Chinese Evidence-based Medicine Center and Cochrane China Center, West China Hospital, Sichuan University, Chengdu, Sichuan, China, China (People's Republic)
Background: Handling time-varying exposure is a challenging issue for assessing treatment effect especially in observational studies. In presence of time-varying exposure, valid effect estimation involves well-defined objectives, rigorous design, and proper analysis. However, existing methodological studies often exclusively address analysis models without a holistic perspective.
Objectives: This study aims to propose an evidence-based framework on how to deal with time-varying treatment in pharmacoepidemiology for observation study in order to address treatment effect.
Methods: We first identified critical domains, objective, design and analysis, to handle time-varying treatment and developed a framework regarding the elements of objective, design and analysis. Then we systematically reviewed the methodological study related to time-varying treatment from 2000 to 2021. Information regarding general study characteristics, research objectives, study design, analysis strategies were collected and summarized.
Results: There were 36 publications met the inclusion criteria, among which 23 (63.9%) were reviews, 10 (27.8%) were original articles. As for research objective, 75% of publications had stated treatment effect of interest, while the remaining 25% didn’t specify the target treatment effect. There is no article that clearly sorts out the category of target treatment effect in presence of time-varying exposure which is fundamental for subsequent study design and analysis strategies. As for study design, key components of a study, including PECOT (population, exposure, control, outcome, and follow-up time), should be well-defined. The analysis domain is most studied which is discussed by 24 publications (66.7%). In this study, the analysis strategy contains two parts, dataset preparation and statistical methods. Analogous to trials, datasets were classified as intention-to-treat (ITT), Per-Protocol (PP) and As-treated (AT). Only one study mentioned how to the construct dataset. Based on dataset, corresponding methods were selected. Methods for AT analysis was most studied, which were divided into three main categories: time-dependent regression, longitudinal matching and G-methods. Each category of method has different application scenarios.
Conclusions: There is no standardized process for addressing treatment effect in pharmacoepidemiology for observational studies with time-varying exposure, and some key concepts are not well-defined. This study hence sorts out the treatment of interest and corresponding analysis method, then proposed a basic framework and reference workflow to improve the future work dealing with time-varying exposure.
