Hospital-based prostate cancer screening in vietnamese men with lower urinary tract symptoms: a classification and regression tree model

Abstract

Abstract Background Prostate cancer (PCa) is a common disease in men over 65 years of age, and should be detected early, while reducing unnecessary biopsies. This study aims to construct a classification and regression tree (CART) model (i.e., risk stratification algorithm) using multivariable approach to select Vietnamese men with lower urinary tract symptoms (LUTS) for PCa biopsy. Methods We conducted a case-control study on 260 men aged ≥ 50 years who visited MEDIC Medical Center, Vietnam in 2017–2018 with self-reported LUTS. The case group included patients with a positive biopsy and the control group included patients with a negative biopsy diagnosis of PCa. Bayesian Model Averaging (BMA) was used for selecting the most parsimonious prediction model. Then the CART with 5-fold cross-validation was constructed for selecting men who can benefit from PCa biopsy in steps by steps and intuitive way. Results BMA suggested five potential prediction models, in which the most parsimonious model including PSA, I-PSS, and age. CART advised the following cut-off points in the marked screening sequence: 18 < PSA < 33.5 ng/mL, I-PSS ≥ 19, and age ≥ 71. Patients with PSA ≥ 33.5 ng/mL have a PCa risk was 91.2%; patients with PSA < 18 ng/mL and I-PSS < 19 have a PCa risk was 7.1%. Patient with 18 ≤ PSA < 33.5ng/mL and I-PSS < 19 have a PCa risk is 70% if age ≥ 71; and is 16% if age < 71. In overall, CART reached high predictive value with AUC = 0.915. Sensitivity, specificity, positive predictive value, negative predictive value, and accuracy of CART at the 20% diagnosis probability threshold were 91.5%, 86.2%, 86.9%, 91.2%, and 88.9% respectively; at 80% diagnosis probability threshold were 79.2%, 92.3%, 91.2%, 81.6%, and 85.8% respectively. Conclusion CART combining PSA, I-PSS, and age has practical use in hospital-based PCa screening in Vietnamese men with lower urinary tract symptoms.