 2 (0%)|| 6 (0%)||<0.05|
|Hotel services||1076 (70%)|| 684 (44%)|| 891 (66%)||1955 (73%)||1762 (51%)||<0.001|
|Total||1532 (100%)||1462 (100%)||1358 (100%)||2695 (100%)||3425 (100%)|| |
Unsurprisingly, the major component of costs per hospitalised TB patient was staff and hotel services. Diagnostic costs were a relatively small fraction of patient costs but varied across hospitals. Such variations in proportion of diagnostic costs across hospitals are explained by differences in intensity of laboratory use. Diagnostic costs were relatively low in hospital 4 because of poor structural quality. Shortages of laboratory reagents and breakdown of equipment occurred frequently. For a period of 5 months in 1998 no laboratory tests for TB were carried out (personal communication with the Chief Laboratory Technologist). In this hospital diagnosis for TB was mainly through sputum microscopy and chest X-rays.
The results of this study are consistent with those from the earlier costing studies (Mills et al. 1997; Hanson et al. 2000; Hongoro 2001). Worth highlighting is the relatively high cost of inpatient care, US$28 and US$33 for malaria and TB, respectively, in a country where annual per capita health expenditure is less than US$20. The confounding effect of HIV-related co-morbidities cannot be ignored when interpreting such cost results. We assessed the contribution of other drugs not related to the principal diagnosis13 to the overall cost per case and we found it to be minimal. This was unexpected with the high prevalence of HIV amongst TB patients. Evidence from a Kenyan study showed marked differences in unit costs between HIV-negative patients (US$16.62) and HIV-positive-patients (US$32.94) (Nunn et al. 1993). Although we did not control the HIV status, its effect might have occurred through its impact on the length of stay.
To establish whether observed cost variations were because of differences in service efficiency requires an appreciation of service quality across hospitals. Results of cost-quality assessments reported elsewhere (Hongoro 2001) show that hospitals 1–3 were better performers in terms of tracer quality levels achieved for the cost incurred. Hospitals 4–6 had consistently high mean costs for comparable or lower quality levels, which is suggestive of sub-optimal use of resources (or relative inefficiency). Differences in service efficiency seem to be explained more by the average length of stay, which varied significantly across hospitals. Those hospitals that had higher length of stay tended to have higher costs per case.
Hospital case mix is partly influenced by the performance of the referral system and if it is not functioning properly, as was found in this study and elsewhere (Sanders et al. 1998), hospitals may expend resources on inappropriate cases—that is, on cases that do not technically deserve hospitalisation. For instance, admission of more simple malaria cases may deflate average costs and the converse might be true. However, we found that case severity was independent of referral status suggesting that inappropriate hospital admissions did not necessarily affect observed hospital cost profiles. Self-referrals for malaria cases seemed to be a general phenomenon.
Notable similarities in cost distribution per case were observed with some differences in some cost components across hospitals (for example staff costs). The proportion of drug costs to total patient costs was small. For malaria, this might be explained by the high proportion of simple malaria cases that are treatable with inexpensive drugs like chloroquine or pyrimethamine and sulphadoxine. De Jonghe et al. (1994) found a higher proportion of drug costs per TB case of 6% (compared to 1%) because drug costs were considered per episode of TB and not just for the initial hospitalisation phase.
Costs of hotel services formed a major part at 60% of the cost per case. Similar results were found in Malawi, where kitchen costs, for example, ranged from 56 to 80% per inpatient day (Mills et al. 1991). A cross-country study (involving Malawi, Mozambique and Tanzania) on TB hospital costs by De Jonghe et al. (1994) found kitchen costs ranging from 30 to 53% of patient costs. The results show that it is not only costly to treat but to keep a patient in hospital.
The contribution of staff costs per case across hospitals ranged from 21 to 50% for malaria and 19 to 44% for TB. Low staff costs observed in this study were owing to reliance on technical and auxiliary staff, and student nurses (especially in hospital 1). All hospitals had training schools for nurses, which provided a cheap source of labour. Time and motion studies (not conducted in this study) might have exposed this issue better. Nonetheless, the results show that efficiency improvement activities in hospitals need to focus on hotel and staff costs because they constitute a major part of the inpatient cost structure.
The major caveat to using the study results in making relative hospital efficiency interpretations is that it was based on only three tracers, and quality of care issues are not presented here. The conditions used in our study are among the top five diseases of public health significance in the country. The presence of information gaps in the study affected unambiguous ranking of hospitals but did not affect the extent to which the tracer method exposed resource-use patterns and unit costs.
The study shows that it is possible to calculate detailed costs and examine hospital cost structure more closely using the tracer method. Use of a representative sample of tracers can potentially address case mix and case severity problems in hospital costing in data-constrained environments. Further empirical research is required on this issue. There is also scope for using tracers to explore x-inefficiency through comparing standard costs derived from normative management protocols and actual costs.
Cost data generated in this study are useful for hospital management as they highlight the major cost components that should be targeted for improving efficiency in providing inpatient care services. In addition, the study results provide baseline data for future use in economic analyses of malaria and TB, which continue to be major causes of morbidity and mortality in Zimbabwe.
We thank Justin Parkhurst for reviewing an early draft of the paper and Isabel Sinha for formatting the manuscript.
Dr Charles Hongoro, Health Systems Development Programme, Department of Public Health and Policy, London School of Hygiene and Tropical Medicine, Keppel Street, London WC1E 7HT, UK. E-mail: firstname.lastname@example.org (corresponding author).
Dr Barbara McPake, Department of Public Health and Policy, Health Policy Unit, London School of Hygiene and Tropical Medicine, Keppel Street, London WC1E 7HT, UK. E-mail: email@example.com