Modeling Disease-specific Survival in Observational Studies with Missing Cause of Death: Leveraging Information from Clinical Trial Data.

Modeling disease-specific survival (DSS) is essential for evaluating the risk of death due to the disease itself, offering valuable insights into disease progression, identifying high-risk subgroups, and informing treatment decisions. The National Cancer Database (NCDB), one of the largest cancer registries in the United States, has the potential to be a valuable resource for assessing DSS in the general cancer population. However, the NCDB does not record cause-of-death information due to practical limitations. In contrast, randomized cancer clinical trials provide detailed documentation of the cause of death. To bridge this gap, a framework is developed to leverage cause-of-death information from clinical trials to infer DSS in the NCDB population through a proportional cause ratio model. This model accommodates both temporal variation and patient-level characteristics influencing the distribution of causes of death. To account for population heterogeneity between trial participants and registry patients, information transfer is restricted to the cause-of-death mechanism rather than the overall survival distribution. Estimation procedures are established, and the asymptotic properties of the resulting estimators are rigorously derived. Extensive simulation studies demonstrate the validity of the proposed estimators and associated inference under both the proportional cause ratio model and the proportional hazards model for DSS. The proposed method is applied to our motivating study in which cause of death information from the National Surgical Adjuvant Breast and Bowel Project (NSABP B-06 trial) is transferred to the NCDB to infer breast cancer-specific survival based on demographics of the patient, clinical characteristics, and treatments received.