Tat-Thang Vo: Department of Statistics and Data Science, The Wharton School, University of Pennsylvania, Philadelphia. Electronic address: tatthang@wharton.upenn.edu.
Aidan Cashin: Centre for Pain IMPACT, Neuroscience Research Australia, Sydney, Australia.
Cecilia Superchi: CRESS, INSERM, INRA, Université de Paris, Paris, France.
Pham Hien Trang Tu: Department of Family Medicine and Population Health, University of Antwerp, Antwerp, Belgium.
Thanh Binh Nguyen: Department of Pharmacy, Vietnam National Cancer Hospital, Hanoi, Vietnam.
Isabelle Boutron: CRESS, INSERM, INRA, Université de Paris, Paris, France.
David MacKinnon: Department of Psychology, Arizona State University, Tempe.
Tyler Vanderweele: Departments of Epidemiology and Biostatistics, Harvard School of Public Health, Harvard University, Boston.
Hopin Lee: Centre for Statistics in Medicine, Nuffield Department of Orthopaedics, Rheumatology and Musculoskeletal Sciences, University of Oxford, Oxford, United Kingdom.
Stijn Vansteelandt: Department of Applied Mathematics, Computer Science and Statistics, Faculty of Science, Ghent University, Ghent, Belgium.
OBJECTIVE: To describe the bias assessment practice in recently published systematic reviews of mediation studies and to evaluate the quality of different bias assessment tools for mediation analysis proposed in the literature. METHOD: We conducted an overview of systematic reviews by searching MEDLINE (OvidSP), PsycINFO (OvidSP), Cochrane Database of Systematic Reviews (OvidSP), and PubMed databases for systematic reviews of mediation studies published from 2007 to 2020. Two reviewers independently screened the title, abstracts, and full texts of the identified reports and extracted the data. The publications of all mediation-specific quality assessment tools used in these reviews were also identified for the evaluation of the tools' development and validation. RESULT: Among 103 eligible reviews, 24 (23%) reviews did not assess the risk of bias of eligible studies, and 48 (47%) assessed risk of bias using a tool that was not specifically designed to evaluate mediation analysis. 31 (30.1%) reviews assessed the risk of mediation-specific biases, either narratively or by using specific tools for mediation studies. However, none of these tools were consensus-based, rigorously developed or validated. CONCLUSION: The quality assessment practice in recently published systematic reviews of mediation studies is suboptimal. To improve the quality and consistency of risk of bias assessments for mediation studies, a consensus-based bias assessment tool is needed.
