Assessing the usability of ACROBAT-NRSI for studies of exposure and intervention in environmental health research

Article type
Authors
Morgan RL1, Thayer K2, Guyatt G1, Blain R3, Eftim S3, Ross P3, Santesso N1, Holloway AC4, SchĂĽnemann HJ1
1Department of Clinical Epidemiology and Biostatistics, McMaster University, Canada
2National Toxicology Program Office of Health Assessment and Translation, National Institutes for Environmental Health Sciences, USA
3ICF International, USA
4Department of Obstetrics and Gynecology, McMaster University, USA
Abstract
Background: Recently, A Cochrane Risk Of Bias Assessment Tool (ACROBAT) was released to examine internal validity (risk of bias) in non-randomized studies of interventions (NRSI). The applicability of ACROBA- NRSI to studies dealing with exposures or interventions in environmental health has not yet been explored.
Objectives: This study evaluated the application of ACROBAT-NRSI in studies of environmental health exposure.
Methods: Two independent reviewers sequentially applied ACROBAT-NRSI to two systematic reviews containing 14 and 17 individual studies of environmental exposures. Material available to reviewers included instructions for application of the ACROBAT-NRSI, including a list of possible confounders specific to the exposures considered in the two reviews. After completing the first and second assessments with the tool, reviewers provided feedback regarding clarity of the instrument and ease of application, highlighting barriers to use. Modifications to enhance usability for studies of exposures (rather than interventions) were made to the tool.
Results: Based on feedback from reviewers, two modifications were required:
1. replacement of 'intervention' with 'exposure' throughout the document; and
2. provision of additional explanation and examples for application of the tool to studies using a cross-sectional design.
Modification to the tool increased reviewers’ understanding during the second application; however, disagreements occurred when judgments on domain-level risk of bias were made, specifically in domains assessing bias due to confounding and bias in the selection of the reported result.
Conclusion: Modifications to the tool were limited and improved understanding, as well as reduced disagreement, among reviewers. We identified areas requiring more clarification to further improve consistency of the judgments. We will perform additional work examining studies of exposure with the goal of providing a definitive instrument to assess risk of bias in studies of environmental exposures.