PURPOSE: Generalized competing event (GCE) models improve stratification of patients according to their risk of cancer events relative to competing causes of mortality. The potential impact of such methods on clinical trial power and cost, however, is uncertain. We sought to test the hypothesis that GCE models can reduce estimated clinical trial cost in elderly patients with cancer. METHODS: Patients with nonmetastatic head and neck (n = 9,677), breast (n = 22,929), or prostate cancer (n = 51,713) were sampled from the SEER-Medicare database. Using multivariable Cox proportional hazards models, we compared risk scores for all-cause mortality (ACM) and cancer-specific mortality (CSM) with GCE-based risk scores for each disease. We applied a cost function to estimate the cost and duration of clinical trials with a primary end point of overall survival in each population and in high-risk subpopulations. We conducted sensitivity analyses to examine model uncertainty. RESULTS: For the purpose of enriching subpopulations, GCE models reduced estimated clinical trial cost compared with Cox models of ACM and CSM in all disease sites. The relative cost reductions with GCE models compared with ACM and CSM models, respectively, were -68.4% and -14.4% in prostate cancer, -38.8% and -18.3% in breast cancer, and -17.1% and -4.1% in head and neck cancer. Cost savings in breast and prostate cancers were on the order of millions of dollars. The GCE model also reduced relative clinical trial duration compared with CSM and ACM models for all disease sites. The optimal risk score cutoff for clinical trial enrollment occurred near the top tertile for all disease sites. CONCLUSION: GCE models have significant potential to improve clinical trial efficiency and reduce cost, with a potentially large impact in prostate and breast cancers.
PURPOSE: Generalized competing event (GCE) models improve stratification of patients according to their risk of cancer events relative to competing causes of mortality. The potential impact of such methods on clinical trial power and cost, however, is uncertain. We sought to test the hypothesis that GCE models can reduce estimated clinical trial cost in elderly patients with cancer. METHODS:Patients with nonmetastatic head and neck (n = 9,677), breast (n = 22,929), or prostate cancer (n = 51,713) were sampled from the SEER-Medicare database. Using multivariable Cox proportional hazards models, we compared risk scores for all-cause mortality (ACM) and cancer-specific mortality (CSM) with GCE-based risk scores for each disease. We applied a cost function to estimate the cost and duration of clinical trials with a primary end point of overall survival in each population and in high-risk subpopulations. We conducted sensitivity analyses to examine model uncertainty. RESULTS: For the purpose of enriching subpopulations, GCE models reduced estimated clinical trial cost compared with Cox models of ACM and CSM in all disease sites. The relative cost reductions with GCE models compared with ACM and CSM models, respectively, were -68.4% and -14.4% in prostate cancer, -38.8% and -18.3% in breast cancer, and -17.1% and -4.1% in head and neck cancer. Cost savings in breast and prostate cancers were on the order of millions of dollars. The GCE model also reduced relative clinical trial duration compared with CSM and ACM models for all disease sites. The optimal risk score cutoff for clinical trial enrollment occurred near the top tertile for all disease sites. CONCLUSION:GCE models have significant potential to improve clinical trial efficiency and reduce cost, with a potentially large impact in prostate and breast cancers.