Online Program

Cost Effectiveness Methods in Oncology: Bootstrapping a Risk Adjusted, Censored ICER

*Gerhardt Michael Pohl, Eli Lilly and Company 

Keywords: ICER, Propensity, Bootstrap

The Incremental Cost Effectiveness Ratio is a frequently used tool in health technology assessment. As commonly applied, however, it frequently fails to take into account the variability inherent in the component mean cost differences and mean effectiveness differences. Furthermore, it often does not explicitly adjust for differing risk factors present among patients nor for administrative censoring that may occur among patients who survive beyond the temporal end of a database. These issues may be particularly important in oncology research where patient survival may be impacted by covariates such as age and tumor stage at time of diagnosis and where using only complete records, that is, data only from patients observed until death, may bias estimates. We propose a bootstrapped approach using direct stratification on patient characteristics at baseline to provide a sense of variability and to adjust for risk factors. Within each stratum, censoring of survival is accomplished via Kaplan-Meier methods; and censoring of costs, by the well-known methods of Lin, et al., Biometrics, 1997. The within-stratum results then are pooled proportional to the stratum size and the resulting overall mean cost and survival differences combined to form an overall ICER. The relationship of this pooling method to inverse propensity weighting using the within-stratum fraction of treated patients as a local propensity score is described as well as connections to concepts in causal inference such as average overall treatment effect. Alternative frameworks for causal inference as also interpreted within the context of different cross-stratum weightings.