CANTRANCE: FROM CANCER COMPARATIVE EFFECTIVENESS STUDIES TO DISEASE-SPECIFIC MORTALITY

Tuesday, October 25, 2011
Grand Ballroom AB (Hyatt Regency Chicago)
Poster Board # 26
(ESP) Applied Health Economics, Services, and Policy Research

Jeanette K. Birnbaum, MPH, University of Washington, Seattle, WA, Jeffrey Katcher, Fred Hutchinson Cancer Research Institute, Seattle, WA and Ruth Etzioni, PhD, Fred Hutchinson Cancer Research Center/ University of Washington, Seattle, WA

Purpose: Although disease-specific mortality is a primary endpoint of interest in cancer comparative effectiveness (CE) studies, practical considerations often dictate the study of intermediate endpoints such as disease incidence, detection, severity, treatment, or recurrence.  In response, many models that are highly application-specific have been developed to extrapolate to mortality.  CANcer TRANslation for Comparative Effectiveness (CANTRANce) is a user-friendly modeling framework that formalizes and streamlines mortality projections for cancer CE studies.

Methods: We constructed a simulation framework with a web interface where users can enter the results of their CE study, choose from several options how to model their intermediate endpoint, and specify key additional information needed to extrapolate from their endpoint to mortality.  Disease-specific mortality is then modeled as an exponential process with other-cause death as a competing risk.  We applied this framework to two CE studies with different purposes, interventions, and intermediate endpoints: (1) an adjuvant therapy for colon cancer recurrence and (2) a diagnostic test for therapeutic decision-making in breast cancer. 

Results:  The interface transmits the user’s data and modeling specifications to a simulation program that returns estimates of all-cause survival as well as crude and net disease-specific survival over a user-specified time period.  All-cause survival is compared across intervention groups to determine the number of years of life saved or lost by the intervention.  In the colon cancer example, a treatment that decreased the risk of recurring within 5 years by approximately 20% translated into a gain of about 2.5 years in median all-cause survival and over 100 years of life saved in a population of 2000 over 10 years.  In contrast, the change in treatment distribution due to the diagnostic test for targeted therapy in breast cancer corresponded to no significant reduction in mortality, but allowed low-risk women to avoid unnecessary treatment. 

Conclusions: CANTRANce is a promising tool for cancer researchers to estimate the impact of interventions on mortality without developing their own complex models.  Initial applications suggest that evidence-based modeling assumptions can produce very reasonable results.  Future work will expand the suite of examples to additional intermediate endpoints and deploy an open-access user interface.