Inaccuracies in Estimates of Life Expectancies of Patients with Bronchial Cancer in Clinical Decision Making

Abstract

Approximations of life expectancy in clinical decision making frequently assume constant disease-specific ("excess") mortality hazards over age at diagnosis and over time from diagnosis. This assumption is inconsistent with the longer relative survival of younger patients with bladder cancer and with the declines in mortality hazards from bladder and breast cancers over time from diagnosis. To estimate the error that may result from these as sumptions, the authors derived excess mortality hazards from the Surveillance, Epidemiology and End Result (SEER) tumor registry for bronchial cancers stratified by age at diagnosis and time from diagnosis. They compared the life expectancies calculated by a model using an average constant annual cancer-specific mortality hazard over time from diagnosis with those calculated using data-derived cancer-specific annual mortality hazards that varied as a function of time from diagnosis. For younger patients with less advanced disease, the constant-average-mortality model underestimated life expectancies by up to 50% relative to those predicted by the time-variant model. For those over 75 years old at diagnosis, and for all patients with advanced disease, the constant-average-mortality model overestimated life expectancies by up to 65% relative to those predicted by the time-variant model. The authors conclude that predictions of life expectancy with bronchial cancer, and probably with other neoplasms, are limited by the widespread use of oversimplified methods of calculation and by the lack of data describing mortality hazards as a function of time from diagnosis. The assumption of a constant average cancer-specific mortality over time from diagnosis may result in inaccurate estimates of life expectancy and bias the results of clinical decision analyses. Key words: life expectancy; mortality hazards; prediction; time. (Med Decis Making 1993;13:237-244)