Predicting clinically significant prostate cancer with or without digital rectal exam and MRI data using ClarityDX...

Area of Science:

• Urology
• Medical Imaging
• Machine Learning in Healthcare

Background:

• Accurate prediction of clinically significant prostate cancer (csPCa) is crucial for appropriate patient management.
• Existing diagnostic methods may have limitations in distinguishing csPCa from indolent disease.

Purpose of the Study:

• To develop and validate optimized random forest models for predicting csPCa.
• To evaluate the incremental value of digital rectal examination (DRE) and magnetic resonance imaging (MRI) data in csPCa prediction models.

Main Methods:

• Prognostic study utilizing aggregated observational data from six international organizations.
• Development and validation of calibrated random forest models (ClarityDX Prostate) incorporating PSA, free PSA, biopsy status, age, DRE, and MRI data.
• Models were trained on cohorts (n=1626-2191) and validated on separate cohorts (n=378-1318) from diverse clinical settings.

Main Results:

• All developed models demonstrated high predictive accuracy with ROC AUC values ≥0.80.
• Incorporating DRE improved model performance (ROC AUC=0.82).
• Models utilizing MRI features achieved ROC AUC values of 0.87 (without DRE) and 0.88 (with DRE) in the validation cohort.

Conclusions:

• The ClarityDX Prostate models, particularly those incorporating MRI, offer high accuracy for predicting csPCa.
• These models demonstrate robust performance across variable clinical settings, aiding in clinical decision-making.