Analytical strategies for characterizing chemotherapy diffusion with patient-level population-based data

To inform assessments of the quality of cancer care, we describe analytical approaches to characterizing trends in diffusion of chemotherapy drugs subsequent to their US FDA approval. The economics and medical literature provide two distinct sets of empirical methods for investigating diffusion of innovations: aggregate models, which use the level of market penetration as an estimator of diffusion; and disaggregate models, which evaluate diffusion based on the time required for different individual units to adopt innovations. When patient-level population-based data are available, disaggregate methods make the best use of the available information. We propose a method that employs time-to-event techniques to describe the probability of utilization of a drug within a specified timeframe subsequent to the diagnosis of cancer. By mapping the relationship between this probability and calendar time of a patient's diagnosis, we can assess trends in diffusion. Our approach accounts for the dependent censoring for death, as well as for the clustering of patients within physicians. The method proposed is illustrated using Surveillance, Epidemiology, and End Results (SEER)-Medicare data applied to two case studies: gemcitabine, approved for stage III/IV pancreatic cancer; and irinotecan, approved as a second-line therapy for stage IV colorectal cancer.