The Four-Kallikrein Panel Is Effective in Identifying Aggressive Prostate Cancer in a Multiethnic Population

Prostate-specific antigen (PSA) is a prognostic factor for prostate cancer, and PSA levels during midlife have been shown to be predictive of subsequent aggressive and lethal disease in men of European and African ancestry (1, 2). However, studies have found that PSA screening leads to considerable overdiagnosis of indolent disease that is unlikely to progress and a large number of men undergoing biopsies and unnecessary invasive treatments that lower quality of life (3–6). Given these limitations, most guidelines recommend against systematic population-based PSA screening (7). Screening tools with improved discriminative ability for aggressive and lethal disease than PSA could greatly enhance clinical decision making and reduce the number of unnecessary biopsies performed and treatments received.

The four-kallikrein (4K) panel, commercially available as the 4Kscore test (OPKO Health Inc.), which consists of blood measures of total, free, and intact PSA and human kallikrein-related peptidase 2 (hK2), has been demonstrated to improve detection of aggressive prostate cancer compared with PSA among men with moderately elevated PSA levels (8–16). A meta-analysis of 11,134 predominantly White participants across 12 studies reported that the 4K panel predicted high-grade prostate cancer with an AUC of 0.81 [95% confidence interval (CI), 0.79–0.83], with AUC increments of improvement ranging from 0.03 to 0.12 compared with PSA (17). However, only three previous 4K investigations included men of African descent (390 men total or 3.5% of the 11,134 men in the meta-analysis; refs. 12, 13, 18), and none included substantial numbers of participants from other non-White populations. It is of particular importance to comprehensively understand the 4K panel effectiveness in multiethnic populations, as prostate cancer risk differs widely by ancestral backgrounds. Among U.S. men, African Americans have an approximately two-fold higher prostate cancer mortality rate compared with Whites and Hispanics (38.9 per 100,000 men vs. 18.1 and 15.8, respectively), whereas Asians/Pacific Islanders have the lowest (8.6 per 100,000; ref. 19). Further, it has been suggested that prostate tumors progress more quickly in men of African ancestry (20).

In this investigation, we evaluated the discriminative performance of the previously established 4K model (10) in a nested case–control prostate cancer study among African American (AA), Latino (LA), Japanese (JA), Native Hawaiian (NH), and White (WH) men in the Multiethnic Cohort (MEC) with moderately elevated PSA levels, as these men are the target population for clinical use of the 4K panel. In secondary analyses, we assessed whether the inclusion of a prostate cancer polygenic risk score (PRS) enhances the discriminative ability of the 4K panel.

The MEC is a large multiethnic prospective cohort established between 1993 and 1996 to investigate risk of cancer and other chronic conditions among individuals living in Hawaii and Los Angeles (21). Between 2001 and 2006, blood samples were collected from ∼67,000 participants for nested case–control cancer studies. For this investigation, prediagnostic plasma levels of free, intact, and total PSA and hK2 were measured among 2,224 incident prostate cancer cases and 2,230 controls, of which 1,667 cases and 691 controls had elevated total PSA (≥2 ng/mL). Cancer cases were identified through linkage to cancer Surveillance, Epidemiology, and End Results (SEER) registries in Hawaii and California, whereas cancer mortality was determined though routine linkages to death certificates through state death files and the National Death Index for deaths that occurred outside of Hawaii and California. With the exception of death due to prostate cancer, disease aggressiveness was defined at the time of diagnosis based on histological status of disease (e.g., Gleason score, stage), which was obtained from SEER registries. Controls were matched to cases using incidence density sampling based on race and ethnicity, birth year, area (Hawaii or Los Angeles), hours of fasting, and year and time of collection. Controls were alive and prostate cancer free at the age of the case diagnosis (see Supplementary Information). Participants included AA, LA, JA, NH, and WH men, with race/ethnicity acquired from a questionnaire completed by participants at baseline (22).

Five primary outcomes were assessed: overall prostate cancer, Gleason 8+ prostate cancer, aggressive prostate cancer, prostate cancer death, and nonaggressive prostate cancer. Gleason 8+ cases were those with a Gleason score ≥8. Aggressive cases were those with either a Gleason score ≥8, non-localized disease, or who later died due to prostate cancer. Nonaggressive cases were defined as cases with a Gleason score ≤7, localized disease, and who did not die due to prostate cancer. The outcomes of aggressive prostate cancer and prostate cancer death excluded 19 ineligible controls who died prior to the matched case's prostate cancer death. We additionally compared aggressive cases to nonaggressive cases. Because of changes in the SEER grading system over time we were unable to differentiate Gleason 7 tumors for a substantial proportion of cases, and thus, we focused the analysis on men with Gleason 8+ tumors.

We obtained informed written consent from all participants, and study protocols were approved by the Institutional Review Boards at the University of Hawaii and the University of Southern California.

Laboratory methods have been described previously (see Supplemantary Information; ref. 12). Samples were processed in two batches, with 38 samples processed in both. Measurements between batches for these 38 samples were highly correlated (using all 38 samples, not limited to those with PSA ≥2 ng/mL): Pearson r(total PSA) = 0.99, r(free PSA) = 0.98, r(intact PSA) = 0.94, and r(hK2) = 0.79 (after excluding one outlying sample with discordant hK2 levels, correlation for hK2 became r = 0.90). In the combined batches, the coefficient of variation was 49.9% for total PSA at the median 2.15 ng/mL, 40.5% for free PSA at the median 0.51 ng/mL, 43.8% for intact PSA at the median 0.28 ng/mL, and 43.3% for hK2 at the median 0.05 ng/mL. Coefficients of variation within batches were comparable (Supplementary Table S1). 4K markers were positively correlated (Supplementary Table S2).

Ethnic-specific differences in 4K marker levels and associations between 4K marker levels and potential prostate cancer-risk factors were assessed using the full sample (not limited those with PSA ≥2 ng/mL). The 4K score was calculated by combining the four biomarkers into a single score, reflecting each participant's predicted probability of receiving a positive biopsy, using the prespecified model (10) equivalent to the commercially available 4Kscore (OPKO Health Inc.). Linear regression models including each of the 4K markers and the 4K score as separate outcomes were used to assess ethnic-specific differences in marker levels, stratifying by case–control status and controlling for matching factors, body mass index (BMI) at blood draw, and laboratory batch. Similar linear models were used to test associations between 4K markers and type 2 diabetes (T2D), BMI, and age at blood draw.

We evaluated the ability of the 4K panel to discriminate between prostate cancer cases and controls by calculating AUCs, implemented in the pROC R package (23), of logistic regression models with covariates included for matching factors, BMI at blood draw, and batch. Model AUCs were compared between models that additionally included (i) total PSA, (ii) total and free PSA, and (iii) the 4K score as variables, focusing on relative AUCs. Analyses were repeated within racial/ethnic groups to assess ethnic-specific AUCs. Ethnic-specific differences in the discriminative ability of the 4K panel were investigated by evaluating model fit using a likelihood ratio test (LRT), comparing the 4K model to a model adding an interaction term for 4K score*ethnicity. Theoretical considerations suggest that the 4K panel would not be of value to men with low PSA due to the difficulty of measuring extremely low levels of PSA-forms (e.g., intact PSA), and indeed, previous studies have shown that this model does not enhance the discriminative ability of PSA in this subgroup (24); thus, primary analyses were restricted to individuals with PSA ≥2 ng/mL. Because this PSA restriction breaks the matching of case and controls, we ran sensitivity analyses limited to sets of matched cases and controls within each outcome, where each participant had PSA ≥2 ng/mL. Sensitivity analyses were also performed excluding cases diagnosed <1 year from the blood draw (n = 148). Analyses were additionally performed stratified by the median age at blood draw (69 years).

In secondary analyses, we investigated whether a prostate cancer PRS could improve the discriminative ability of the 4K panel. Of the 2,358 MEC participants with PSA ≥2 ng/mL, 1,776 had genome-wide imputed genotype data. A weighted PRS was calculated for each participant as the sum of the number of risk alleles carried by an individual, weighted by previously estimated variant-specific effects for 135 prostate cancer-associated variants (see Supplementary Information). AUCs were recalculated for all models in this subset of participants with PRS and for a fourth model adding the PRS as a covariate to the 4K model, along with genetic ancestry using the first 10 principal components from a principal component analysis (see Supplementary Information).

Participant characteristics for those with PSA ≥2 ng/mL used in the primary analysis to assess the discriminative ability of the 4K panel are described in Table 1. The mean ages of cases and controls at blood draw were 69 (range: 47–86) and 72 (54–87), respectively. Samples were drawn an average of 4.1 years (range: <1–18 years) prior to a prostate cancer diagnosis. Characteristics for the full sample used to compare individual 4K marker levels are in Supplementary Table S3 (note that the distribution of participants by race/ethnicity was similar before and after applying the PSA exclusion criterion).

4K marker levels and 4K scores at time of the blood draw were significantly higher in men subsequently diagnosed with prostate cancer compared with those who did not develop prostate cancer (Supplementary Table S4; Supplementary Fig. S1). Among prostate cancer controls, individuals with T2D had slightly lower total PSA and 4K scores than those without T2D (Supplementary Table S5). Older age was associated with higher individual 4K marker levels and 4K scores (Supplementary Table S6). Among cases, African Americans had higher levels of total, free, and intact PSA than Latinos and lower levels of hK2 than Native Hawaiians (Fig. 1), potentially influenced by slightly different distributions of aggressive disease between ethnicities (Table 1). Among controls, 4K scores were slightly higher in African Americans than Japanese, Latinos, and Native Hawaiians (Fig. 1).

Model fit for overall prostate cancer was significantly improved when adding an interaction term for 4K*ethnicity (LRT, P < 0.001), suggesting differences in 4K model performance by ethnicity. Thus, we report overall and ethnic-specific results. Among men with elevated PSA (≥2.0 ng/mL; 1,667 cases and 691 controls), the AUC for overall prostate cancer was 0.748 for the 4K panel compared with 0.711 for free plus total PSA and 0.669 for total PSA alone (Fig. 2; Supplementary Table S7). Discrimination was similarly enhanced for the 4K panel when comparing controls to aggressive disease subsets of cases: the AUC was 0.777 for Gleason 8+ prostate cancer (vs. 0.739 for free plus total PSA and 0.685 for total PSA alone) and 0.782 for aggressive prostate cancer (vs. 0.739 for free plus total PSA and 0.678 for total PSA alone). We observed improved discriminative ability of the 4K panel over PSA when comparing nonaggressive prostate cancer cases to controls (4K AUC of 0.747 vs. 0.726 for free plus total PSA and 0.693 for total PSA alone); however, the magnitude of these improvements was lower for nonaggressive prostate cancer than aggressive outcomes (AUC increment comparing 4K to total PSA: +0.104 and +0.054 for aggressive and nonaggressive prostate cancer, respectively; Fig. 2; Supplementary Table S7). Overall, the 4K panel was less predictive of nonaggressive prostate cancer (AUC = 0.747) than aggressive prostate cancer (AUC = 0.782). When comparing aggressive to nonaggressive cases, the AUC for the 4K panel had slightly improved discriminative ability over PSA, with an AUC of 0.645 for the 4K panel compared with 0.639 for free plus total PSA and 0.632 for total PSA alone (Fig. 2; Supplementary Table S7).

Improvements of the 4K panel over PSA were consistently observed in each racial/ethnic group for overall, high grade, and aggressive prostate cancer, and prostate cancer death; however, prostate cancer death AUCs within populations were unstable due to the smaller number of events. AUC improvements between the 4K and free plus total PSA models were similar between populations, ranging from +0.009 to +0.058 across overall, Gleason 8+, and aggressive prostate cancer. AUC improvements between the 4K and total PSA only models ranged from +0.032 to +0.183, with African Americans having the lowest and Native Hawaiians the highest AUC improvement. Japanese and Latino men consistently had large AUC improvements (Fig. 2; Supplementary Table S7). The 4K model had the highest discriminative ability in Native Hawaiians, followed by Latinos, Japanese, Whites, and African Americans. Similar to the free plus total PSA model, the 4K model consistently had the lowest discriminative ability in African Americans.

AUCs calculated in sensitivity analyses limited to matched cases and controls with PSA ≥2 ng/mL for each outcome were extremely similar to those observed in the complete sample; results stratified by population were highly similar as well, suggesting that our results are robust to the inclusion of all controls in analyses of disease aggressiveness, although some sample sizes were too small to calculate reliable AUC estimates (Supplementary Table S8). Upon removing cases diagnosed less than 1 year after blood draw, results were similar, and the interpretation was identical (results not shown). As expected and consistent with previous findings (24), the 4K model did not notably enhance the discriminative ability of PSA when including individuals with PSA <2 ng/mL (Supplementary Tables S9 and S10).

In analyses stratified by median age at blood draw, the 4K panel typically had improved performance compared with PSA and the combined total plus free PSA across outcomes and populations in men providing a blood draw at >69 and ≤69 years of age. For overall prostate cancer, the discriminative ability of all three models (total PSA, total plus free PSA, and the 4K panel) was higher in older individuals, and the AUC improvement of the 4K panel (compared with PSA and total plus free PSA) was greater in older individuals. This was also observed for the more aggressive categories when including all populations and less consistently within populations. Among Japanese men providing a blood draw at younger ages, the 4K panel had lower AUCs compared with total plus free PSA for overall prostate cancer, Gleason 8+, and aggressive prostate cancer (Supplementary Table S11).

When including the PRS in the 4K model among men with elevated PSA (≥2 ng/mL) and genomic data (1,252 cases and 524 controls), the AUC increased by ∼0.01 in all subgroups (to 0.766 for overall prostate cancer, 0.789 for Gleason 8+ prostate cancer, 0.761 for nonaggressive prostate cancer, and 0.801 for aggressive prostate cancer). Improvements were greater within Native Hawaiians and minimal among Japanese (Supplementary Fig. S2; Supplementary Table S12).

In sensitivity analyses, AUCs were recalculated using 10-fold cross-validation on the full sample. Results were similar and the interpretation was identical (results not shown).

In this prospective investigation, we demonstrate that the 4K panel can be extended to non-White men and that it effectively discriminates overall and aggressive prostate cancer cases from controls among men with PSA ≥2 ng/mL in multiethnic populations. It also discriminates nonaggressive prostate cancer cases from controls, although to a lesser degree than observed for aggressive prostate cancer. The 4K panel performed consistently better than free plus total PSA and total PSA alone in all racial and ethnic populations tested. Compared with total PSA alone, the 4K panel had the least amount of improvement among African American men, with the AUC increment of improvement being about half the increment observed in the full sample across outcomes. We also noted slight but consistent improvements in the discriminative ability of the 4K model with the inclusion of a prostate cancer PRS.

Improved prostate cancer discrimination is of particular importance among African American men, as they have the highest prostate cancer incidence and mortality rates in the United States (19, 21, 25) and tumors with more aggressive characteristics (20, 26). Despite the lower performance of the 4K panel in African Americans compared with other populations, our results suggest that African American men are likely to benefit from the 4K panel versus PSA alone. Three previous 4K investigations that included African American men did not identify significant differences between the performance of the 4K panel in African Americans versus Whites, potentially due to limited sample sizes and the use of samples collected closer to the time of diagnosis (12, 13, 18).

Among the five populations tested, Japanese and Native Hawaiians have the lowest prostate cancer incidence and mortality (27). Despite the 4K model being based on estimates from European men, in our investigation, it had slightly better discriminative ability in Native Hawaiians, Japanese, and Latinos than Whites. However, in analyses stratified by the median age at blood draw, the 4K panel had improved discriminative ability in older Japanese men but did not have improved performance compared with the combined total plus free PSA in younger Japanese men for overall, Gleason 8+, and aggressive prostate cancer. The 4K panel performed particularly well in Native Hawaiians, with AUCs ranging from 0.875 to 0.929 among the four outcomes and large AUC improvements compared to PSA alone. Although this is the largest multiethnic investigation of the 4K panel to date, replication with larger sample sizes, particularly among Native Hawaiians, is needed to verify the comparative performance of the 4K panel between racial and ethnic populations.

Our multiethnic 4K findings are consistent with other largely-European 4K investigations. A study of 4,765 biopsied European ProtecT participants with PSA ≥3 ng/mL reported a high grade prostate cancer (defined as Gleason score ≥7) AUC of 0.82 (95% CI, 0.80–0.84) and overall prostate cancer AUC of 0.72 (95% CI, 0.70–0.73; ref. 10), comparable to the MEC Gleason 8+ prostate cancer AUC of 0.78 (95% CI, 0.75–0.81) and overall prostate cancer AUC of 0.75 (95% CI, 0.73–0.77). The AUC increment of the 4K panel compared with free plus total PSA and PSA alone in this previous investigation is also comparable to our results.

This study was not without limitations. The PSA exclusion criterion greatly reduced our sample size and power for within population analyses, particularly for Native Hawaiians. Only 93 participants in our sample died due to prostate cancer, limiting our power to assess the 4K panel's ethnic-specific ability to discriminate between controls and prostate cancer cases who likely have the most lethal tumors and whose prognosis cannot be determined by stage or grade. Although PSA screening rates were comparable between ethnicities (61% of WH, 59% of AA, 51% of JA, 51% of LA, and 32% of NH reported undergoing clinical PSA screening), lower rates in Native Hawaiians could contribute to the higher 4K performance in this group. The 4K panel may not have performed as well in African Americans because the 4K model was developed within a predominantly White sample (10) and Whites have dissimilar prostate cancer risk from African Americans. This could potentially be improved by developing a new model using a large African American population. Since participants were not all biopsied and we relied on SEER registries and death certificates for prostate cancer outcomes, we cannot eliminate the possibility of misclassification bias.

Given the high mortality burden of prostate cancer, excessive false-positive rates resulting from standard prostate cancer screening measures, and the huge health disparity of this disease, a screening tool that can be effectively and universally applied across racial and ethnic populations would be valuable. Our multiethnic prospective investigation suggests that the 4K panel has superior discriminative ability over total and free PSA to detect prostate cancer, especially for more aggressive disease and to a lesser degree for nonaggressive disease, within and across racial and ethnic populations, implicating the broad clinical applicability of the 4K panel and its potential to improve current prostate cancer screening practices. Future investigations to improve the performance of the 4K model for African American men should be a priority. Until then, race and ethnicity should be taken into consideration when the 4K panel is utilized clinically.

A.J. Vickers has ownership interest (including patents) in Arctic Partners and OPKO Health. H.G. Lilja has ownership interest (including patents) in OPKO Health. No potential conflicts of interest were disclosed by the other authors.

Conception and design: H.G. Lilja, C.A. Haiman

Development of methodology: B.F. Darst, A. Chou, A.J. Vickers, H.G. Lilja

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Pooler, L.R. Wilkens, L. Le Marchand, C.A. Haiman

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): B.F. Darst, A. Chou, P. Wan, E.A. Vertosick, D.V. Conti, L.R. Wilkens, A.J. Vickers, H.G. Lilja, C.A. Haiman

Writing, review, and/or revision of the manuscript: B.F. Darst, A. Chou, E.A. Vertosick, D.V. Conti, L.R. Wilkens, L. Le Marchand, A.J. Vickers, H.G. Lilja, C.A. Haiman

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P. Wan, X. Sheng, E.A. Vertosick, L. Le Marchand, H.G. Lilja

Study supervision: H.G. Lilja, C.A. Haiman

B.F. Darst was supported in part by an award from the Achievement Rewards for College Scientists Foundation Los Angeles Founder Chapter. This work was supported in part by the NIH/NCI with a Cancer Center Support Grant to Memorial Sloan Kettering Cancer Center (P30 CA008748), a SPORE grant in Prostate Cancer to Dr. H. Scher [P50-CA92629], R01CA160816 to H.G. Lilja and A.J. Vickers, the Sidney Kimmel Center for Prostate and Urologic Cancers, David H. Koch through the Prostate Cancer Foundation, and K99CA246063 to B.F. Darst. This work was also supported in part by the Swedish Cancer Society (CAN 2017/559) and the Swedish Research Council (VR-MH project no. 2016-02974). The MEC was supported by NIH/NCI grant U01 CA164973. The authors thank all participants from the Multiethnic Cohort study. They also thank Anqi Wang for support with manuscript preparation.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.