A pre-specified model based on four kallikrein markers in blood improves predictions of adverse pathology and biochemical recurrence after radical prostatectomy

A pre-specified model based on four kallikrein markers in blood, commercially available as 4Kscore, predicts Gleason Grade (GG) 3 + 4 or higher prostate cancer on biopsy. However, sampling error and variation in pathology reporting may miss aggressive disease. The 4Kscore was measured in cryopreserved blood from 2330 men obtained before prostatectomy at a single institution between 2002 and 2010. Adverse surgical pathology and biochemical recurrence (BCR) were pre-specified to be assessed in all men, biopsy GG 3 + 3, and 3 + 4. Adjusted for established clinical predictors, the 4Kscore was significantly associated with adverse pathology (OR 1.49; 95% CI 1.32, 1.67; p < 0.0001). Adding 4Kscore increased discrimination from (AUC) 0.672 to 0.718 and 0.644 to 0.659 within biopsy GG 3 + 3 and 3 + 4, respectively. Higher 4Kscore was associated with higher risk of BCR (HR 1.16, 95% CI 1.06, 1.26; p = 0.001). Adding 4Kscore improved the prediction of BCR (C-index 0.630–0.660) within GG 3 + 3, but not GG 3 + 4. The 4Kscore can help guide the clinical decision whether additional risk assessment—such as confirmatory biopsy—is needed to decide between active surveillance versus curative therapy. Evidence that the panel could influence management in biopsy GG 3 + 4 is less strong and requires further investigation.

low-and intermediate-risk cancer on biopsy. Our objective was to assess the ability of the 4Kscore to predict adverse pathology at prostatectomy-the gold standard for accurate histological diagnosis-and BCR, with a focus on men diagnosed with Gleason 3 + 3 or 3 + 4 prostate cancer at biopsy.

Study design
This retrospective study included 2330 men with localised prostate cancer undergoing radical prostatectomy at Martini-Klinik in Hamburg, Germany, a tertiary referral centre, between 2002 and 2010. All biopsies were 10-12-core transrectal ultrasound-guided biopsy using a standard template, with biopsy and pathological evaluation conducted at the Martini-Klinik. Kallikrein markers were measured in preoperative blood cryopreserved at −80°C. The rate of active surveillance at the time was very low, with almost all patients treated shortly after diagnosis. We excluded patients with missing pathology data at prostatectomy (n = 22), missing kallikrein measurements (n = 5) and suspected non-specific analytical interference in kallikrein measurements (n = 3).
Test methods Sample aliquots were shipped to Dr. Lilja's laboratory at Lund University in Malmö, Sweden for measurements of kallikrein levels conducted in 2016-2017 blind to outcome. Total and free PSA levels were measured using the AutoDelfia 1235 automatic immunoassay system using the dual-label DELFIA Prostatus total/free PSA-Assay (Perkin-Elmer, Turku, Finland) calibrated against the World Health Organization (WHO) 96/670 (PSA-WHO) and WHO 68/668 (free PSA-WHO) standards. Intact PSA and hK2 were measured with F(ab')2 fragments of the monoclonal capture antibodies to reduce the frequency of non-specific assay interference, as described in detail previously. 17,18 To reduce interobserver variability of pathologic specimen, all Gleason Grade 3 + 3 and 3 + 4 biopsies and prostatectomy specimens were read at the Institute of Pathology of the University Clinic Hamburg Eppendorf. Markers were assayed blind to clinical outcome and vice versa.
Statistical methods Adverse pathology at prostatectomy was defined by Brand et al.: 19 primary Gleason pattern 4, any pattern 5 or non-organ-confined disease: seminal vesicle invasion, ECE or lymph node invasion. BCR was defined as a PSA level ≥0.20 ng/mL. Logistic and Cox regression were used to study the association between 4Kscore and adverse pathology and BCR, respectively. Discrimination was assessed by the AUC for adverse pathology and C-index for BCR, comparing the improvement in discrimination by adding 4Kscore to a preoperative clinical base model.
We first defined a clinical logistic model-age at blood draw, total PSA, biopsy Gleason Grade (3 + 3 vs 3 + 4 vs 4 + 3 vs >4 + 3), and clinical tumour stage (<T2b vs ≥T2b)-to predict adverse pathology. Next, we calculated the 4Kscore-using the prespecified formula developed in the ProtecT cohort 20 -for each patient. The 4Kscore-only model was defined using a univariate logistic model to predict adverse pathology. Lastly, we defined a full model by including both 4Kscore and the variables in the clinical model, to predict adverse pathology. For all models, the logit transformation of the 4Kscore was used. Since patients with blood sample available prior to surgery were not representative of the distribution of all radical prostatectomy patients at Martini-Klinik, with a larger proportion having lower Gleason Grade, all logistic models included sampling weights equal to the inverse of the probability of patients with blood sample available based on Gleason Grade.
To confirm whether the 4Kscore offers additional predictive ability after adjusting for the clinical model, we reported the estimates for the 4Kscore from the full model. The predictive accuracy of the clinical model, 4Kscore-only model and the full model was ascertained by calculating bootstrapped (using 200 bootstrap samples) optimism-corrected AUC, pre-specified to be assessed in the two groups of men where a clinical decision needs to be made: biopsy Gleason Grade 3 + 3 and 3 + 4, separately. Clinical utility was assessed using decision-curve analysis. 21 Sensitivity analyses included (i) using a more restrictive definition of adverse pathology with ECE excluded from the definition, (ii) defining the clinical model to include the number of positive cores and millimetres of cancerous tissue on biopsy, (iii) excluding patients with low 4Kscores who may have never had been biopsied and diagnosed, had they received a 4Kscore, (iv) incorporating transrectal ultrasound volume into the clinical models and (v) excluding radical prostatectomy cases from 2002 to 2004, which were graded prior to the 2005 ISUP Modified Gleason System, and therefore patients considered to have pattern 3 disease may be regraded to pattern 4 on the modern grading system. All possible combinations of adverse pathology, clinical models and subgroups as defined above, were assessed.
To assess the association between 4Kscore and BCR, we excluded 195 patients with missing data on recurrence, and 71 men who underwent adjuvant treatment, defined as any additional treatment within 6 months of surgery. Twenty-six men who underwent salvage treatment prior to the recorded date of BCR were considered to have had BCR at the time of treatment. Among the remaining 2064 patients, we used a univariable Cox regression model to assess the association, then created two multivariable Cox models to ascertain whether 4Kscore offered additional predictive ability after adjusting for a preoperative prediction model (PSA, clinical stage and biopsy Gleason Grade) and post-operative prediction model (PSA, Gleason Grade on pathology, ECE, seminal vesicle invasion, lymph node invasion and surgical margin status). We then assessed the association between 4Kscore and BCR in the preoperative setting among men with biopsy Gleason Grade 3 + 3 and 3 + 4, separately, and evaluated the discriminative accuracy by calculating the change in the C index when including 4Kscore. An exploratory analysis assessing whether the association between 4Kscore and outcomes differed based on the expression of five molecular markers-ERG, PTEN, EZH2, FOXA1 and HOXB13-in biopsy tissue is described in full in the supplementary material with distribution of the molecular markers shown in Supplementary Table 1 and the results shown in  Supplementary Tables 2-5. All statistical analyses were conducted using STATA 15.0 (StataCorp, College Station, TX, USA) and R version 3.5.1 (R foundation for Statistical Computing, Vienna, Austria). Table 1. The median age at blood draw was 64 years (IQR 59, 67). Nearly two-thirds of men who underwent prostatectomy had Gleason 3 + 3 cancer at biopsy. A total of 709 men (30%) were found to have adverse pathology. The estimated rate of adjuvant treatment in this cohort was 21%.

Patient characteristics are displayed in
On multivariable regression, 4Kscore was significantly associated with adverse pathology, after adjusting for clinical variables (OR 1.49; 95% CI 1.32, 1.67; p < 0.0001; Table 2). The results are shown in Table 2, with all sensitivity analyses shown in Supplementary Table 6a-6d. The optimism-corrected AUC for the clinical model was 0.672 and 0.644 among patients with biopsy Gleason 3 + 3 and 3 + 4, respectively. Adding 4Kscore to the clinical model increased the AUC to 0.718 and 0.659, respectively ( Table 2). Sensitivity analyses did not importantly change these findings, with clear evidence of benefit in biopsy Gleason 3 + 3 patients and smaller and less consistent benefit in biopsy Gleason 3 + 4 disease ( Table 2 and Supplementary  Table 7a-7d). Decision-curve analysis illustrated the improvement in net benefit of the 4Kscore in appropriate ranges for a decision threshold of 5-20% for biopsy Gleason 3 + 3, and to a lesser extent, 20-60% for biopsy Gleason 3 + 4 (Fig. 1a, b). To better illustrate its clinical relevance, Fig. 2 displays the risk of adverse pathology by 4Kscore for men with biopsy Gleason 3 + 3 or 3 + 4, highlighting the relevance of the 4Kscore for decisions about the confirmatory biopsy and definitive treatment, respectively. The clinical performance of proceeding with a confirmatory biopsy among biopsy Gleason 3 + 3 men at various illustrative cut points is shown in Supplementary Table 8. For example, performing a confirmatory biopsy in men with a clinical + 4Kscore risk greater than 10% in 10,000 men would reduce the number of biopsies by 3086. Of these men avoiding confirmatory biopsy, 195/40 would have adverse pathology with/without ECE.

DISCUSSION
We assessed whether the 4K panel (commercialised by OPKO Health Inc. as the 4Kscore test), helpful in detecting the presence of high-grade cancer within the prostate before a biopsy, 20 could be expanded to help clinicians better predict the presence of adverse pathology within the prostate in men with biopsy grade 3 + 3 or 3 + 4 cancers. Such a tool could substantially improve decision-making by clinicians and patients with the decision of whether to have additional testing, such as confirmatory biopsy, prior to active surveillance (among men with biopsy grade 3 + 3 cancers), or to start active surveillance or have immediate radical surgery (among men with biopsy grade 3 + 4 cancers). We found that 4Kscore was strongly associated with both adverse pathology and BCR among men with biopsy Gleason 3 + 3, and improved the clinical utility of preoperative risk models across an appropriate range of risk thresholds. However, 4Kscore does not improve the value of post-operative risk models and, therefore, does not appear useful for counselling men after prostatectomy regarding the likelihood of recurrence. These findings support the use of the 4Kscore for biopsy decision-making as they suggest that, where grade 3 + 3 or 3 + 4 is to be found, the 4Kscore obtained at the time of biopsy decision-making could be used to make subsequent decisions about clinical management.
Our findings are supported by several prior studies. We have previously demonstrated that free PSA and hK2 enhance the predictive accuracy of clinical models predicting adverse pathology and BCR. [22][23][24][25] In a cohort of 392 men from the Rotterdam arm of the European Randomized Study of Screening for Prostate Cancer treated with radical prostatectomy, predictions based on levels of four kallikrein markers accurately distinguished between pathologically insignificant and aggressive disease (addition of the kallikrein panel increased the AUC to 0.84, p < 0.0005). 26 In a prospective multi-institutional study comprising 1312 men treated with radical prostatectomy at 26 sites in the United States, Punnen et al. showed that the 4Kscore was associated with Gleason score and ECE in the prostatectomy specimen. However, the 4Kscore was not found to improve the prediction of aggressive cancers when added to clinical prediction models, possibly due to small sample size. 27 Regarding the utility of the 4Kscore in the active surveillance setting, Lin et al. prospectively evaluated 718 men enrolled in the multi-institutional Canary PASS trial, demonstrating that 4Kscore improved predictions of high-grade prostate cancer at confirmatory biopsy, but did not add substantive predictive value at subsequent surveillance biopsies. 28 Similar findings were seen in a Spanish study of 137 men on active surveillance, where 4Kscore risk was associated with reclassification at confirmatory biopsy. Among men with 4Kscore below 7.5%, reclassification to Gleason 3 + 4 was missed in 2 men (6%) with no reclassification to Gleason 4 + 3. 29 With 2330 patients, the present study is the largest series evaluating the role of the 4Kscore in predicting adverse pathology

All values are median (IQR) or frequency (proportions).
A pre-specified model based on four kallikrein markers in blood improves. . . A Haese et al. at radical prostatectomy. It is also the first study to evaluate the utility of the 4Kscore for the endpoint of BCR. One limitation of this study is that detailed biopsy pathology with the number of positive cores and millimetres of cancerous tissue was available for 58% of the cohort. Moreover, the percentage of Gleason 4 was lacking in most Gleason 3 + 4 biopsies. As quantitative Gleason grading provides substantial prognostic information in Gleason 3 + 4 carcinomas, 30 the impact of the 4Kscore may be dampened. However, the findings from sensitivity analyses that included detailed biopsy pathology data were similar to the main findings. This suggests that the 4Kscore adds important information about the risk of adverse pathology above and beyond that contained in detailed reporting of biopsy pathology, such as the number of cores and tumour length, which is not routinely reported by pathologists. A second possible limitation is that our study was restricted to a single centre. Our findings on Gleason 3 + 3, related to confirmatory biopsy, replicate those of a prior study; 28 our findings on treatment decision-making in Gleason 3 + 4 disease require further investigation. Finally, our cohort and results are in the pre-MRI era, and the association between 4Kscore, outcomes and MRI is not fully established and requires further research.

CONCLUSION
The 4Kscore strongly predicts adverse pathology and BCR in men with low-grade cancer on biopsy. In practice, the 4Kscore, along with additional tests such as MRI, could assist physicians and their patients in making the critical clinical decision for Gleason 3 + 3 cancers: whether to engage in additional risk assessment, such as a confirmatory biopsy, before initiating active surveillance. Evidence that the 4Kscore improves decision-making in biopsy Gleason 3 + 4 cancer (e.g. active surveillance vs definitive treatment) is less strong, but worthy of further study, especially in cohorts with low volume of Gleason pattern 4. Odds ratios are for a one-point increase when taking the logit of the 4Kscore.  Hazard ratios are for a one-point increase when taking the logit of the 4Kscore.
A pre-specified model based on four kallikrein markers in blood improves. . . A Haese et al.