Purpose: To examine the impact of modeling chemotherapy effectiveness with a cure assumption (i.e., a proportion of patients will have no risk of recurrence; the remainder will experience no additional benefit) vs. a proportional hazards (PH) assumption (i.e., all patients will experience a constant reduction in the recurrence rate). We used the example of adjuvant therapy with oxaliplatin plus fluorouracil/leucovorin (FL) (FOLFOX) compared with FL alone for patients with stage III colorectal cancer.

Methods: We developed a Markov model with the following health states: recurrence-free, local recurrence only, metastatic recurrence, and dead. We estimated key model parameters by calibrating each model to predict the reported 3-year disease-free survival probability and its upper and lower bounds. The base-case effectiveness PH estimate of 0.53 (47% reduction in recurrence rate) corresponded to a cure probability of 0.23. All other parameters, such as the cancer-specific hazard rate among patients with metastatic recurrence, were equivalent across models and were based on the literature. Costs were derived from an analysis of SEER-Medicare linked data based on patients diagnosed in 2002-4.

Results: Discounted life expectancy gains (FOLFOX vs. FL) were approximately 12-18% greater for the cure assumption compared with the PH model (1.04 vs. 0.90 discounted years for base case). The lifetime risk of a metastatic recurrence was 0.30 and 0.28 for the PH and cure models, respectively. Discounted net costs were slightly lower with the cure model, and depended on the magnitude of the costs following metastatic recurrence relative to non-metastatic costs. In our base case, the metastatic costs were $1841 higher and the non-metastatic costs were $1646 lower for the PH model compared with the cure model. The incremental cost-effectiveness ratio for FOLFOX vs. FL for the base case was $27,500 per life-year saved (LYS) and $23,700/LYS for the PH and cure assumptions, respectively. Using the upper-bound estimate for 3-year disease-free survival, the ratios were $110,000/LYS (PH) and $92,900/LYS (cure).

Conclusions: Incorporation of a cure assumption tends to result in larger life expectancy gains and lower net costs, with incremental cost-effectiveness ratios about 15% lower with the cure model. Because of the uncertainties about the true mechanisms of chemotherapy effectiveness, the exploration of these two assumptions should be conducted as a structural sensitivity analysis.