Factors influencing the trustworthiness of non-randomized studies of interventions: a survey of international experts

ABSTRACT Background Perceived trustworthiness of research may be influenced by factors beyond the risk of bias, including study-related characteristics, research context, and external circumstances. Identifying these factors is essential for gauging the credibility of non-randomized studies of interventions (NRSIs) as they are interpreted and used in systematic reviews, and for improving their design to ensure that they provide reliable evidence for decision-making. Our objective was to identify factors, not covered in risk of bias assessment tools, that could influence the trustworthiness of NRSIs. Methods We conducted a cross-sectional survey of international experts. We defined trustworthiness as the proper, justified or rational trust in the study findings. Using convenience sampling, we recruited participants who were top-cited scientists in the field of epidemiology, members of the Cochrane Bias Group and Cochrane Non-Randomized Studies Methods Group, authors of initiatives related to observational studies and corresponding authors of NRSIs. Through an online survey, we asked them to identify factors that they believe could influence the trustworthiness of NRSIs. We analyzed qualitative data using an inductive thematic approach. We first coded the responses, which were redefined into factors and grouped under themes. We summarized findings in frequencies and percentages. Results 130 participants out of 1488 contacted completed the survey. Of the 130 participants, 40 (31%) were methodologists and 61 (47%) had 21-40 years of experience in research. The level of expertise in NRSIs ranged from intermediate (35%) to advanced (30%) and expert (30%).We identified a total of 56 factors, with a median of 6 factors per participant (IQR 3; 9, range 0-20). We grouped the factors under 20 domains, when relevant, and eventually under eight overarching themes: Open Science (e.g., transparency, registration), Research Question (e.g., appropriate rationale and hypothesis), Study Methodology (e.g., study design, participants, statistical considerations), Data Source (e.g., quality), Findings and Interpretation (e.g., plausibility of effect estimate), Writing (e.g., appropriate writing), Oversight (e.g., investigators, journal), and Artificial Intelligence (e.g., no suspicion of use in writing or synthesis). Conclusions Our findings provide insight to gauge and improve the quality and uptake of NRSIs, with important implications for strengthening evidence-based decision-making in both research and practice.