More consideration is needed when conducting non-randomised studies of interventions
DHIMAN P., Lee H., KIRTLEY S., COLLINS G.
Objective: Evaluate the methodological conduct, reporting, and risk of bias of non-randomised studies of interventions (NRSIs) published by UK National Institute for Health Research Biomedical Research Centres (NIHR-BRCs). Study design and setting: We conducted a systematic review, searching the Medline and Web of Science databases between 2012-2018, for NRSIs funded by NIHR-BRCs. Eligible studies were published between April 2012 and December 2017. We selected a contemporary subset of NRSIs published in 2017. We extracted study design, methods for overcoming confounding bias from non-randomisation, analysis methods, and items for assessing risk of bias. Risk of bias was the primary outcome, assessed using Risk Of Bias In Non-randomised Studies – of Interventions (ROBINS-I). Results: Fifty-two NSRI publications were included, of which over half were cohort studies and 29% before-and-after studies. 77% analysed non-purposefully-collected data. All had serious or critical risk of bias. Regression adjustment was most commonly used to address confounding bias (50%). Few (12%) studies accounted for missing data and 42% reported different numbers of outcomes in their methods and results. Conclusion: Most reviewed NRSIs had serious or critical risk of bias. Although NRSIs can evaluate treatment effects when appropriately conducted, this review shows that their design, analysis, and reporting require more consideration.