Risk of bias in network meta-analyses of interventions for knee and hip osteoarthritis published in leading general medical and specialty journals: A methodological study using the RoB NMA tool.

OBJECTIVE: Network meta-analyses (NMAs) frequently inform recommendations for osteoarthritis (OA), but methodological flaws may introduce bias and yield misleading results. This methodological study aimed to evaluate the risk of bias in a restricted sample of systematic reviews incorporating NMAs of health care interventions for knee and hip OA. METHOD: We searched PubMed to identify NMAs published in 22 leading general/internal medical and specialty journals ('Rheumatology/Orthopedics', 'Rehabilitation/Sports Sciences') that evaluated pharmacological or non-pharmacological interventions for knee or hip OA pain or physical function. Risk of bias was assessed independently by two reviewers using the ROBIS tool in combination with the Risk of Bias in Network Meta-Analysis (RoB NMA) tool. RESULTS: A total of 21 NMAs were included. Of these, 11 (52%) evaluated pharmacological interventions, 7 (33%) non-pharmacological interventions, and 3 (14%) a combination. Most NMAs (13; 62%) were rated high risk of bias, four (19%) raised some concerns, and four (19%) were low risk. Risk of bias was most frequently related to intervention grouping and assessment of transitivity, including limited consideration of effect modifiers and statistical consistency. Assessments took a median of 45 minutes (range: 20-110 minutes), with 67-71% inter-rater agreement and slight to fair agreement statistics. CONCLUSION: Knee and hip OA NMAs published in leading journals frequently exhibit limitations in intervention grouping and assessment of transitivity, potentially affecting their validity. Future OA NMAs should implement transparent node-making and assess transitivity a priori by examining key clinical and methodological study characteristics to determine whether valid statistical synthesis is feasible.