Introduction:

The 4Kscore test provides a probability of Grade Group 2 or higher (GG2+) prostate cancer on prostate biopsy and has been validated to improve biopsy decision making in several populations. However, until the 2018 publication of the 4Kscore in a Veterans Affairs (VA) cohort, most participants in these studies have been Caucasian with sparse African American (AA) representation. Since AA men have a higher incidence of aggressive prostate cancer, we examined the calibration of the 4Kscore in AA men with an aim to externally validate a pre-specified coefficient to adjust for the increased risk of GG2+ prostate cancer in AA men.

Methods:

Two statisticians evaluated the literature on AA race and prostate cancer risk, including assessing the calibration of the 4Kscore for detecting GG2+ prostate cancer in 85 AA men who underwent biopsy in the initial prospective US validation study. They pre-specified a coefficient to adjust the 4Kscore for the increased risk of GG2+ prostate cancer in AA men prior to any data collection. Our primary objective was to validate the pre-specified race coefficient in an independent cohort of AA men referred for prostate biopsy in the VA study. We compared the performance of the 4Kscore with and without the race coefficient in AA men from the VA study in addition to AA and non-AA men from the VA study and initial prospective study combined. We assessed model discrimination using AUC from ROC plots, model calibration using calibration plots, and clinical utility using decision curve analysis (DCA).

Results:

In a combined cohort of AA and non-AA men from the prospective US and VA study, the rate of GG2+ disease on prostate biopsy was higher among AAs (38% vs 23%) (Table 1). A pre-specified race coefficient of 0.60 on the log-odds scale (OR 1.82) was used to adjust the 4Kscore for the increased risk of GG2+ prostate cancer in AA men. After the modification of the 4Kscore, the median probability of GG2+ disease increased from 20% to 31% in AA men.  Among AA men in the VA study, including this coefficient improved calibration with slopes approaching the ideal 45° line (Figure 1). The coefficient improved net benefit on DCA within typical threshold ranges for performing a biopsy. In the combined cohort, there was no difference in model discrimination (AUC 0.825 vs 0.828, p = 0.14), calibration, or net benefit suggesting the test performance did not change for most men.

Conclusion:

We externally validated a pre-specified race coefficient to improve the calibration of the 4Kscore in AA. While the performance of the test remained unchanged for most men, this modification provides a more accurate estimation of the risk of GG2+ prostate cancer among AA men. This coefficient for AA men should be incorporated into the 4Kscore.

Funding: OKPO diagnostics