Article type
Abstract
Background: Problems in the design and execution of research studies lead to concern over the validity of their findings. Systematic assessment of risk of bias in studies is required to draw conclusions about the strength of the evidence for causal effects of interventions on health outcomes. The ROBINS-I (Risk Of Bias In Non-randomised Studies - of Interventions) tool provides a structured approach to assessing risk of bias in non-randomised studies of interventions (NRSI). The tool involves comparison of the NRSI with a theoretical, perfectly conducted, randomised-controlled trial. Risk of bias is assessed over seven domains, with judgements guided by responses to signalling questions. The published tool and guidance focus mainly on studies with a cohort-type design. However, the NRSI designs most commonly included in Cochrane Reviews are interrupted time series (ITS) and controlled before-after (CBA) studies.
Objective: Adapt the ROBINS-I tool, including its signalling questions and accompanying guidance, to assess risk of bias in ITS and CBA studies.
Methods: Working groups considered risks of bias specific to ITS and CBA study designs. They met through a series of teleconferences and at a face-to-face meeting. Modifications to the ROBINS-I tool were developed by expert consensus. Preliminary tools for each study design were piloted within the working groups. Feedback from piloting informed further modifications.
Results: Additional signalling questions were added to the confounding domain for both study designs. For ITS studies, questions related to the ability of observed trends pre-intervention to predict what would have occurred post-intervention in the absence of intervention. For CBA studies, questions related to the ability of the control group to mimic what would have occurred in the intervention group in the absence of the intervention. New signalling questions also address biases that may arise in designs where an intervention is cluster-allocated and where populations are cross-sectionally sampled at different time points.
Conclusions: These modifications to the ROBINS-I tool will allow its use to assess risk of bias in ITS and CBA studies.
Objective: Adapt the ROBINS-I tool, including its signalling questions and accompanying guidance, to assess risk of bias in ITS and CBA studies.
Methods: Working groups considered risks of bias specific to ITS and CBA study designs. They met through a series of teleconferences and at a face-to-face meeting. Modifications to the ROBINS-I tool were developed by expert consensus. Preliminary tools for each study design were piloted within the working groups. Feedback from piloting informed further modifications.
Results: Additional signalling questions were added to the confounding domain for both study designs. For ITS studies, questions related to the ability of observed trends pre-intervention to predict what would have occurred post-intervention in the absence of intervention. For CBA studies, questions related to the ability of the control group to mimic what would have occurred in the intervention group in the absence of the intervention. New signalling questions also address biases that may arise in designs where an intervention is cluster-allocated and where populations are cross-sectionally sampled at different time points.
Conclusions: These modifications to the ROBINS-I tool will allow its use to assess risk of bias in ITS and CBA studies.