A machine learning-based screening model for the early detection of prostate cancer developed using serum microRNA data from a mixed cohort of 8,741 participants.

Objective: Early detection of prostate cancer (PCa) can improve the prognosis of patients. Currently, the role of the prostate specific antigen test for PCa screening remains debatable. We aimed to develop an efficient and clinically applicable method for the screening of PCa by the noninvasive screening of several serum microRNA (miRNA) levels.

Methods: A mixed cohort, including PCa, multiple other cancers (OCa), benign prostate disease (BPD) and health population (HP), study with 8,741 samples was conducted. Six machine learning algorithms were employed to develop a screening model for PCa using the training dataset. The performance of models was assessed using the testing dataset, and the model with the optimal predictive power was chosen for further analysis. Furthermore, the detection performance of the screening model was evaluated using validation set, external set and clinical subset analysis.

Results: We first constructed a PCa early screening model named PCa4miR using only four miRNAs (miRNA-1290, miRNA-6777-5p, miRNA-1343-3p and miRNA-6836-3p), the overall sensitivity was 88.5%, and the specificity was 95.9%. The testing set (AUC = 0.972), validation set (AUC = 0.981) and external set (AUC = 0.811) were used to evaluate the model. The model has excellent discriminative ability for identifying PCa, distinguishing them from individuals with BPD, OCa, and HP as shown by the detection index calculations. The results of subgroup analysis showed that the model had excellent detection accuracy in different clinical subgroups of PCa. Additionally, we used the ratio of miR-1290/miRNA-6836-3p, miR-6777-5p/miRNA-6836-3p and miR-1343-3p/miR-6087 to establish a screening scoring model for PCa (PCaSS) with excellent detection efficacy. The PCaSS model is the first scoring model for PCa using serum miRNA ratios.

Conclusion: Using the largest known sample size and the most complex mixed cohort, we have successfully devised efficient screening models for prostate cancer, namely PCa4miR and PCaSS. These models have demonstrated exceptional screening accuracy, underscoring their capacity for the early detection of prostate cancer.