Accurate prediction of need for invasive treatment in alpha 1-blocker treated patients with benign prostatic hyperplasia not possible: Bootstrap validation analysis

Abstract

Objectives Frequently, statistically significant prognostic factors are reported in published studies with suggestions that disease management should be modified. However, the clinical relevance of such factors is rarely quantified. We evaluated the accuracy of predicting the need for invasive treatment among patients with benign prostatic hyperplasia treated conservatively with alpha 1-blockers. Methods Information on eight prognostic factors was collected from 280 patients treated with alpha 1-blockers. Using the proportional hazards regression coefficients, a risk score for retreatment was calculated for each patient. The analyses were repeated on 1000 groups of 280 patients sampled from the original case series. The results from these “bootstrap analyses” were compared with the original results. Results Three statistically significant predictors of retreatment were identified. The 20% of patients with the greatest risk score had an 18-month risk of retreatment of only 20% (this should ideally approach 100%). Analyses of less than one half of all the bootstrap samples resulted in the same three significant prognostic factors. The 20% of patients with the greatest risk score in each of the 1000 samples experienced a highly variable risk of retreatment of 0% to 42%. Conclusions Strongly significant predictors for retreatment suggest the need for a change in disease management, but 4 of the 5 high-risk patients would be overtreated with a modified policy. The subclassification of patients with a relatively low risk and high risk of retreatment appeared far from accurate. Internal validation procedures may warn against the invalid translation of statistical significance into clinical relevance.