Sputum culture conversion definitions and analytic practices for multidrug-resistant TB

Abstract

Dear Editor, Conversion of sputum culture from positive for Mycobacterium tuberculosis to negative is commonly used to monitor treatment response for rifampinand multidrug-resistant TB (RR/MDR-TB).1,2 Sputum culture conversion is also used as a surrogate or proxy for final outcome in observational cohort studies.3,4 Analyses of culture-based endpoints in observational cohorts requires analytic decisions that can potentially introduce bias and affect study interpretation. For example, only patients with a positive baseline culture are included in conversion analyses, requiring investigators to define ‘‘baseline’’, i.e., the allowable interval before and/or after treatment initiation a positive culture result is considered for inclusion. Other examples include the operationalization of the conversion definition and the handling of deaths and losses to follow-up (LTFU) occurring prior to conversion. Variability in definitions and analytic practices across studies, and in their reporting, hinders interpretation, assessment of potential bias, and comparisons across studies. We sought to describe recently published definitions and analytic practices in RR/MDR-TB observational cohorts using culture-based endpoints within three key domains: 1) defining ’’baseline’’ cultures, 2) defining and analyzing conversion endpoints, and 3) reporting and analyzing deaths and LTFU occurring prior to conversion. We searched PubMed/MEDLINE for observational cohorts of RR/MDR-TB patients reporting a culture-based endpoint published between January 1st, 2015 and December 31st, 2020. We focused our search on observational cohorts because, although trials are not immune to bias, factors such as pre-treatment inclusion criteria and enforcement of standardized protocols minimize some opportunity for hypothesized biases. We additionally excluded systematic reviews because they would not include methodologic details of individual studies, studies that did not include a culture conversion endpoint, reviews or case reports, and studies on drug-susceptible TB. We selected a search period spanning 2015–2020 because our objective was to identify recent practices in MDR-TB cohort studies and because the use of culture-based endpoints has increased in recent years. We used search terms related to RR/MDR-TB and culture-based endpoints (e.g., culture-conversion, culture conver*) and searched 15 major international clinical and epidemiology journals. A single reviewer abstracted study characteristics, including the objective, whether the study was a retrospective cohort with routinely collected medical records or a prospective cohort with a standard protocol, study setting and size, and the type of culture-based endpoint. Reviewers used a standardized data collection form and met regularly to discuss studies to improve consistency across reviewers. For each domain, we abstracted definitions and analytic practices used and provided relevant quantitative metrics. Our search strategy yielded 107 publications. We excluded 47 that were results of randomized trials or published study protocols (n1⁄4 15), systematic reviews or meta-analyses (n1⁄412), did not report a culture-based endpoint (n1⁄4 8), were not on RR/MDR-TB (n1⁄4 5) or did not have original cohort data (n1⁄4 7). We reviewed the full text and extracted data from 60 published studies. Seventeen studies (28%) reported on prospective observational cohorts; 43 (72%) were retrospective medical record reviews. Approximately a third of studies reported on ,100 patients (Table). Only 4 (7%) studies defined the ‘‘baseline’’ interval for sputum collection before or after treatment initiation. In these, the allowable interval ranged from 6 months pretreatment to 30 days post-treatment initiation. The proportion without a positive baseline culture (e.g., due to a negative result, contamination, or missingness) was reported in 26 (43%) studies and comprised a median of 18% of the cohort. Forty-five studies (72%) defined the culture-based endpoint used in the study: most commonly (n 1⁄4 21, 35%), this was conversion established by two negative cultures 30 days apart; four additional studies required patients could not have a subsequent positive culture. Six studies required only one negative culture to establish conversion. The timefixed absolute proportion with conversion (e.g., 6 months) and relative survival probability of conversion (i.e., hazard ratio) were the most common outcomes, each reported in 48% of publications. Only 14 (23%) studies reported how deaths occurring prior to culture conversion were analyzed. In 4 (7%), authors treated deaths as censoring events in time-to-conversion analyses; in 4 (7%), patients who died were excluded; and in 5 (8%) patients who died prior to conversion were considered not to have converted. Approaches were similar for LTFU. Only 2 (2%) studies conducted sensitivity analyses to determine whether a different approach changed results.