A systematic, concept‐based method of developing the exposure measure for drug safety and effectiveness studies

Abstract

Purpose: In drug safety and effectiveness studies based on secondary data, the choice of an appropriate exposure measure for a given outcome can be challenging.Different measures of exposure can yield different estimates of treatment effect and safety. There is a knowledge gap with respect to developing and refining measures of drug exposure, to ensure that the exposure measure addresses the study question and is suitable for statistical analysis.Methods: We present a transparent, step-by-step approach to the development of drug exposure measures involving secondary data. This approach would be of interest to students and investigators with initial training in pharmacoepidemiology. We illustrate the approach using a study about Parkinson's disease.Results: We described the exposure specifications according to the study question.Next, we refined the exposure measure by linking it to knowledge about four major concepts in drug safety and effectiveness studies: drug use patterns, duration, timing, and dose. We then used this knowledge to guide the ultimate choice of exposure measure: time-varying, cumulative 6-month exposure to tamsulosin (a drug used to treat prostate hyperplasia). Conclusions:The proposed approach links exposure specifications to four major concepts in drug safety and effectiveness studies. Formulating subject-matter knowledge about these major concepts provides an avenue to develop the rationale and specifications for the exposure measure.

Handling time‐varying treatments in observational studies: A scoping review and recommendations

“…Most reports (31, 86.1%) dealt with issues about statistical methods, with the primary aim to address time‐varying confounding affected by past treatments, 3,7–9,16,17,19–33 indication/selection confounding, 23,25,26 inefficient statistical power, 22,34,35 and selection bias by nonrandomized censoring 22,26,34 . Fourteen (38.9%) discussed epidemiological designs which primarily tackled immortal time bias 26,36–41 …”

“…22,26,34 Fourteen (38.9%) discussed epidemiological designs which primarily tackled immortal time bias. 26,[36][37][38][39][40][41] The discussions about biases were outlined in Table 2. We categorized these sources of bias into two types in the context of time-varying treatment.…”

“…The ITT approach carries forward the treatment of interest without factoring the actual treatment change over time. 7,17,21,22,31,34,37,40,41 The standard ITT approach in observational studies only considers the treatment status of baseline and ignores subsequent treatment change. As there is no time-varying treatment, conventional survival analysis methods including Cox regression, Kaplan-Meier estimator are applicable for data analysis.…”

“Take up to eight tablets per day”: Incorporating free‐text medication instructions into a transparent and reproducible process for preparing drug exposure data for pharmacoepidemiology

“…The application of algorithms in this paper represents extension of previous work to clarify sequential steps taken to prepare drug exposure data from EHRs 8 . It adds to a body of work aimed at increasing transparency in decisions required to refine measures of drug exposure, prior to embarking on statistical analysis 25,26 . The current algorithm does not derive drug dose, which may be especially important in studies where is a likely dose‐dependent effect on the outcome.…”

Considerations for defining medication exposure when analyzing real‐world data

“…Although a systematic and concept‐driven approach has been developed for developing exposure measures in drug safety and effectiveness studies, 7 there is currently no established framework for conceptualizing or implementing medication use. To address this gap, we have created a structure that outlines the considerations that need to be considered when developing medication use definitions.…”