Researchers used the recently developed Rotterdam European Randomized Study of Screening for Prostate Cancer-Magnetic Resonance Imaging Risk Calculator (ERSPC-MRI RC) to demonstrate the way in which adjustment of the intercept in the risk-prediction model can ensure accuracy and improve clinical decision making. Their findings were published in European Urology Oncology.

The ERSPC-MRI RC was created with data from 961 European men who received multiparametric magnetic resonance imaging (mpMRI) and MRI-targeted biopsy and systematic biopsy between 2012 and 2017. Because this group was preselected and had undergone mpMRI at a large academic institution, the clinically significant prostate cancer (csPC) prevalence in the development cohort for the calculator was found to be relatively high at 36% (csPC prevalence among currently available MRI-based cohorts ranges from 22% to 43%).

Related Articles

The researchers used a hypothetical cohort with a lower prevalence to demonstrate how the model performed both with and without intercept adjustment. Their findings showed that without adjustment there is poor calibration, which could lead to patients with low-risk csPC being labeled as high risk. “After adjusting for the a priori likelihood, the predicted and actual probabilities in the cohort are highly comparable,” the authors wrote. “This example illustrates that the prevalence adjustment considerably influences the [net benefit].”

Given the risk of overdiagnosis and sepsis associated with prostate biopsy, the use of a multivariable risk stratification tool can be useful for making clinical decisions. Because risk calculators are developed using unique and specific sets of data, the researchers emphasized that clinicians should consider the way in which different factors such as biopsy protocols and clinical work-up may affect a cohort’s heterogeneity. Their findings emphasize the need for centers to adjust risk calculators after external validation. If external validation has not been performed, clinicians can adjust the intercept in a risk calculator to avoid potentially harmful treatment decisions.

The researchers stressed that clinicians who use risk calculators must be aware of center differences — and make adjustments accordingly — as they pertain to a priori likelihood of cancer and risk factor distribution.  

“Clinical data should be readily available when deciding who to biopsy,” the authors concluded. “To fulfill this promise, we caution that a prediction model may need to be recalibrated to correctly represent a patient’s individual risk.”

Reference

Verbeek JFM, Nieboer D, Steyerberg EW, Roobol MJ. Assessing a patient’s individual risk of biopsy-detectable postate cancer: be aware of case mix heterogeneity and a priori likelihood [published online August 17, 2019]. Eur Urol Oncol. doi: 10.1016/j.euo.2019.07.012