Screening Model for Bladder Cancer Early Detection With Serum miRNAs Based on Machine Learning: A Mixed-Cohort Study...

Area of Science:

• Biomarkers and diagnostics
• Oncology
• Machine learning in healthcare

Background:

• Early detection of bladder cancer (BCa) significantly improves patient prognosis.
• Current BCa screening methods lack widespread acceptance.
• There is a need for efficient, noninvasive early BCa screening tools.

Purpose of the Study:

• To develop an efficient, clinically applicable, and noninvasive method for early bladder cancer screening.
• To identify specific serum microRNA (miRNA) levels indicative of BCa.
• To utilize machine learning for creating predictive screening models.

Main Methods:

• A large-scale mixed-cohort study (n=16,189) including BCa patients, other cancers, benign diseases, and healthy individuals.
• Development of BCa screening models using five machine learning algorithms on a training dataset.
• Evaluation of model performance using ROC curve and decision curve analysis on testing, validation, and external datasets.

Main Results:

• The BCaS3miR model, employing three serum miRNAs (miR-6087, miR-1343-3p, miR-5100) and KNN algorithm, demonstrated superior performance.
• BCaS3miR achieved over 90% sensitivity and specificity across testing, validation, and external sets, with AUCs ranging from 0.917 to 0.990.
• A BCa screening scoring model (BCaSS) was developed, showing predictability and advantages in subgroup analyses.

Conclusions:

• Effective bladder cancer screening models, BCaS3miR and BCaSS, were developed using a large mixed cohort.
• These models exhibit remarkable screening accuracy, offering potential for early BCa detection.
• The noninvasive nature and high performance suggest clinical applicability for BCa screening.