Sources of bias in non-randomized comparative studies of surgical procedures

IntroductionAlthough non randomized comparative studies provide weak evidence for the effectiveness of health interventions, the surgical literature is dominated by these studies. Non randomized studies are prone to selection and measurement biases that exaggerate estimates of the effectiveness of surgical procedures. We sought to quantify the extent of bias associated with characteristics of non randomized studies of surgical procedures.MethodsWe identified English language studies comparing laparoscopic-assisted and conventional surgery for colon cancer that estimated surgical complications as an outcome measure. A single reviewer determined study eligibility and extracted the data. Subgroup meta-analyses for five study characteristics were performed using R. odds ratios (ORs) and 95% confidence intervals (95% CIs) were estimated using random-effects models.Results155 comparative studies were identified from 6,261 abstracts. Sixty-three studies that provided an estimate of surgical complications were included. The summary OR for surgical complications associated with laparoscopic-assisted surgery in 21 randomized controlled trials was 0.76 (95% CI 0.62-0.94). The summary OR for surgical complications associated with laparoscopic-assisted surgery in 42 non randomized studies was 0.70 (95% CI 0.64-0.76). Studies with consecutively recruited controls and studies with matched controls estimated a larger benefit of laparoscopic-assisted surgery on complications than studies without these characteristics (see Table). Studies with concurrent controls and systematic assessment of outcomes estimated smaller benefits of laparoscopic-assisted surgery.Characteristics of non randomized studies (n=42)Characteristic presentCharacteristic absentOR⁎95% CIOR⁎95% CIProspective data collection0.670.60-0.760.680.52-0.89Consecutive recruitment of patients0.620.52-0.740.740.66-0.82Concurrent (vs historical) controls0.700.64-0.760.610.22-1.63Matching of groups0.520.35-0.780.710.66-0.78Systematic outcome assessment0.760.72-0.800.600.49-0.74⁎Favoring laparoscopic-assisted surgeryConclusionsAmong non randomized studies comparing surgical complications after laparoscopic-assisted and conventional surgery for colon cancer, use of matched controls is strongly associated with exaggerated estimates of safety.