Evaluating the use of methods to mitigate bias from non‐transient medications in the case‐crossover design: A systematic review

Abstract

AbstractPurposeThe case‐crossover design is a self‐controlled study design used to compare exposure immediately preceding an event occurrence with exposure in earlier control periods. The design is most suitable for transient exposures in order to avoid biases that can be problematic when using the case‐crossover design for non‐transient (i.e., chronic) exposures. Our goal was to conduct a systematic review of case‐crossover studies and its variants (case‐time‐control and case‐case‐time‐control) in order to compare design and analysis choices by medication type.MethodsWe conducted a systematic search to identify recent case‐crossover, case‐time‐control, and case‐case‐time‐control studies focused on medication exposures. Articles indexed in MEDLINE and EMBASE using these study designs that were published between January 2015 and December 2021 in the English language were identified. Reviews, methodological studies, commentaries, articles without medications as the exposure of interest, and articles with no available full text were excluded. Study characteristics including study design, outcome, risk window, control window, reporting of discordant pairs, and inclusion of sensitivity analyses were summarized overall and by medication type. We further evaluated the implementation of recommended methods to account for biases introduced by non‐transient exposures among articles that used the case‐crossover design on a non‐transient exposure.ResultsOf the 2036 articles initially identified, 114 articles were included. The case‐crossover was the most common study design (88%), followed by the case‐time‐control (17%), and case‐case‐time‐control (3%). Fifty‐three percent of the articles included only transient medications, 35% included only non‐transient medications, and 12% included both. Across years, the proportion of case‐crossover articles evaluating a non‐transient medication ranged from 30% in 2018 to 69% in 2017. We found that 41% of the articles that evaluated a non‐transient medication did not apply any of the recommended methods to account for biases and more than half of which were conducted by authors with no previous publication history of case‐crossover studies.ConclusionUsing the case‐crossover design to evaluate a non‐transient medication remains common in pharmacoepidemiology. Researchers should apply appropriate design and analysis choices when opting to use a case‐crossover design with non‐transient medication exposures.