eSupplementary data for Manuscript entitled Understanding Bias in Topical Steroid Withdrawal Studies Through Causal Diagrams

Reviews, Delphi consensus and Perspective pieces were excluded because they represent opinion pieces and/or reviews of the primary studies.</li><li>Technical directed acyclic graph illustrating the complex longitudinal relationship between topical corticosteroid exposure (TCS) and topical steroid withdrawal (TSW) in clinical practice.</li></ol><p dir="ltr"><b>Figure 1*. </b>Directed Acyclic Graph (DAG) illustrating the complex longitudinal relationship between topical corticosteroid exposure (TCS) and topical steroid withdrawal (TSW) in clinical practice.

TCS exposure and disease severity are represented as time-varying variables, with disease severity influencing subsequent TCS exposure and TSW-related symptoms, while prior TCS exposure may itself influence subsequent disease severity. This treatment–confounder feedback creates a time-dependent causal structure that complicates the causal estimand of TCS exposure on TSW. The pink arrows denote open biasing pathways, encompassing conventional and time-varying confounding as well as pathways related to exposure and outcome measurement and ascertainment; therefore, not all pink pathways represent confounding in the strict causal sense.

The DAG distinguishes true TCS exposure from recorded exposure and the putative TSW phenotype from patient symptoms, symptom reporting, and observer-recorded or diagnosed TSW, highlighting potential exposure and outcome misclassification. The latent “Unmeasured determinants of TCS exposure and TSW” represents unmeasured common causes of exposure and outcome and therefore a potential source of unmeasured confounding. Collectively, the DAG illustrates how observed associations between TCS exposure and TSW may arise from causal effects and multiple sources of bias, including measured and unmeasured confounding, time-varying confounding, measurement error, and diagnostic or ascertainment processes.</p><p><br></p><p dir="ltr">*The directed acyclic graph was parsed and evaluated using DAGitty (<a href="dagitty.net/dags.html" target="_blank">dagitty.net/dags.html</a>) and was subsequently re-drawn on Adobe Illustrator version 30.8.1 to improve visual presentation without altering its causal structure.</p>