A meta-epidemiological study of bias in randomized clinical trials of open and laparoscopic surgery.

Abstract

BACKGROUND\nBlinding, random sequence generation, and allocation concealment are established strategies to minimize bias in RCTs. Meta-epidemiological studies of drug trials have demonstrated exaggerated treatment effects in RCTs where such methods were not employed. As blinding is more difficult in surgical trials it is important to determine whether this applies to them. The study aimed to investigate this using systematic meta-epidemiological methods.\n\n\nMETHOD\nThe Cochrane Database of Systematic Reviews was searched for systematic reviews of RCTs that compared laparoscopic and open abdominal surgical procedures. Each review was then scrutinized to determine whether at least one of the included trials was blinded. Eligible reviews were updated and individual RCTs retrieved. Extracted data included the primary outcomes of interest (length of stay and complications), secondary outcomes and a risk of bias assessment. A multistep meta-regression analysis was then performed to obtain an overall difference in the reported outcome differences between trials that employed each bias-minimization strategy, and those that did not.\n\n\nRESULTS\nSome 316 RCTs were included, reporting on eight different procedures. Patient-blinded RCTs reported a smaller difference in length of stay between laparoscopic and open groups (difference of standardized mean differences -0·36 (95 per cent c.i. -0·73 to 0·00)) and complications (ratio of odds ratios 0·76 (95 per cent c.i. 0·61 to 0·93)). Blinding of postoperative carers and outcome assessors had similar effects.\n\n\nCONCLUSION\nLack of blinding significantly altered the treatment effect estimates of RCTs comparing laparoscopic and open surgery. Blinding should be implemented in surgical RCTs where possible to avoid systematic bias.