Description of individual patient dataset

Our data generation relied on a real-life IPD-MA dataset that evaluated the associations between MDR-TB treatments and treatment success. Only patients with confirmed MDR-TB were included. Patients with extensively drug-resistant tuberculosis (XDR-TB) were excluded. We refer to this dataset as the MDR-TB-IPD. Ahuja et al. described in detail the study identification, selection and participation in IPD-MA [12].

The MDR-TB-IPD contained both individual-level and study-level information including patients’ age, sex, HIV status and acid fast bacilli (AFB) as well as smear status at the start of the MDR-TB treatment regimen, history of TB, cavitation status of the patient, information on 15 pharmacological agents, resistance status for each treatment, and the treatment outcome.

Specifically, the MDR-TB-IPD dataset is comprised of 31 studies with a total of 9290 patients. The number of patients per study varied from 25 to 2211, while the median number of patients across studies was 104 (see Table 1). The study-specific mean age of patients varied from 7 to 48 years, with a median mean age of 41 years across studies. Overall, most patients were males with the proportion of male ranging from 47% to 96% across studies. AFB smear status varied from 70% to 100%, while the median proportion of AFB smear positive across studies was 75%. Most patients were HIV-negative; indeed, 22 studies had no HIV-positive patients. Treatment prevalence for the drug “ethionamide” varied from 0 to 100% - because some treatments were not available in some study centres.

Characteristics of study participants by study and overall

N.(%): The number and percentage of patients in each study.

Median, P₂₅ and P₇₅ indicates the median, 25th and 75th Percentiles which have been calculated for the number of patients, mean age of patients, proportion of male patients, proportion of smear status, HIV status across studies, and proportion of treatment Ethionamide

We conducted a simulation study to evaluate the performance of propensity score matching-based approaches to estimate treatment effects in the context of MDR-TB-IPD.