As cancer treatments have improved, a growing proportion of patients may eventually be considered cured. For these patients, their subsequent risk of death can become the same as a person who hasn’t had cancer. Not accounting for these cured patients can bias economic evaluations of treatment effects. In a recent issue of Value in Health, Drs. Megan Othus and Scott Ramsey in the Public Health Sciences Division assessed the cost-effectiveness of various treatments in a clinical trial of advanced melanoma. Using a mixture cure modeling approach, rather than standard methods, provided more accurate estimates of overall survival and cost-effectiveness.
Standard survival modeling methods may not be equipped to address situations where therapies essentially cure some of the patients treated. Within a clinical trial population, for example, there may be both a group of cured patients and a group of noncured patients. These two groups of patients are likely to have different mean survival times and different costs. While both cured and noncured patients may share the same baseline mortality risk, for example, noncured patients have additional mortality risk related to their cancer. Similarly, cured patients have higher costs related to long-term follow-up and surveillance compared to noncured patients. Because of this heterogeneity of the patient population, analyses that take a simple average of the total survival time will lead to biased results. Instead, cure models providing a weighted average of the mean survival times or cost of each of the cured and noncured groups should be calculated, weighted by the relative proportion of the two groups.
Image from Dr. Megan Othus
To evaluate the effect of cured patients on survival and cost-effectiveness measures, the authors analyzed data for two treatment arms in clinical trial data of patients with unresectable stage III and IV melanoma. Patients in these arms were either treated with ipilimumab (Ipi) or a cancer vaccine derived from the melanosomal glycoprotein 100 (gp100). When evaluating overall survival, a standard analysis not considering cured patients suggested that the Ipi arm had a mean 8 months longer survival than the gp100 arm. Using a mixture cure analysis, however, showed that the mean overall survival among noncured patients (10 months) and cured patients (26 months) was similar in both arms, and that the differences in survival between the two arms was driven by 15% more cured patients in the Ipi arm.
The authors also compared estimates of the incremental cost-effectiveness ratio (ICER) per quality-adjusted life year (QALY). The ICER statistic summarizes the additional value of a particular treatment compared to an alternative, taking into consideration the costs and effects of two treatments. When not taking cured patients into account, the ICER comparing Ipi to gp100 was $324,000/QALY. When the proportion of cured patients was taken into account, however, the ICER comparing Ipi to gp100 was only $113,000/QALY. This substantial difference in estimates was attributed to both the greater proportion of cured patients in the Ipi arm and the significantly higher QALYs in the cured patients.
Importantly, the cure models described in this manuscript are also of potential relevance to other cancer types, though the degree of difference between standard modeling and cure modeling depends on factors such as the costs of the therapies and the proportion of cured patients. Said lead author Dr. Othus, “with the new cancer therapies becoming available, we expect more patients in more cancers to be long-term survivors or cured of their cancer. This manuscript demonstrated that accounting for this is important in health economic analyses, and can potentially change the conclusions regulatory agencies make when reviewing such therapies.”
Also contributing to this study from the Fred Hutch were Dr. Aasthaa Bansal and Ms. Lisel Koepl.
Funding for this study was provided by Bristol-Myers Squibb.
Othus M, Bansal A, Koepl L, Wagner S, Ramsey S. Accounting for Cured Patients in Cost-Effectiveness Analysis. Value Health 2017; 20(4):705-709. doi: 10.1016/j.jval.2016.04.011.