Modeling Covariate-Adjusted Survival for Economic Evaluations in Oncology
OBJECTIVES: Survival data from randomized controlled trials (RCT) is routinely extrapolated for economic evaluations in oncology. Imbalances in prognostic and/ or predictive factors across treatment arms should be adjusted to generate unbiased estimates. To date no formal guidance has been developed...
Saved in:
Published in: | Value in health Vol. 20; no. 9; p. A408 |
---|---|
Main Authors: | , , , , |
Format: | Journal Article |
Language: | English |
Published: |
Lawrenceville
Elsevier Science Ltd
01-10-2017
|
Subjects: | |
Online Access: | Get full text |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | OBJECTIVES: Survival data from randomized controlled trials (RCT) is routinely extrapolated for economic evaluations in oncology. Imbalances in prognostic and/ or predictive factors across treatment arms should be adjusted to generate unbiased estimates. To date no formal guidance has been developed regarding how such adjustments should be made. We compared various covariate-adjusted survival modeling approaches, based on parametric regression and propensity score matching, applied to the ENDEAVOR RCT in multiple myeloma that assessed carfilzomib-dexamethasone (Cd) versus bortezomib-dexamethasone (Vd). METHODS: Overall survival (OS) data and baseline characteristics were used for a subgroup (bortezomib-naive/one prior therapy) reflecting the population where Cd is recommended in England and Wales. The following adjusted survival modeling approaches were compared: multiple Weibull regression model including prognostic/predictive covariates jointly fitted to the two arms to predict survival i) using the mean value of each covariate and ii) using the average of patient-specific survival predictions; iii) applying an adjusted hazard ratio derived from a Cox proportional hazard model to the baseline risk estimated for Vd with a Weibull model; iv) propensity score matching followed by fitting a Weibull model to the two arms of the balanced data including treatment group as the only covariate (matched data approach). RESULTS: The difference in mean OS estimated by the matched data approach was 2.06 years (0.02-5.01) with the smallest variance among the estimates. Despite other approaches estimated similar differences, the mean OS appeared biased (using the mean value of each covariate yielded skewed survival estimates), had limited external validity (implausible long-term OS predictions), and required assumptions not statistically appropriate, e.g. proportional hazards were not satisfied for all covariates. CONCLUSIONS: Adjusted survival modeling based on matched data approaches provides a flexible and robust method to correct for covariate imbalances in economic evaluations. The conclusions of our study may be generalizable to other settings. |
---|---|
ISSN: | 1098-3015 1524-4733 |
DOI: | 10.1016/j.jval.2017.08.061 |