A systematic methodological review of non-randomised interventional studies of elective ventral hernia repair: clear definitions and a standardised minimum dataset are needed.

Background: Ventral hernias (VHs) often recur after surgical repair and subsequent attempts at repair are especially challenging. Rigorous research to reduce recurrence is required but such studies must be well-designed and report representative and comprehensive outcomes.
Objective: We aimed to assesses methodological quality of non-randomised interventional studies of VH repair by systematic review.
Methods: We searched the indexed literature for non-randomised studies of interventions for VH repair, January 1995 to December 2017 inclusive. Each prospective study was coupled with a corresponding retrospective study using pre-specified criteria to provide matched, comparable groups. We applied a bespoke methodological tool for hernia trials by combining relevant items from existing published tools. Study introduction and rationale, design, participant inclusion criteria, reported outcomes, and statistical methods were assessed.
Results: Fifty studies (17,608 patients) were identified: 25 prospective and 25 retrospective. Overall, prospective studies scored marginally higher than retrospective studies for methodological quality, median score 17 (IQR: 14-18) versus 15 (IQR 12-18), respectively. For the sub-categories investigated, prospective studies achieved higher median scores for their, 'introduction', 'study design' and 'participants'. Surprisingly, no study stated that a protocol had been written in advance. Only 18 (36%) studies defined a primary outcome, and only 2 studies (4%) described a power calculation. No study referenced a standardised definition for VH recurrence and detection methods for recurrence varied widely. Methodological quality did not improve with publication year or increasing journal impact factor.
Conclusion: Currently, non-randomised interventional studies of VH repair are methodologically poor. Clear outcome definitions and a standardised minimum dataset are needed.