JMIR Publications

ABSTRACT

Background:

Mobile health applications (apps) have emerged as useful tools for patients and clinicians alike, sharing health information or assisting in clinical decision-making. Two prostate cancer (PCa) risk calculator apps, the Rotterdam and the Coral apps, have both demonstrated better predictive accuracy of PCa and high-grade PCa than traditional Prostate-Specific Antigen (PSA) and/or digital rectal examination (DRE). However, the epidemiology of PCa varies among different populations and therefore the applicability of such apps needs to be evaluated. The present study attempts to be the first to validate the PCa risk calculator apps with both biopsy and prostatectomy cohorts in Taiwan.

Objective:

To validate the PCa risk calculator apps using a Taiwanese cohort of patients. Additionally, to utilize post-prostatectomy pathology outcomes to assess the accuracy of both apps with regard to high-grade PCa.

Methods:

All male patients who had undergone transrectal ultrasound prostate biopsies in a single Taiwanese tertiary medical center between 2012 and 2018 were identified retrospectively. The probabilities of PCa and high-grade PCa (Gleason score ≧7) were calculated utilizing the Rotterdam and Coral apps, and compared with biopsy and prostatectomy results. Calibration was graphically evaluated with the Hosmer-Lemeshow goodness-of-fit test. Discrimination was analyzed utilizing the area under the receiver operator characteristic curve (AUC). Decision curve analysis was performed for clinical utility.

Results:

Overall, 246/1134 (21.7%) patients were diagnosed with PCa, and 155 out of 246 (63%) patients had high-grade PCa according to the biopsy results. After confirmation with prostatectomy pathological outcomes, 25/53 (47.2%) patients were upgraded to high-grade PCa and 1/84 (1.2%) patients were downgraded to low-grade PCa. Only the Rotterdam app demonstrated good calibration for detecting high-grade PCa in the biopsy cohort. The discriminative ability for both PCa (AUC: 0.779 vs 0.687, DeLong’s method: P< 0.001) and high-grade PCa (AUC: 0.862 vs 0.758, DeLong’s method: P< 0.001) was significantly better for the Rotterdam app. In the prostatectomy cohort, there was no significant difference between both apps (AUC: 0.857 vs 0.777, DeLong’s method: P= 0.128).

Conclusions:

The Rotterdam and Coral apps can be applied to the Taiwanese cohort with accuracy. The Rotterdam app outperformed the Coral app in the prediction of PCa and high-grade PCa. Despite the small size of the prostatectomy cohort, both apps to some extent demonstrated the predictive capacity for true high-grade PCa, confirmed by the whole prostate specimen. Without a PCa risk calculator specifically for the Taiwanese population, the Rotterdam app might be a good alternative to enhance the predictive accuracy of current methods for detecting PCa and high-grade PCa. Clinical Trial: VGHKS19-CT3-13