Predicting the likelihood of success of phase 3, randomized, controlled trials for cancer therapeuticsDespite significant investments in cancer drug development worldwide, a large proportion of Phase 3 randomized controlled trials (RCTs) fail to demonstrate clinical success, with conventional wisdom being that these failures relate to a lack of drug efficacy. This honors thesis investigates whether failure in oncology RCTs can be systematically predicted based on the structural and socioeconomic characteristics of a given phase 3 cancer trial’s design and conduct. Using a curated dataset of 155 Phase 3 cancer trials in the first-line metastatic setting for a selected group of highly lethal cancers, a logistic regression model was developed to predict trial outcomes as successes or failures. Predictor variables included trial design features, sponsorship, regional attributes, accrual metrics, and the existence of non-randomized evidence to support the decision to conduct the Phase 3 trial, such as the experimental drug’s Phase 2 trial status. To assess the model’s performance and predictions, multicollinearity and variance inflation factor (VIF) diagnostics were performed, revealing moderate collinearity among certain structural predictors such as condition, region, and sponsor type. The final draft of the model yielded an overall prediction accuracy of 81%, with a sensitivity of 62.5% and a specificity of 89.7%. Additional diagnostic statistics, including the kappa statistic, McNemar's test, and positive and negative predictive values, underscored the model’s strength in detecting failed trials while maintaining strong negative classification performance. Our results suggest that trial failure in RCTs in the first-line metastatic setting is not entirely caused by experimental drug inefficacy, as some trial design and contextual factors are significantly associated with the likelihood of success. This highlights a systemic need to incorporate historical trial information and feasibility assessments into future trial construction, participant enrollment, and data collection. However, limitations related to the dataset size, model assumptions, and binary outcome classification may necessitate the future use of flexible predictive frameworks to understand cancer trial success. Overall, this study contributes to a growing body of research seeking to enhance clinical trial efficiency and reduce preventable failures in cancer drug development.Paul, Kyle A.creatorWarner, Jeremythesis advisorUzun, EcereaderBrown University. Biology and Medicinesponsor2025text_resourcesborn digitalEnglishSenior thesis (ScB)--Brown University, 2025Cell and Molecular BiologythesesCancerLogistic regression analysisClinical trialsdata scienceAll rights reservedIn CopyrightAll Rights Reserved10.26300/vxh5-4x54