MRI-based prostate cancer risk calculators prone to underprediction

New research published in JAMA Network Open shows that magnetic resonance imaging (MRI)-based risk calculators can predict prostate cancer risk among adults in Europe and North America with some accuracy, but remain limited when applied to advanced serum biomarker cohorts.

The researchers indicated that MRI-based models have the potential to augment or replace traditional tools for predicting prostate cancer risk. However, few studies have compared the performance of existing MRI-based risk tools in different screening pathways or countries.

To bridge this research gap, the team compared four MRI-based risk calculators – the Prospective Loyola University Multiparametric MRI (PLUM) tool; the University of California, Los Angeles (UCLA)-Cornell model; the Van Leeuwen model and the Rotterdam Prostate Cancer Risk Calculator–MRI (RPCRC-MRI) – in 2181 patients across three cohorts.

Two of the cohorts consisted of European and North American patients who received an MRI prior to prostate biopsy, while the third cohort relied on an advanced serum biomarker, the Prostate Health Index (PHI), before MRI or biopsy.

Across these cohorts, each model was tasked with identifying patients at risk of being diagnosed with clinically significant prostate cancer. Model performance was measured in terms of receiver operating characteristics for area under the curve (AUC) estimates, calibration plots and decision curve analysis.

All four models demonstrated good diagnostic discrimination in the European and North American cohorts.

In the European cohort, the models achieved an AUC of 0.90, but performance varied slightly when the tools were applied to North American patients.

In the North American cohort, the UCLA-Cornell, Van Leeuwen and RPCRC-MRI models reached AUCs of 0.83, whereas the PLUM tool achieved an AUC of 0.85. However, the RPCRC-MRI and PLUM models demonstrated somewhat better calibration than their counterparts.

Within the PHI cohort, all models were prone to underestimate clinically significant prostate cancer risk. For this cohort, the UCLA-Cornell model and the PLUM model had the best calibration and discrimination, with AUCs of 0.83 and 0.82, respectively.

In the decision curve analysis, each model provided similar net benefit in the European cohort, but the PLUM and RPCRC-MRI models provided higher benefit in the North American cohort. For the PHI cohort, the UCLA-Cornell model demonstrated highest net benefit.

The researchers concluded that these findings support the use of RPCRC-MRI and PLUM in MRI-based screening pathways for prostate cancer. But they underscored that risk models specific to screening pathways incorporating advanced biomarkers are needed to address MRI-based models’ propensity to underpredict risk in some cohorts.

Advances in data analytics and precision medicine are increasingly enabling research teams to design risk prediction models for cancer.

Last week, researchers from Washington University School of Medicine in St. Louis detailed the development of a deep learning approach to help predict which patients with non-small cell lung cancer (NSCLC) are likely to experience brain metastasis.

Brain metastases can occur in a significant portion of patients with early and locally advanced NSCLC, but no method currently exists to identify at-risk patients who may benefit from more aggressive treatments.

To address this, the research team developed a deep learning model that uses lung biopsy images to identify abnormal features associated with brain metastasis. The model was significantly more accurate than four expert pathologists, emphasizing the potential for high-quality risk prediction models in oncology.