Early detection of bladder cancer (BCa) can have a positive impact on patients' prognosis. However, there is currently no widely accepted method for early screening of BCa. We aimed to develop an efficient, clinically applicable, and noninvasive method for the early screening of BCa by detecting specific serum miRNA levels. A mixed-cohort (including BCa, 12 different other cancers, benign disease patients, and health population) study was conducted using a sample size of 16,189. Five machine learning algorithms were utilized to develop screening models for BCa using the training dataset. The performance of the model was evaluated using receiver operating characteristic curve and decision curve analysis on the testing dataset, and subsequently, the model with the best predictive power was selected. Furthermore, the selected model's screening performance was evaluated using both the validation set and external set. The BCaS3miR model, utilizing only three serum miRNAs (miR-6087, miR-1343-3p, and miR-5100) and based on the KNN algorithm, is the superior screening model chosen for BCa. BCaS3miR consistently performed well in both the testing, validation, and external sets, exceeding 90% sensitivity and specificity levels. The area under the curve was 0.990 (95% CI: 0.984-0.991), 0.964 (95% CI: 0.936-0.984), and 0.917 (95% CI: 0.836-0.953) in the testing, validation, and external set. The subgroup analysis revealed that the BCaS3miR model demonstrated outstanding screening accuracy in various clinical subgroups of BCa. In addition, we developed a BCa screening scoring model (BCaSS) based on the levels of miR-1343-3p/miR-6087 and miR-5100/miR-6087. The screening effect of BCaSS is investigated and the findings indicate that it has predictability and distinct advantages. Using a mixed cohort with the largest known sample size to date, we have developed effective screening models for BCa, namely BCaS3miR and BCaSS. The models demonstrated remarkable screening accuracy, indicating potential for the early detection of BCa.
Read full abstract