Prevalence and predictors of bias in the reporting of primary efficacy and toxicity endpoints in randomized clinical trials of radiation oncology

Introduction: To determine the prevalence and predictors of bias in reporting of primary efficacy and toxicity endpoints in randomized trials (RCTs) of radiation oncology. Methods: We searched MEDLINE for eligible RCTs published from January 1994 to October 2014. Bias in reporting of primary efficacy endpoint was defined as reporting that treatment was beneficial based on secondary endpoints despite a statistically non-significant difference in primary endpoint. Bias in reporting of toxicity endpoint was defined as not reporting toxicity findings in the abstract, discussion or results table. Logistic regression multivariate models were used to determine predictors of biased reporting. Results: We found that 13% of 323 RCTs have bias in the reporting of primary efficacy endpoint with non-cooperative group trials as a significant predictor of bias (odds ratio (OR) 2.04, 95% confidence interval (CI) 1.03-4.00, P = 0.04). Thirty-five per cent of 279 RCTs were judged to have bias in the reporting of toxicity endpoint with trials not listed in Clinicaltrials. gov as a significant predictor of bias (OR 3.23, 95% CI 1.43-7.14, P = 0.004). Conclusion: The prevalence of bias in reporting of primary efficacy and toxicity endpoint for radiotherapy RCTs was 13% and 35% respectively. Noncooperative group trials were more likely to have bias in the reporting of primary efficacy endpoint. Trials not listed in Clinicaltrials. gov were more likely to have bias in the reporting of toxicity endpoint.