Comparison of cost effectiveness of directly observed treatment (DOT) and conventionally delivered treatment for tuberculosis: experience from rural South AfricaBMJ 1997; 315 doi: http://dx.doi.org/10.1136/bmj.315.7120.1407 (Published 29 November 1997) Cite this as: BMJ 1997;315:1407
- Katherine Floyd, lecturer in health economicsa,
- David Wilkinson, specialist scientistb,
- Charles Gilks (), senior clinical lecturer in tropical medicinea
- a Division of Tropical Medicine, Liverpool School of Tropical Medicine, Liverpool L3 5QA
- b Centre for Epidemiological Research, South Africa (Hlabisa), Medical Research Council, PO Box 187, Mtubatuba 3935, South Africa
- Correspondence to: Dr Gilks
- Accepted 29 July 1997
Objective: To conduct an economic evaluation of directly observed treatment (DOT) and conventionally delivered treatment for the management of new cases of tuberculosis in adults.
Design: Community based directly observed treatment, which has been implemented in the Hlabisa district of South Africa since 1991, was compared with a conventional approach to tuberculosis treatment widely used in Africa. Each was assessed in terms of cost, cost effectiveness, and feasibility of implementation within existing resource constraints.
Setting: Hlabisa Health District, South Africa.
Subjects: Adult patients with new cases of tuberculosis on smear testing; the number of cases increased from 20 per month to over 100 from 1991 to 1996.
Main outcome measures: Cost of case management in 1996, cost effectiveness in terms of the cost per case cured, and bed requirements in comparison with bed availability for the 1990, 1993, and 1996 caseload. Costs are expressed in US dollars at values for 1996.
Results: Directly observed treatment was 2.8 times cheaper overall than conventional treatment ($740.90 compared with $2047.70) to deliver. Directly observed treatment worked out 2.4-4.2 times more cost effective, costing $890.50 per patient cured compared with either $2095.60 (best case) or $3700.40 (worst case) for conventional treatment. The 1996 caseload of tuberculosis required 47 beds to be dedicated to tuberculosis to implement directly observed treatment, whereas conventionally delivered treatment would have required 160 beds; the current number of beds for tuberculosis treatment in Hlabisa is fixed at 56.
Conclusions: Because of the reduced stay in hospital, directly observed treatment is cheaper, more cost effective, and more feasible than conventional treatment in managing tuberculosis in Hlabisa, given the existing hospital bed capacity and the escalating caseload due to the HIV/AIDS epidemic. Such results may hold elsewhere, and wherever conventional tuberculosis management is practised a switch to directly observed treatment will increase hospital capacity to cope with a growing caseload.
Tuberculosis is a problem of global importance, and the number of cases is rising as a consequence of population growth, worsening poverty, and the HIV/AIDS epidemic
Conventional approaches to management are increasingly difficult to implement, especially when caseloads are rapidly increasing
This study found that community based directly observed treatment, a novel approach to treating tuberculosis, was considerably cheaper and more cost effective than a conventional approach entailing prolonged admission to hospital
Because directly observed treatment considerably reduces hospital stay, its implementation will increase the capacity of hospitals to cope with a rising tuberculosis caseload wherever the conventional approach is currently used
This South African model of directly observed treatment for tuberculosis is worthy of serious consideration by policymakers and programme managers elsewhere
Tuberculosis is a problem of global importance,1 and population growth, increasing poverty, and the AIDS epidemic mean that the number of reported cases continues to grow.2 3 Coping with this rising caseload is difficult: delivery of treatment is not straightforward, and successful programmes in Africa have typically relied on lengthy admissions to ensure patient compliance.4 Given that already constrained resources preclude either the construction of new wards or the hiring of additional staff and that hospitals are already operating at or beyond capacity,5 6 this approach seems to be increasingly unviable. Other approaches to case management are urgently needed that can achieve high patient compliance while being cheaper per patient and less dependent on hospital care.
Community based directly observed treatment (DOT) is currently the standard approach to care in the United States. Though expensive,7 it is probably less costly than admission, and high cure rates have been achieved. To evaluate whether directly observed treatment could be an attractive economic option in a resource poor setting, we studied such a programme in Hlabisa district, South Africa—one of the few sites in a developing country where it has been implemented. We have previously reported the costs, cost effectiveness, and feasibility of implementing several strategies within existing resource constraints.8 In this paper we compared directly observed treatment with the conventional approach that has been widely used elsewhere in Africa.
Subjects and methods
Hlabisa is a rural district where the tuberculosis caseload has increased from 25 to over 100 per month during 1991-6, largely because of the emergence of the HIV epidemic. In 1993, 35% of adult patients with tuberculosis tested positive for HIV infection9; in 1997 the figure was 70% (unpublished data). There are seven clinics and a district hospital, and the yearly income per head is $1730 (£1081).10
The two strategies were compared after diagnosis of new cases of tuberculosis in patients with positive results on smear testing. Before diagnosis the strategies were similar.
Directly observed treatment has been described in detail elsewhere.11 12 After diagnosis, the time patients spend in hospital is determined by their clinical condition and the time taken to arrange community care. In 1996 the average length of stay was 17.5 days. While in hospital, patients receive daily treatment with four drugs (isoniazid, rifampicin, pyrazinamide, and ethambutol) and, with the help of field workers, identify someone who can supervise treatment after discharge. The emphasis is on choosing a supervisor who makes accessing directly observed treatment convenient for the patient. In 1995, 56% of patients were supervised by unpaid non-health workers, most of whom were storekeepers; 21% were supervised by health clinic staff; 21% by community health workers; and 2% by staff in tuberculosis wards. On discharge, field workers transport patients to their supervision point in a hospital vehicle. Supervisors are given the prepacked drugs required for completion of the six month treatment course. The drugs used after discharge are the same as those given in hospital, except that they are given at higher dose and are taken on an intermittent basis twice weekly. Patients therefore visit their supervisor twice a week to take their drugs under direct observation, visiting on average 48 times. Field workers visit supervisors monthly to check compliance, collect patient outcome data, and trace absconders. A hospital doctor manages the programme.
Conventional approach—Patients stay in hospital for the first two months of treatment under the conventional management strategy for tuberculosis. This lengthy stay is designed to ensure compliance with treatment during the intensive phase (at least), when patients are most infectious, and with some drug regimens is necessitated by the use of streptomycin. At discharge, patients are given a one month supply of drugs, and they subsequently collect their drugs once a month from their nearest clinic. Drug regimens vary, but we assumed that the six month course recommended by the World Health Organisation would be used. This consists of two months of isoniazid, ethambutol, pyrazinamide, and rifampicin, followed by four months of rifampicin and isoniazid, with exact dosages dependent on body weight. Two sputum smears are examined at 2, 4, and 6 months.
We assessed both average and marginal costs.13 We report average costs because they are a better reflection of true variable costs and therefore of greater use for planners and policymakers concerned with national policy for a large number of facilities, such as a country's tuberculosis control strategy.14 Costs were assessed from a societal perspective in United States dollars at values for 1996.
Health system costs
For each strategy component except drugs, total annual recurrent health system costs for the financial year 1995–6 were calculated using one or more of the following: expenditure files which provided a breakdown for inpatient, outpatient, and health clinic care; the hospital payroll; the rate paid by the ministry of health for vehicle usage; vehicle logbooks; and interviews with staff. Gross salaries were used to calculate staff costs. Average drug costs per patient were established by combining a costing of each regimen by weight category (using the hospital pharmacy price list) with data from the tuberculosis register on the weight of patients.
Total annual capital costs were calculated using quoted 1996 purchase prices, reasonable assumptions concerning expected useful life (5 years for a vehicle, 10 years for equipment, and 30 years for buildings), and a discount rate of 8% (the difference between the interest rate paid on government treasury bonds and the inflation rate).
For hospital costs, 92.5% of the costs of administrative and support staff and of capital costs were allocated to inpatient care and 7.5% to outpatient care, since staff involved in direct outpatient care accounted for 7.5% of the total costs of staff involved in direct patient care. We allocated 4.1% of laboratory costs (staff and equipment not used solely for tuberculosis work) to sputum smear testing since smears comprised 4.1% of the total number of laboratory tests in 1995.
Annual output data were for 1995—that is, the number of days in hospital; the number of visits to outpatient departments, health clinics, and community health workers; the number of patients receiving directly observed treatment for tuberculosis; the total number of patients with tuberculosis; and the number of laboratory tests done. Sources included hospital admission records, the medical superintendent, programme reports, and the regional laboratory manager.
For average cost calculations we assumed that visits to outpatient departments, health clinics, and community health workers by a patient with tuberculosis would cost the same as visits for other health problems. For spending a day as an inpatient we assumed that all costs (except those associated with staff allocated to the tuberculosis ward, drugs, laboratory investigations, and x ray examinations) would be the same for a patient with tuberculosis as for any other inpatient. The cost of a visit for directly observed treatment was calculated by combining the cost per visit with the pattern of supervision in 1995 as costs vary according to the site chosen for supervision of treatment.
A questionnaire was administered to all patients eligible for treatment under the directly observed treatment programme at the outset of the study (48 in total). The time and travel costs associated with visits to hospital and health clinics were recorded. When different sites were chosen for directly observed treatment, the time and travel costs incurred to visit them were also noted. Time costs were translated into monetary costs using monthly incomes reported in a second questionnaire (designed one week into the study and administered to the 35 original respondents in the tuberculosis ward at this time). Calculations assumed patients would work 25 days a month and 8 hours a day. Travel and time costs were then summed and average costs associated with visits to hospital and health clinics and for directly observed treatment by specific type of supervisory site calculated. The average costs of directly observed treatment by supervisory site in 1995 were combined to calculate the average cost incurred by a patient per directly observed treatment visit.
Time constraints meant that the only data collected on community costs were those incurred when patients were accompanied on hospital visits. These were comparatively small and were included in the patient cost per hospital visit. Costs incurred by non-health worker supervisors were considered to be nil because the time commitment was negligible and supervision creates no additional costs for supervisors (unpublished data).
We chose cure as the measure of effectiveness—the WHO's criterion for measuring a programme's success.15 To estimate a best and worst case scenario for the likely effectiveness of the conventional strategy, we analysed data on the outcomes of a six month cohort of patients with positive results on smear testing in Tanzania in 199016 and annual reports with detailed outcome data from Malawi 1989-933 because these programmes used a similar case management approach. For directly observed treatment we used data from the 1991–4 audits of the Hlabisa programme.12 Cure is not routinely assessed bacteriologically in Hlabisa, but a retrospective analysis of 109 patients found that 95% were cured.17 In this analysis, the rate of completion of treatment was therefore multiplied by 95% to give an estimated cure rate. For consistency, the same assumption was applied to the data from Malawi and Tanzania for the small number of patients who completed treatment but for whom cure was not confirmed.
Patients who died or left the district during treatment were excluded from the analysis. Differences in the death rate may reflect the varying impact of the AIDS epidemic. The proportion of patients leaving the district during treatment is related to population transience. Neither of these is a direct function of the tuberculosis programme itself, so omitting these categories prevented unnecessary distortion of the analysis.
Calculation of cost effectiveness
Cost effectiveness was calculated in three steps. First, the proportion of patients who completed treatment was multiplied by the cost of managing a patient up to the completion of treatment. Second, the cost of a patient not completing treatment (calculated by assuming that default would occur at hospital discharge) was multiplied by the proportion of patients not completing treatment. Third, the resulting two costs were summed and divided by the cure rate.
Directly observed treatment was considerably cheaper than the conventional strategy (Table 1). It was 2.7 times cheaper for the health system, 3 times cheaper for the patient, and 2.8 times cheaper overall. In both cases, health system costs accounted for 87% of total costs. Admission to hospital was the most costly item. For directly observed treatment it accounted for 75% of health system costs, 76% of patient costs, and 75% of costs overall; for the conventional approach it accounted for 94% of health system costs, 88% of patient costs, and 93% of costs overall. The only other important cost items were visits for directly observed treatment (14% of the total cost of the directly observed treatment strategy) and organisation of supervision and supervision of supervisors (5% of the total cost of the directly observed treatment strategy). No other item—including drugs—accounted for more than 5% of total costs.
The data for individual strategy components (tables 2 and 3) showed that a day in hospital was costly ($27.80 per day for the health system and $4 for the patient), as was an outpatient visit ($16.70 for the health system and $9.70 for the patient). In terms of time, a hospital visit cost a patient five hours. Visits to health clinics and community health workers cost less. The different drug regimens cost about the same, and sputum smear examination was a minor cost.
The organisation of supervision and the supervision of supervisors were less expensive than expected, at $38.90 per patient. Visits for directly observed treatment were also notable for their low cost. Visits to non-health workers were cheapest, costing the health system nothing and the patient just 35 minutes, estimated to equate to $0.30. On average, each visit for directly observed treatment cost the health system $1.70 and the patient $0.45. Overall, the input of non-health workers reduced the total cost of directly observed treatment by 26% from the health system's perspective, by 53% from the patient's perspective, and by 31% overall in comparison with what costs would be if health clinics alone were used for supervision.
Directly observed treatment based in the community was therefore inexpensive. In combination, arrangements for supervision, supervision of supervisors, and 48 visits cost the health system $120.50 per patient (equivalent to 4.3 days spent in hospital receiving treatment for tuberculosis) and the patient $21.60 (equivalent to five days in hospital).
Directly observed treatment also seemed to use resources efficiently (Table 4). At $890.50 per patient cured, it was between 2.4 and 4.2 times more cost effective than the conventional approach. Moreover, directly observed treatment was feasible within existing resource constraints (Table 5). Both strategies were possible in 1990, but in 1996 the increased caseload meant that directly observed treatment was the only strategy that could be implemented without displacing patients from other wards, reducing the quality of care provided, and needing extra investments in infrastructure (and, probably, staff).
Directly observed treatment is an attractive economical option in Hlabisa. It is cheap, cost effective, and implementable within existing resource constraints.
This strategy is cheap because treatment based in the community that is supervised by field workers is cheaper than lengthy hospital supervision of care (all other cost items were insignificant in terms of impact on total costs). This is also likely to be true elsewhere in South Africa because the costs of the major inputs—staff, vehicles, and fuel—do not vary much; population densities in most districts are unlikely to fall to the point where the costs of vehicles and fuel would become inordinately expensive; and non-health worker input is considered acceptable by the Department of Health.18
Even in very poor African countries, the cost of hospital care is unlikely to fall enough to make directly observed treatment comparatively expensive. In Malawi in 1995, a day in hospital cost $2.097 (13.3 times less than in Hlabisa), so 60 days in hospital would cost the health system $125.40. On the assumption that health clinic and community health worker visits and field worker and driver costs would also be 13.3 times less but that vehicle and fuel costs would be the same as those in Hlabisa, 17.5 days in hospital, 48 visits for directly observed treatment, and field worker supervision would cost $64.30 (36.60+6.10+21.60). Since the population density is higher in Malawi than Hlabisa (93 v 67 per square kilometre),19 the cost of field worker supervision might be cheaper still. Furthermore, the low cost of visits for directly observed treatment suggests that if input by non-health workers were unacceptable and the costs of directly observed treatment more than doubled, this strategy would still be cheap from the perspective of the health system.
For the patient, in contrast, the acceptability of non-health worker input is critical. When this is feasible, 48 visits for directly observed treatment will always be cheaper than 42.5 days in hospital because the patient can choose a supervisor who makes access to treatment convenient. Moreover, recent observation in Hlabisa indicates that the increased caseload secondary to the HIV epidemic may be best absorbed through supervision by volunteers.20 When using non-health workers is not acceptable and access to health facilities is limited, however, 48 visits for directly observed treatment will incur large costs and treatment in hospital could be preferable.
The likely effectiveness of directly observed treatment elsewhere is harder to assess. Nevertheless, a nearby district has found early success in implementing a similar programme (G Dean, personal communication); and the programme in Hlabisa has continued to prove highly effective even after management was no longer the responsibility of the programme innovator. Evidence also suggests that contacts of patients with tuberculosis treated in the community are not at increased risk of infection compared with those managed in hospital.21 22 23
Wherever the conventional approach is currently relied on, a switch to directly observed treatment will help to decongest hospitals and increase the capacity to cope with a growing tuberculosis caseload and a more widespread increase in demand for hospital care. Indeed, with the HIV related tuberculosis epidemic across Africa and limited extra resources for tuberculosis control programmes, it may become programme managers' only option.
We thank the staff at Hlabisa District Hospital for their helpfulness and Xoli Mfeka (research intern at the Centre for Health and Social Studies, University of Natal, Durban) for her help.
Funding: This research was done in partnership with and with support from the South African Medical Research Council and the KwaZulu-Natal Provincial Health Department. It was undertaken as part of the work of the HIV/AIDS research work programme at the Liverpool School of Tropical Medicine, Liverpool, which is funded by the Department for International Development. The department cannot accept responsibility for any information provided or views expressed.
Conflict of interest: None