Accurate estimation of medical care costs raises a host of issues, both practical and methodological.
This article reviews methods for estimating the long-term medical care costs associated with a cancer diagnosis.
The authors consider data from administrative claims databases and describe the analytic challenges posed by these increasingly common resources. They present a number of statistical methods that are valid under censoring and describe methods for estimating mean costs and controlling for covariates. In addition, the authors compare two different approaches for estimating the cancer-related costs; namely, the portion of the long-term costs that may be attributed to the disease. Examples from economic studies of breast and colorectal cancer are presented.
In an analysis of data on colorectal cancer costs from the SEER-Medicare database, the two methods used to estimate expected long-term costs (one model based, one not model-based) yielded similar results. However, in calculating expected cancer-related costs, a method that included future medical costs among controls yielded quite different results from the method that did not include these future costs.
Statistical methods for analyzing long-term medical costs under censoring are available and appropriate in many applications where total or disease-related costs are of interest. Several of these approaches are non-parametric and therefore may be expected to be robust against the non-standard features that are often encountered when analyzing medical cost data.
*From the Fred Hutchinson Cancer Research Center, Seattle, Washington.
†From the Health Care Financing Administration, Baltimore, Maryland.
‡From the National Cancer Institute, Bethesda, Maryland.
Supported by contracts from the National Cancer Institute (N01-CN-05320 and N01-CN-650342).
Address correspondence and reprint requests to: Ruth Etzioni, Fred Hutchinson Cancer Research Center, Public Health Sciences Division, 1100 Fairview Avenue North, Box 19024, MP-665, Seattle, WA.