Predicting the Failure of Randomized Controlled Trials in Radiation Oncology: What Can We Learn?

Abstract

While randomized controlled trials (RCTs) form the cornerstone of evidence-based medicine, many cancer RCTs fail to complete. Incomplete RCTs are detrimental to scientific progress, wasting resources and precluding patient participation in other important studies. As RCT failure is poorly understood in radiation oncology, we sought to review the relevant literature to identify predictors of trial failure. We undertook a review of the trial registry ClinicalTrials.gov to assess factors influencing the completion of radiotherapy RCTs. Eligibility included: 1) trials registered in clinicaltrials.gov between 09/27/2007 to 12/31/2010, 2) radiation (brachytherapy or external beam radiotherapy) required in at least 1 study arm, and 3) trials that were closed, either due to successful completion or failure to complete. Data were abstracted by 2 independent investigators, with discrepancies settled through consensus. Descriptive statistics were generated based on abstracted trial characteristics. Studies were stratified by completion status and characteristics were compared using the chi-square test, Fisher’s exact test, two-sample t-test or Wilcoxon rank sum test as appropriate. Univariable and multivariable logistic regression analyses were performed to determine factors predictive of RCT failure. The initial search yielded 460 studies, from which 138 studies met inclusion criteria after detailed review. All were RCTs, with either phase I (2%), II (43%) or III (55%) design. Most studies were conducted in North America (46.7%) followed by Europe (37%) and Asia (12%). In total, 41 (30%) RCTs failed to complete, most commonly due to a lack of accrual (59%), followed by inadequate funding (15%), drug unavailability (7%) and interim data monitoring committee recommendations (7%). A higher rate of RCT failure was observed in the more recent era (p=0.009), with rates increasing from 11% (1987-2006) to 39% (2009-2012). On univariable analysis, independent predictors of failure included RCTs with a surgical comparator (odds ratio [OR] of failure 6.45; p=0.010), government sponsorship (OR 3.68; p=0.024), inclusion of a safety endpoint (OR 3.03; p=0.014) and study start year (OR 1.18; p=0.030). On multivariate analysis, surgical RCTs were strongly predictive of failure (OR 7.90; p=0.015), while behavioral RCTs were less likely to fail (OR 0.14; p=0.047). Rates of RCT failure in radiation oncology are high and have been increasing over time. Trials with a surgical comparator are exceptionally prone to failure, whereas behavioral trials are more likely to succeed. These factors can help to inform the design of future RCTs, and develop strategies to mitigate the risk of failure in future radiation RCTs that include a surgical comparator.