Predicting Clinically Significant Prostate Cancer with or Without Digital Rectal Exam and MRI Data Using Claritydx Prostate Models

Abstract

Background: In patients suspected of prostate cancer (PCa), the decision to biopsy hinges on available clinical data. Magnetic resonance imaging (MRI) and digital rectal exam (DRE) data can be informative but may be unavailable. We report creating models to predict clinically significant PCa (csPCa, defined as grade group ≥2 PCa), which use total prostate-specific antigen (PSA), free PSA, prior negative biopsy status, age, and DRE and MRI data if available. 

Methods: This was a prognostic study to create optimized ensembles of calibrated random forest models predicting csPCa using total PSA, free PSA, prior negative biopsy status, and age, with or without DRE and MRI data (prostate volume and PI-RADS score). Observational data was aggregated from cohorts in five centers including the University of California (UCLA), Los Angeles, USA, Johns Hopkins University (JHU), Baltimore, USA, University of Calgary (UC), Canada, University of Alberta (UA), Edmonton, Canada, and Thomayer University Hospital (TUH), Prague, Czechia. Eligibility requirements included patients 1) undergoing a prostate biopsy due to an elevated PSA or palpable DRE abnormality, 2) without a prior PCa diagnosis, and 3) having available data for model inputs including age, total PSA, free PSA, as well as prostate volume from MRI and PI-RADS for models using MRI data. Prostate biopsies were performed between September 2009 and April 2023. The risk models (ClarityDX Prostate DRE, ClarityDX Prostate MRI, and ClarityDX Prostate DRE + MRI) were derived (training cohorts n=1626 to 2191) and validated (validation cohorts n=317 to 1257) from different clinical sites. 

Findings: In this prognostic study the created models had ROC AUC values ≥0.80. Adding DRE improved the model’s ROC AUC to 0.82 while models using MRI features, with or without DRE, had ROC AUC values of 0.87 in the validation cohort. The primary outcome was the accurate prediction of csPCa with or without MRI or DRE data in individuals prior to prostate biopsy. 

Interpretation: These optimized risk models provide high accuracy for predicting csPCa in multiple clinical settings and support the use of the four ClarityDX Prostate models for predicting csPCa in individuals in variable clinical settings. 

Funding: Alberta Innovates –ASBIRI Program, Bird Dogs - Alberta Cancer Foundation: Prostate Cancer Research Plan, TELUS Ride for Dad and the Prostate Cancer Fight Foundation, and Prostate Cancer Canada.

Declaration of Interest: The authors declare that M.E.H., A.F., and C.J.W. are Nanostics consultants. A.A. is a Nanostics board member. R.J.P, D.P, C.V., P.H.B., and J.D.L. are Nanostics employees. All other authors have no competing interests to disclose.

Ethical Approval: The Health Research Ethics Board of Alberta approved the collection of patient data from UC and TUH (HREBA.CC-18-0241) plus UA (19-0109). The UCLA Institutional Review Board approved the collection of patient data from UCLA (IRB #11-001580 and IRB #19-001136). The collection of data from JHU was approved by research project CR00040216.

Keywords: clinically significant prostate cancer, Diagnostic Tests Routine, machine learning, prostatic neoplasms, risk score