Community-based tuberculosis treatment is likely to be necessary for successful global implementation of the World Health Organization‚s DOTS strategy . It may be operationally more feasible to give treatment twice each week, especially when treatment is given in the community , and so it is important to determine fully the effectiveness of intermittent therapy.
Twice weekly therapy can be delivered under direct observation (DOT) in rural South Africa  in a sustained  and highly cost-effective  manner. A drug regimen containing rifampicin throughout and given twice weekly under DOT in the community, with supervision largely by volunteers, was curative . In this study, 416 new and 113 previously treated adults with culture-positive pulmonary tuberculosis (58% HIV infected, 9% combined drug resistance) were assessed for treatment effectiveness. Adherence was defined as taking 85% or more of the prescribed therapy; 83% of the cohort achieved this. Cure among adherent HIV-infected and uninfected new patients, and among previously treated patients, was 96-97% .
We have continued to follow this cohort and here we report post-treatment relapse and mortality rates, stratified by HIV status and by previous treatment history, and we consider the predictive value of a positive smear at 2 months for relapse and death.
The study was done in the Hlabisa health district of KwaZulu/Natal, South Africa. The district is largely rural and is home to approximately 210000 Zulu-speaking people. The economy is based on migrant worker remittances, pensions and subsistence farming. Per capita income was US$1730 in 1994, the literacy rate was 69% and life expectancy was 63 years (Economic Research Unit, University of Natal). The control programme treats all people with tuberculosis in the district. All tuberculosis suspects are admitted for diagnosis and initiation of therapy. Treatment is given daily as an inpatient for 2-3 weeks, patients being discharged when clinically stable and a treatment supervisor has been identified. Approximately 90% of patients are discharged  and they then receive twice-weekly treatment, given as DOT in the community, to complete 6-month‚s therapy. Patients are supervised by a nurse in a community clinic or by a paid community health worker (45%) or a community volunteer (55%), who is usually a store-keeper .
We have previously reported the results of the prospective cohort study carried out to determine the effectiveness of twice-weekly DOT for tuberculosis in HIV-infected and uninfected patients, irrespective of previous treatment history . While hospitalized, patients received daily treatment with isoniazid (H), rifampicin (R), pyrazinamide (Z) and ethambutol (E) according to weight: under 50kg, H 250mg, R 450mg, Z 1500mg and E 800mg; over 50kg H 300mg, R 600mg, Z 2000mg and E 1200mg. In the community, twice-weekly therapy was given: <50kg H 800mg, R 600mg, Z 3000mg and E 2000mg; >50kg H 900mg, R 600mg, Z 3500mg and E 2500mg. These four drugs were continued until 2 months, after which H+R was continued in the same dosage. Fixed-dose combination capsules were not available. Adherence was monitored through direct observation of treatment by supervisors who recorded every dose taken on a treatment card, and this was supplemented by monthly visits to supervisors by programme staff. Adherence was deemed inadequate if eight or more consecutive doses (4 weeks) were missed, and the patient was then classed as ‚treatment interrupted‚. Treatment completion was defined as taking 85% or more of prescribed drugs within 7 months of starting treatment.
In line with recommendations for evaluation of new tuberculosis drug regimens , cure was defined as a negative sputum culture at the end of treatment. Here we report post-treatment mortality and relapse. We followed all cured patients by monitoring hospital readmission for suspected tuberculosis and by home visits every 6 months to cured patients. Relapsed tuberculosis was defined by a positive culture for Mycobacterium tuberculosis on a single sputum sample; death was defined as any cured patient who died during follow up; patients were classed as ‚left the area‚ if after two home visits they could not be found and if relatives reported they had left the area. At each home visit patients were questioned about symptoms, especially cough, and were asked to produce a sputum sample if possible. If asymptomatic, or if the sputum culture was negative, we classified that patient as not having tuberculosis.
Culture and susceptibility testing for mycobacteria was done in the Department of Medical Microbiology, University of Natal. Sputum was liquefied using NALC-NaOH, concentrated by centrifugation for 20min at 3000×g and cultured on Lowenstein-Jensen medium and Middlebrook 7H11 agar as previously described . Susceptibility testing was carried out according to the 1% proportional method on Middlebrook 7H10 agar at the following concentrations: isoniazid 0.2 and 1.0mg/l, rifampicin 1.0mg/l, streptomycin 2.0 and 10.0mg/l and ethambutol 7.5mg/l. Pyrazinamide susceptibility was determined according to the Bactec broth method (100 mg/l: pH 6.0) for all isolates resistant to any of the other drugs.
In line with our analysis plan , we calculated relapse and mortality rates per 100 person-years of observation (PYO) under follow-up. Each individual‚s time under observation was defined as the time from documented cure to the last home visit, or death, or diagnosis of relapse. This approach ensures a participant contributes appropriately to both numerator and denominator. The 95% confidence intervals (CI) of PYO relapse and mortality rates were calculated assuming a Poisson distribution.
Survival curves were constructed using the Kaplan-Meier method. Results were stratified by HIV status and by previous treatment history and compared with the log rank test. Statistical significance was defined as P<0.05. Data analysis was done using SAS statistical software (SAS Institute Inc., Cary NC, USA).
The 403 patients were followed for a mean (SD) duration of 1.2 (0.4) years, equivalent to a total of 499 PYO. Of these, 215 patients (53%) were HIV infected, 175 (43%) were HIV uninfected and in 13 (4%) HIV status was unknown. Total PYO was similar among the HIV-infected (259 years) and HIV-uninfected (225 years) patients.
Over the follow up period, 78 patients (19%) left the area, 58 (14%) died, in 248 (62%) no active tuberculosis was diagnosed and 19 patients (5%) relapsed. Fifteen relapses were identified through presentation at hospital, and four were identified during home visits. Of the 248 patients in whom no active tuberculosis was diagnosed, 190 were asymptomatic when followed up and 58 produced sputum that did not grow M. tuberculosis.
Relapse and mortality post-treatment
Among the whole cohort, the relapse rate was 4 per 100 PYO (95% CI 2.3-5.7), and the mortality rate was 11.6 deaths per 100 PYO (95% CI 8.6-14.6). The rate of loss to follow-up was similar among HIV-infected and HIV-uninfected patients (15.8 and 16 patients per 100 PYO, respectively).
The relapse rate was similar among HIV-infected patients [3.9 per 100 PYO (95% CI 1.5-6.3)] and HIV-uninfected patients [3.6 per 100 PYO (95% CI 1.1-6.1)] (P=0.7). The probability of relapse at 18 months (Fig. 1) was estimated as 5% in both groups.
However, mortality was increased fourfold among the HIV-infected patients: 17.8 and 4.4 deaths per 100 PYO for HIV-infected and uninfected, respectively (P<0.0001). The probability of survival at 18 months was estimated as 76 and 96%, respectively, and at 24 months was estimated as 59 and 81%, respectively (Fig. 2).
Effect of previous treatment history on outcome
Of the 390 patients with known HIV status, 83 (21%) had been previously treated for tuberculosis. The proportion lost to follow-up was similar among those with a history of previous treatment (23%) and those with no such history (18%) (P=0.3).
In all, three of the 83 previously treated patients (4%) relapsed, while 15 of the 307 with no previous history of tuberculosis treatment (5%) relapsed. The rate of relapse was similar in those previously treated (3 per 100 PYO) and those not previously treated (4 per 100 PYO) (P=0.4).
A total of 10 previously treated patients died post- treatment (12%) compared with 46 (15%) of those not previously treated (P=0.3). The mortality rate among those previously treated was 3 deaths per 100 PYO, compared with 4 deaths per 100 PYO in those with no history of previous treatment.
Predictive value of a positive smear at 2 months for adverse treatment outcome
The sensitivity and positive predictive value of a positive smear at 2 months for mortality, for relapse, and for these two adverse outcomes combined were both poor (Table 1). None of the few patients with a positive smear at this stage died post-treatment, and none of the patients who died had a positive smear at 2 months.
While the proportion of those with a positive smear at 2 months that relapsed (2/12, 16%) was higher than the proportion of those with a negative smear at 2 months that relapsed (16/370, 4%), this difference was not statistically significant (P=0.4).
The data described in this study demonstrate an acceptably low relapse rate among patients cured following twice weekly rifampicin-containing DOT, irrespective of HIV status and previous treatment history. We confirm the excess mortality experienced by HIV-infected patients following successful tuberculosis treatment, further reinforcing the need for interventions to reduce this. We also show that having a positive sputum smear at 2-3 months during treatment does not predict relapse or mortality post-treatment.
Perriens et al. tested a similar regimen in Zaire  using HRZE daily for 2 months (DOT 6/7 days each week) and HR twice weekly for 4 months (1 dose each week of the twice weekly doses was given under DOT). They followed 119 HIV-infected and 180 HIV-uninfected patients and estimated relapse rates at 18 months of 9 and 5.3%, respectively (P=0.06). In our sample, we estimated a relapse rate of 5% in each group at 18 months. Perriens et al. also reported substantially increased mortality among HIV-infected patients post-treatment, at rates very similar to ours. Although our estimate of relapse at 18 months (5%) is lower than that of Perriens et al. (9%), this difference is not statistically significant. Therefore, the effectiveness of the two drug regimens is similar.
A relapse rate of 5% is regarded as ‚acceptable‚ . Most studies of effective short-course regimens  included fewer than 300 patients, and many included less than 100 . These studies quote relapse rates of 0-4%, but with relatively small sample sizes, the CI around these point estimates frequently exceeds 5%. Our results, taken with the data reported on cure , therefore, fit the criteria necessary for an effective short-course regimen .
Although taking almost all doses of an intermittent regimen is probably important, our analysis included some patients who had taken only 85% of prescribed therapy, and had taken up to 7 months to do so. There does seem to be at least some flexibility even within a twice-weekly regimen. Furthermore, although none of our patients had multidrug-resistant disease, combined drug resistance was present in 9%. Our findings support those of others  who show that initial drug resistance has little impact on treatment outcomes.
We treated patients with a history of previous tuberculosis treatment with a regimen usually reserved for new patients. There was no difference in relapse or mortality post-treatment at 6 months despite using the same regimen in both groups and despite ignoring positive sputum smears at 2-3 months . Our sample size does not allow us to conclude equivalent effectiveness in the two groups. However, we can be 95% confident that relapse among those previously treated is no greater than 15% (5% in new patients), and that mortality is no greater than 25% (15% in new patients). Loss to follow-up in our study (19%) is probably acceptable for a cohort study in rural Africa, especially in a highly migrant community. However, this rate of loss does mean that relapse and mortality may be higher than our estimates if patients lost to follow-up suffer higher rates of relapse and mortality.
We cannot be sure whether the episodes of relapse we observed were reactivation of inadequately treated disease or reinfection with a new strain. We have shown that recent transmission probably accounts for 29-43% of cases in this setting , so it is likely that several of the relapses are new infections. We are currently analysing our stored specimens with DNA fingerprinting.
It is perhaps not surprising that a positive smear at 2-3 month is not predictive of relapse or mortality. Mortality seems to be dominated by HIV-related disease  and not by failed tuberculosis treatment. None of our patients had a post-mortem so we are unsure how many had undiagnosed tuberculosis. As noted above, relapse is likely to be, at least in part, caused by reinfection, and it is difficult to see how a positive smear during therapy could predict that.
Since 1991 , we have shown that more than 90% of patients with tuberculosis in this part of rural South Africa can be treated in the community following a short hospital stay . Most community supervisors are volunteers, and as caseload increases secondary to the HIV epidemic it seems that volunteers may be able to oversee patients better than health workers . Because this approach to tuberculosis treatment keeps patients out of hospital, it is highly cost effective . We have now shown, in a rigorous and large cohort study, that the drug regimen used is curative  and has an acceptably low relapse rate. The challenge for African tuberculosis control programmes is to develop models of community care  that ensure adherence. These programmes should also be integrated into the district health system so that improved HIV care can be provided  to reduce the unacceptably high HIV-associated mortality both on- and post-treatment.