OBJECTIVES: To comprehensively evaluate clinical predictors of spontaneous acute urinary retention (AUR) across pooled data of placebo-treated patients from clinical trials conducted in men with lower urinary tract symptoms and clinically diagnosed benign prostatic hyperplasia. METHODS: Data from the placebo-treatment groups of several prospective, randomized clinical trials conducted in the United States (n = 3040), Scandinavia, Canada, and worldwide (n = 2295) were combined in the analyses. More than 110 variables were considered individually and in combination as predictors of AUR using logistic regression analysis and classification and regression tree methods with a split-sample approach to cross-validation. RESULTS: The different methods of analysis identified consistent potential predictors of episodes of AUR. When prostate volume was included in the analyses, it was selected as the initial variable discriminating men with and without subsequent AUR. Omitting prostate volume because of its availability in only a subset of men, a logistic model including serum prostate-specific antigen (PSA), urinating more than every 2 hours, symptom problem index, maximum urinary flow rate, and hesitancy of urination had good predictive properties (area under the receiver-operating characteristic curve [AUC] = 0.742 +/- 0.047), as did a model with PSA (AUC = 0.716 +/- 0.045). A classification and regression decision tree with the same variables predicted AUR (AUC = 0.74, sensitivity = 72%, specificity = 67%) as well as did a tree with PSA alone (AUC = 0.70, sensitivity = 75%, specificity = 64%). CONCLUSIONS: Prostate volume and serum PSA are strong predictors of AUR in placebo-treated men with lower urinary tract symptoms and clinically diagnosed benign prostatic hyperplasia who were screened for prostate cancer. From more than 110 variables, logistic models and decision trees with PSA alone were comparable to expanded models that included PSA, urinary frequency and hesitancy, flow rate parameters, and symptom problem index, and to a scoring algorithm.
RCT Entities:
OBJECTIVES: To comprehensively evaluate clinical predictors of spontaneous acute urinary retention (AUR) across pooled data of placebo-treated patients from clinical trials conducted in men with lower urinary tract symptoms and clinically diagnosed benign prostatic hyperplasia. METHODS: Data from the placebo-treatment groups of several prospective, randomized clinical trials conducted in the United States (n = 3040), Scandinavia, Canada, and worldwide (n = 2295) were combined in the analyses. More than 110 variables were considered individually and in combination as predictors of AUR using logistic regression analysis and classification and regression tree methods with a split-sample approach to cross-validation. RESULTS: The different methods of analysis identified consistent potential predictors of episodes of AUR. When prostate volume was included in the analyses, it was selected as the initial variable discriminating men with and without subsequent AUR. Omitting prostate volume because of its availability in only a subset of men, a logistic model including serum prostate-specific antigen (PSA), urinating more than every 2 hours, symptom problem index, maximum urinary flow rate, and hesitancy of urination had good predictive properties (area under the receiver-operating characteristic curve [AUC] = 0.742 +/- 0.047), as did a model with PSA (AUC = 0.716 +/- 0.045). A classification and regression decision tree with the same variables predicted AUR (AUC = 0.74, sensitivity = 72%, specificity = 67%) as well as did a tree with PSA alone (AUC = 0.70, sensitivity = 75%, specificity = 64%). CONCLUSIONS: Prostate volume and serum PSA are strong predictors of AUR in placebo-treated men with lower urinary tract symptoms and clinically diagnosed benign prostatic hyperplasia who were screened for prostate cancer. From more than 110 variables, logistic models and decision trees with PSA alone were comparable to expanded models that included PSA, urinary frequency and hesitancy, flow rate parameters, and symptom problem index, and to a scoring algorithm.