Statistical Analysis

Analysis was based on the complete-case intention-to-treat (ITT) principle. Descriptive statistics including frequency and percentage were used to describe the health state of the study participants. Multiple bar charts with cross-tabulation were used to illustrate distributions of health states by study arms. χ² and Fisher exact tests were employed to compare the 5 EQ-5D-5L health domains by study arms. Both Kolmogorov-Smirnov and Shapiro–Wilk tests revealed that the distribution of EQ-5D-5L index value (ie, the health state utility) was not normal. Median values were used to summarize EQ-5D-5L index value. A nonparametric Mann-Whitney U test was employed to compare the difference in EQ-5D-5L index value among study arms. A log-binomial model was used to identify risk factors for lower HRQoL, which was having at least 1 health problem.

The overall TB treatment cost was estimated by considering costs related to anti-TB drug pick-up, guardian costs, and coping costs over the 2-month intensive phase. Both Kolmogorov-Smirnov and Shapiro-Wilk tests revealed that the distribution of overall TB treatment cost was non-normal, therefore, median values with IQR were used to summarize costs. A nonparametric Mann-Whitney U test was employed to compare cost differences among study arms. The proportion of study participants who faced catastrophic costs at a cut-off point of 20% was estimated. A cross-tabulation was employed to evaluate the distribution of catastrophic cost over the study arms and a χ² test was used to test the association between catastrophic cost and study arms.

The sample size was calculated considering a 2-sided type I error of 5%, a power of 80%, 20% attrition rate, 20% noninferiority margin, a delta of 0.63, and a continuous outcome of percentage adherence over the 2-month intensive phase, with a standard deviation of 41% and 79% of average adherence. The results yielded a sample size of 57 in each arm for a total of 114 participants.