Accuracy of life tables in predicting overall survival in candidates for radiotherapy for prostate cancer.

Abstract

To test the accuracy of life tables (LTs) in predicting survival in men treated with radiotherapy for localized prostate cancer.
We selected the records of 3,176 patients treated with radiotherapy and who had no clinical evidence of disease relapse. Life table-derived life expectancy (LE) was defined for every individual using a population-specific LT. Age, Charlson Comorbidity Index (CCI), and LT-derived LE were then used as predictors of overall mortality in Cox regression models. Predictive accuracy (PA) was estimated with the Harrell's concordance index and was internally validated with 200 bootstrap resamples.
The actuarial median survival was 4.7 years (mean, 6.4 years). At radiotherapy, median age was 70.6 years, median CCI was 2, and median LT-derived LE was 12 years. All variables were statistically significant predictors of overall mortality (all p values <0.001). Age (PA, 60.2%), CCI (PA, 60.1%), and LT-derived LE (PA, 60.2%) were equally accurate. Finally, when age and CCI were combined (PA, 63.2%), both variables provided more accurate mortality predictions than either variable alone (all p values = 0.01).
Life tables have a limited ability to predict LE in patients treated with radiotherapy for prostate cancer. We, therefore, recommend the use of multivariate prognostic models that integrate several variables, such as at least age and comorbidities, to estimate LE. This might help to improve LE estimation during prostate cancer treatment decision making.