Background Estimates of the fraction of patients cured from cancer provide important information to both patients and clinicians. But how reliable are the estimates?
Methods Three statistical approaches, based on similar assumptions that a fraction of patients will be cured from cancer, were used to estimate the fraction cured. The first approach was CANSURV software and the second was developed by De Angelis et al (1999), both using grouped survival data. The third was published by Lambert et al (2007), requiring individual patient records. All three approaches fit mixture cure models; and CANSURV and Lambert's implementation use maximum likelihood, while De Angelis' implementation is based on non-linear least squares. Cansurv is a standalone program whereas the other approaches were implemented using SAS and Stata respectively. SEER-9 data for rectal cancer were used to illustrate the methods.
Results As shown in the Abstract P1-63 table 1, estimates of the cure fraction were similar for the two approaches requiring grouped survival data while Lambert's method provided lower a estimate for patients with localised disease.
Discussion The three approaches provided similar estimates of the cure fraction for patients with regional and distant stage at diagnosis, however there are considerable differences in the estimates for patients with localised disease. Estimates of the cure fraction appear to depend on the choice of statistical model even when the underlying assumptions are very similar.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.