Predictors of a positive prostate biopsy following PSA testing: a retrospective study of Ugandan men aged 40 or more years presenting to the Uganda Cancer Institute for prostate-related care between 2014 - 2021.

Abstract

Introduction: The incidence and mortality due to prostate cancer (PCa) have been rising globally, making it the second most commonly diagnosed malignancy among men and the fourth most common cancer overall. In Uganda, PCa has been increasing at an alarming rate of 5.2% annually, making it the most rapidly growing cancer in the country and across sub-Saharan Africa. Despite the widespread use of prostate-specific antigen (PSA) testing and digital rectal examination (DRE) in PCa screening and biopsy referral, these methods remain inconclusive, and data on additional predictors of a positive biopsy are scarce. Objective: This study aimed at identifying the predictors of a positive prostate biopsy following a PSA test among men aged 40 or more years who presented to the Uganda Cancer Institute (UCI) for prostate related care between 2014 and 2021. Methods: This retrospective study analyzed the medical records of 860 Ugandan men aged 40 or more years, who sought prostate related care at UCI from 2014 to 2021. Only records with baseline PSA results and histological reports were included, excluding those with prior surgery, hormone therapy, or duplicates. Variables included PSA levels, family history of PCa, Age, smoking status, Alcohol use, prostate volume (PV), comorbidities, and imaging results. Binary and Bayesian logistic regression (LR and BLR) models were compared using sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and are under the curve (AUC), with statistical significance set at p<0.05 and 95% Confidence Intervals. Results: Among the patients, 99.5% of those with a positive PSA result also had a positive biopsy. Although the BLR model slightly outperformed the LR model in terms of AUC (0.91 vs. 0.90), the LR model demonstrated superior sensitivity and specificity (98.71% vs. 65% and 29.76% vs. 5%, respectively). The multivariable LR model identified age (OR: 0.95, 95% CI: 0.90-0.99, p=0.02) and MRI results (OR: 0.53, 95% CI: 0.29-0.93, p=0.03) as significant predictors of a positive prostate biopsy. Conclusion:This study found a 99.5% positive biopsy rate following a positive PSA test. While both models demonstrated predictive value, LR’s superior sensitivity and specificity made it more suitable for clinical use. However, BLR’s slightly higher AUC suggests potential for improved accuracy with refinement. Age and MRI findings emerged as key predictors, emphasizing their role in biopsy decisions. These findings support LR in practice while encouraging further exploration of Bayesian approaches for PCa detection in Uganda.