The majority of evidence used by the World Health Organization (WHO) to inform guidelines for the treatment of multidrug-resistant tuberculosis (MDR-TB) is based on findings from observational cohort studies. Observational cohort studies have important limitations when interpreting estimates as causal effects, in contrast to randomized controlled trials (RCTs), the gold standard for assessing efficacy. Specifically, observational cohort studies are at greater risk of common threats to validity, such as selection bias and confounding. The consequences of residual bias in observational cohort studies of MDR-TB patients would be substantial, given WHO guidelines inform the treatment approach for the 500,000 patients estimated to fall sick with MDR-TB globally each year.
The goals of this dissertation are to: assess the comparative effectiveness of adding delamanid to MDR-TB regimens (Aim 1), scrutinize the potential for selection bias when using different approaches to defining the subcohort of MDR-TB patients eligible for studies using sputum culture conversion outcomes (Aim 2), and evaluate differences in the interpretation and estimates of the comparative effectiveness of adding delamanid (as explored in Aim 1) when the time-varying nature of MDR-TB treatment is and is not accounted for (Aim 3).
In all aims, we use data from the observational cohort of the endTB initiative. The endTB initiative was launched in 2015 to rapidly expand access to two new drugs for MDR-TB, bedaquiline and delamanid, for over 2,700 patients in 17 countries. In partnership with national TB programs, a consortium of non-governmental organizations leads the initiative: Partners In Health, Médecins Sans Frontières, and Interactive Research and Development. Participants are treated in accordance with guidelines of WHO and their respective countries under routine programmatic conditions. Study activities are directed by a common protocol, data are collected using standardized forms, and adverse events (AE) are monitored through a unified pharmacovigilance system, which facilitates data consistency across sites.
In Aim 1, we investigate whether adding delamanid to MDR-TB regimens comprised of three drugs likely to be effective improves two- and six-month culture conversion. We apply a censoring approach using inverse probability weighting that accounts for MDR-TB regimen changes over the course of treatment to estimate the observational analogue of the per-protocol effect. We did not identify a difference in two- or six-month culture conversion between participants with delamanid added to their regimen and participants without delamanid. We hypothesize that delamanid did not provide a contribution to effectiveness because regimens already contained multiple efficacious drugs (e.g. linezolid, moxifloxacin/levofloxacin, bedaquiline).
In Aim 2, we used simulated data and data from the endTB observational cohort to evaluate whether extending the allowable baseline sputum culture collection interval past treatment initiation is a source of selection bias in studies using culture conversion outcomes. Two of the most influential factors that increased bias were the proportion of the cohort with a missing pre-treatment culture and the occurrence of death and loss to follow up (LTFU) in this group. These occurred infrequently in the endTB observational cohort; thus we did not observe meaningful differences when the baseline culture definition was extended past treatment initiation. In cohorts with an excess of missing pre-treatment culture data and early non-conversion events such as death and LTFU, extending the allowable interval past treatment initiation may introduce bias. Investigators should scrutinize whether extending the baseline sputum culture collection interval will inadvertently exclude these patients who may have been eligible for inclusion had they had pre-treatment sputum culture data.
In Aim 3, we used the clinical research question from Aim 1 to investigate whether implementing inverse probability of censoring weights to account for the time-varying nature of MDR-TB treatment generated results that were different from those generated through baseline-adjusted analytic approaches. Results were similar using the two types of approaches. This similarity is likely a consequence of the relative modest frequency of regimen changes and the distribution of participants with regimen changes across exposure groups and the outcome. We hypothesize that estimates may differ meaningfully for research questions where treatment changes are highly concentrated in one exposure group and these treatment changes are highly associated with the outcome.
| Identifer | oai:union.ndltd.org:bu.edu/oai:open.bu.edu:2144/43000 |
| Date | 10 September 2021 |
| Creators | Rodriguez, Carly Alicia |
| Contributors | Horsburgh, Jr., C. Robert |
| Source Sets | Boston University |
| Language | en_US |
| Detected Language | English |
| Type | Thesis/Dissertation |
| Rights | Attribution-NoDerivatives 4.0 International, http://creativecommons.org/licenses/by-nd/4.0/ |
Page generated in 0.0024 seconds