In spite of revolutionary advance in treatment for rifampicin- and multidrug-resistant (RR/MDR)
tuberculosis (TB), not all patients experience treatment success. Existing literature on other risk
factors for unfavorable TB treatment outcomes is limited in at least two important ways that
preclude an actionable evidence-based response. First, analyses are typically conducted using
predictive modelling or multivariable regression techniques that do not consider the underlying
causal structure of the factor and TB outcome. These approaches are inadequate for informing
interventions because they are not designed for causal inference and can lead to results that are
biased, and/or uninterpretable. Second, analyses of pre-treatment (i.e., baseline) factors do not
elucidate the mediating mechanism through which an unfavorable treatment outcome occurs. For
example, hepatitis C infection may lead to unfavorable tuberculosis treatment outcomes through
multiple pathways, including advanced liver disease that reduces bioavailability, as well as
suboptimal adherence to TB treatment. An effective adherence intervention would be expected
to improve outcomes under the latter scenario, but not the former. Therefore, analyses elucidating
the extent to which treatment adherence mediates the relationship between a given factor and
treatment outcomes will be required to inform relevant interventions for the patients that need
them most, thereby ensuring that all patients with TB benefit fully from the latest scientific
advances. We propose to elevate existing analyses of factors that lead to unfavorable TB
treatment outcomes through a novel framework for analyzing causes of unfavorable TB treatment
outcomes; one that is grounded in causal inference and evaluates adherence as a potentially
intervenable pathway. Our team of experts in TB, treatment adherence, and epidemiologic
methods will leverage a unique longitudinal dataset of 2,789 patients receiving MDR/RR-TB
treatment under routine program conditions in one of 17 countries on five continents. The overall
goal of the proposal is to generate more valid, relevant, and actionable data with regard to TB risk
factors, with the overall goal of informing interventions to reduce TB-related morbidity and
mortality. Specifically, we will (1) estimate the causal effect of key baseline exposures (e.g., HIV,
hepatitis B and/or C, low body mass index, substance use, cavitary disease with a highly positive
sputum smear, age, and gender) on end-of-treatment outcomes; and (2) determine the extent to
which each causal effect is mediated by suboptimal adherence to treatment, and examine other
potential mediating pathways.
