Initially, the study was designed as a cohort study comparing treatment outcomes for patients who chose home-based treatment with those who chose health facility-based treatment. This design was based on the results from a community assessment in which patients, if given a choice, were asked what would be their preference for the place of treatment. In this assessment, just over 50% would have opted for home-based treatment . During the implementation of the current study it became clear that a much higher proportion of patients opted for home-based treatment, making a formal comparison of treatment outcomes with patients opting for health facility-based treatment statistically more difficult.
The study design was changed into a comparison of treatment outcomes for patients in the PCT cohort with those in a control cohort of all registered new TB patients in the same facilities during the same period of time a year earlier. This also changed the intervention being tested because three variables in the PCT cohort were different from the control cohort; the possibility of having home-based treatment observation, treatment with fixed dose combinations (FDCs) and a regimen with rifampicin for the continuation phase resulting in 6 months duration instead of 8 months. Possible differences between the cohorts should, therefore, be attributed to the overall intervention of PCT, rather than only the home-based setting of treatment observation.
Study population and setting
The study was carried out in the Arusha Municipality, the Kahama district (Shinyanga province) and the Mufindi district (Iringa province) of Tanzania. The decision to choose these sites was based on the number of notified smear-positive TB-patients and their representativeness for urban and rural settings in the country. Within each area, all TB treatment facilities implemented PCT for all new patients who were registered in the second and third quarter of 2006. The only inclusion criterion used was that the patient was defined as 'new', indicating that he/she had never received any TB-treatment that lasted longer than 1 month. The control cohort comprised all new patients in the same health facilities registered in the second and third quarter of 2005 who were treated under the conventional DOTS strategy. Smear-positive TB was recorded when two out of three sputum samples (spot, morning, spot) were Ziehl-Neelsen positive. When clinical symptoms were suggestive of TB, but sputum samples were negative, a suggestive chest X-ray could lead to the diagnosis of smear-negative TB.
The intervention tested was PCT, which consisted of three components. First, each patient was given the choice to receive treatment at home observed by a supporter of his/her choice or to receive daily treatment at the health facility observed by health staff. There were no restrictions on the choice of supporter. Second, treatment for all patients was delivered as a FDC rather than the conventional separate tables for each drug. Third, the treatment contained isoniazid, rifampicin, ethambutol and pyrazinamide in the initial phase of 2 months and isoniazid and rifampicin in the continuation phase of 4 months. At any time, patients were allowed to change from home-based treatment to health facility-based treatment or vice versa. There were no interventions related to default tracing and monitoring in the PCT strategy, other than the routine guidelines from the NTLP, making the control cohort and the PCT cohort comparable in this respect.
Follow-up and data collection
If the patient opted for home-based treatment, he/she was asked to return with the supporter of his choice. The supporter was given instruction by the health care provider on the importance of daily supervision of drug intake, the signs and symptoms of side-effects, and what to do if they occur, and the frequency of the collection of new drugs. Supporters needed to escort the patient to the health facility on a weekly basis in the first 2 months of treatment to collect new drugs, report on the well-being of the patient and to discuss any problem encountered in the support of the patient. The patient was requested to join the supporter at every visit unless too ill to do so. In the remaining 4 months, the visits to the health facility took place twice a month.
Data on the demographics of patients and supporters, drug intake, side effects and laboratory results in the PCT cohort were prospectively recorded in specifically designed registers and cards. Patient and laboratory data in the control cohort were retrospectively retrieved from the TB registers and the patient's treatment cards in the participating health facilities. These routine registers were checked for accuracy (and updated if needed) by the same independent team that collected data in the PCT cohort in order to minimize ascertainment bias. Data collection for both cohorts took place in three rounds from September 2006 to July 2007. Follow-up ended when the last included patient reached the time of treatment completion (April 2007). This meant that the last data collection took place after the treatment outcome of all patients had been recorded.
Outcome measures and explanatory variables
The primary study outcome was the proportion of new smear-positive patients cured at the end of treatment (6 months in the PCT cohort and 8 months in the control cohort). Secondary outcomes were the proportion of patients with treatment success (cured or treatment completed) at 6 months or 8 months, and the proportion of smear-positive patients with smear conversion at 2 months. In addition, we assessed whether the characteristics of the treatment supporters were associated with treatment outcomes in the group of patients with home-based treatment observation. Cure was defined as a smear-positive patient having a negative sputum smear at the end of treatment and on at least one previous occasion. Treatment completion was defined as a patient having completed treatment but not having a final smear examination. The characteristics of the supporters were categorized based on distribution with the purpose of getting groups of similar size. Age in years was defined as <25, 25-34, 35-44 and ≥45. The relationship with the patient was categorized as 'family member' or 'non family member'. Education was categorized as 'none', 'primary' and 'above primary'. The initially collected variable 'household of supporter' was not used due to the strong collinearity with the variable 'relationship' in the analyses.
All sputum smears during diagnosis and follow-up were confirmed by an independent laboratory technician. Discrepancies between the initial test and the re-reading were resolved by a third reader whose results were final.
Baseline characteristics of the two cohorts were compared using the χ
2 test for binary variables and Student's t-test for continuous variables. Outcome measures in the cohorts were compared by calculating the risk ratio (RR) stratified by district. A combined analysis of all three districts was only performed when there was no effect modification by district as assessed by the Cochran-Mantel-Haenszel test.
Supporter characteristics associated with cure and treatment success in the group of patients with home-based treatment observation were assessed by univariable and multivariable logistic regression. All multivariable analyses included, apart from 'place of treatment' the variable 'district' to incorporate the stratified design in the analyses. Other variables were only included if they were significantly associated with the outcome in univariable analyses with a P-value of 0.1 or less. Effect modification by districts in these models was assessed by testing for interactions (multiplicative). Effect modification was assumed statistically significant at the 10% level.
Missing data and loss to follow-up
Due to an unforeseen logistical problem, the cards designed for recording information on the supporters were not available for the first 205 patients. Unfortunately, not all information could be retrieved at a later stage. This did not introduce a bias as the study started in all the three districts at the same time. Furthermore, the patients' freedom of choice for the type of treatment delivery strategy remained. The missing cards only reduced the power of the study for the analyses assessing the effects of supporter characteristics on treatment outcomes in the PCT cohort.
The missing data were categorized as a separate level in the respective variables in order to include all treatment outcomes in the analyses. Therefore, statistical significant associations were assessed on level the Wald test in the logistic regression models rather than the likelihood ratio test.
Based on the total number of new smear-positive patients who opted for home-based treatment and the corresponding number of patients a year earlier, the power of the study for the primary outcome was an 80% ability to detect a statistically significant difference (α = 5%) of 8.5% when cure in the control cohort was estimated at 70%.
The study was approved by the Institutional Review Board of the Ifakara Health Institute in Tanzania. Informed consent was not obtained from the participants as the intervention was based on a change in national treatment guidelines that was applied to all.
Role of the funding source
The funder of the study was involved in the study design and data interpretation. Data collection and data analysis were performed independently of the funder. The corresponding author had access to all data in the study. The final decision to submit for publication was made by the NTLP of Tanzania.