(072) Patterns of Underreporting of Enrollment Dates in Cancer Clinical Trials: A HemOnc Knowledgebase Analysis

Background: The systemic treatment of cancer has evolved greatly over the past 75 years, with improving cancer survival trends over time. Given rapid change in the field, the enrollment period during which a trial was conducted is a critical component of placing results in context. However, reporting is inconsistent, and patterns of underreporting have not been well described.

Objectives: To determine the patterns of underreporting of enrollment period in published manuscripts of cancer clinical trial results.

Methods: Data were sourced from the HemOnc knowledgebase (KB), a repository of cancer and blood disorders treatment information. Prospective trials of systemic anticancer therapy with at least one published manuscript with curated years of trial enrollment were eligible for inclusion. The following variables were determined to be of interest a priori: 1) trial type (randomized vs not); 2) sponsor type (pharmaceutical industry vs other); 3) whether the trial was present in a clinical trial registry (Y/N); 4) pivotal trial for regulatory indication (Y/N); 5) cancer group studied (hematologic vs solid tumor); 6) year of publication; 7) journal of publication (high vs moderate/low impact); and 8) region of first author affiliation (US vs ex-US). Variables with < 5% missingness were considered for inclusion in a multivariable logistic regression (MLR) model. All analyses were performed in R 4.3.2.

Results: As of January 23, 2004, there were 6612 trials in the HemOnc KB, of which 5529 (84%) met inclusion criteria. 531 trials (10%) did not report start year, end year, or both. All variables were 100% complete except for sponsor type, which had 34% missingness. Enrollment was underreported in nonrandomized trials (19% vs 6%), unregistered trials (15% vs 5%), and older trials (pre-2005: 17% vs 2015-onwards: 5%). In MLR, factors associated with an increased likelihood of reporting enrollment were randomized trials (OR 2.89, 95% CI 2.34-3.56), registered trials (OR 1.40, 95% CI 1.07-1.84), newer trials (OR 1.50 per decade, 95% CI 1.36-1.66), high journal impact (OR 1.74, 95% CI 1.41-2.15), and ex-US country of first author affiliation (OR 1.51, 95% CI 1.23-1.85). Pivotal trial status and cancer type studied were not associated with underreporting of enrollment. Variance inflation factors were < 2, and C-statistic was 0.76.

Conclusions: Multiple factors appear to be associated with underreporting. Although there has been improvement over time, some underreporting persists in the modern literature. Trial registration may act as a “Hawthorne effect” for the reporting of trial enrollment details. In the future we plan to investigate additional factors that may be associated with underreporting, such as institutional affiliations of trial leads and drug class(es) being studied.