Prognostic Utility of the Gleason Grading System Revisions and Histopathological Factors Beyond Gleason Grade

Plain Language Summary

The Gleason score, the most powerful prognostic factor in prostate cancer, has undergone major revisions in 2005 and 2014. While these revisions have changed clinical practice around the world, it remains unclear if they have also improved prostate cancer prognostication. Our study shows that the Gleason score revisions have indeed improved prostate cancer death prediction. We also show that other histopathological factors, including comedonecrosis, high-grade prostatic intraepithelial neoplasia and number of cores with ≥50% cancer, have potential prognostic utility beyond the Gleason score.

Introduction

The Gleason score (GS) is a powerful prognostic factor in prostate cancer (PCa). In 2005 and 2014, the International Society of Urological Pathology (ISUP) conducted major revisions of the Gleason grading system. In 2005, ISUP recommended against assigning Gleason pattern 1 and 2, included most cribriform and poorly formed glands in the definition of Gleason pattern 4, and defined the GS as the sum of the most common and the highest Gleason pattern rather than the first and second most common pattern.¹ In 2014, all cribriform and glomeruloid glands were defined as Gleason pattern 4. Furthermore, ISUP endorsed the five-tired Gleason Grade Group (GGs) system as prognostically more accurate and “patient friendly” than the standard GS grouping.^2,3 The ISUP 2014 revision has been adopted by the World Health Organization (WHO 2016).⁴

A key goal of the ISUP Gleason grading system revisions is to improve prognostication. Whether the revisions actually have improved prognostication, and, if so, what specific changes have led to the improvement, remains unclear. Only a few studies have compared the ISUP 2005 GS to the pre-2005 GS and found either small or no improvement in predicting biochemical recurrence (BCR).^5–7 Most studies assessing the prognostic utility of the GGs performed no central re-review of the diagnostic biopsies according to the ISUP 2014/WHO 2016 criteria,^8–16 but rather simply compared the GGs to different groupings of the GS, such as the three-tiered GS (≤6, 7, ≥8).^10–16 Importantly, no study has evaluated if the ISUP 2014 GS in diagnostic biopsies improves prediction of PCa death, a clinically more important outcome than BCR.

It is possible that the prognostic utility of the Gleason grading system can be further improved by incorporating additional histopathological prognostic factors, such as cribriform pattern and/or intraductal carcinoma,^17,18 percentage of Gleason pattern 4,^19–21 measures of tumor extent^22–25 and perineural invasion.^26,27 No study has, however, performed a central re-review according to the latest ISUP/WHO criteria and systematically evaluated if these or other histopathological features can improve PCa prognostication beyond GS.

Therefore, in this large nested case–control study with centrally re-reviewed diagnostic biopsies, we evaluated i) if the ISUP 2014 GS predicts PCa death better than the previous Gleason grading systems, and ii) if additional histopathological features can further improve PCa death prediction.

Materials (Patients) and Methods

Study Design, Study Population and Data Sources

We performed a case–control study nested in the National Prostate Cancer Register (NPCR) of Sweden. The NPCR contains detailed information on cancer characteristics and primary treatment for 98% of all incident PCa cases in Sweden since 1998.²⁸ Information on cause of death is obtained through linkage to the Swedish Cause of Death Register. PCa death (ICD-10: C61) is defined as death where PCa was the underlying cause of death and has 86–96% agreement with medical records review.^29,30

Cases and controls were selected from all men diagnosed with non-metastatic PCa (ie non-M1) between January 1, 1998 and December 31, 2015 in a subset of the full NPCR (Figure 1). To be able to detect a 0.05 difference in the area under the ROC curve (AUC) for the ISUP 2005 GS vs pre-2005 GS in predicting PCa death with a power of 80% (two-sided z-test, α=0.05), we needed to include 372 cases and 372 controls in the study.³¹ As we further anticipated a 30% reduction in sample size due to inability to retrieve the diagnostic biopsy slides, we decided to include 500 cases and 500 controls in the study. Cases were randomly selected among all men who had died from PCa by December 31, 2015. Controls were then randomly selected among men who were alive at the date of death of the respective case, matched to cases on year and county of diagnosis.

Figure 1 Flow-chart of the patients’ selection. Abbreviation: NPCR, the National Prostate Cancer Register of Sweden. Notes: aBased on the data extracted from NPCR June 5, 2020, but restricted to match conditions at April 11, 2017, when ProMort II was sampled. bIncludes duplicate subjects (cases selected as controls (N=1) and controls selected more than once (N=3)).

The diagnostic biopsy slides were successfully retrieved from the Pathology wards across Sweden for 830 men (83%) (Figure 1). The slides were scanned at 40X at Örebro University Hospital, Örebro, Sweden, using a Pannoramic 250 Flash II digital slide scanner (3DHistech Ltd., Budapest, Hungary). In total, we scanned 5536 slides. The scanned images were uploaded to an internally developed validated virtual microscopy system (Supplementary Methods).³²

Information on age, prostate-specific antigen (PSA), clinical TNM stage, primary and secondary Gleason pattern, and planned primary treatment was abstracted from the NPCR. In addition, for the 830 men whose diagnostic slides we had successfully scanned, we performed a medical chart review. Medical charts were successfully reviewed for 282 cases (69.8%) and 297 controls (69.7%). The medical chart information was used when information abstracted from the NPCR was missing.

Histopathological Review

Two genitourinary pathologists (F.G. and M.F.), with 10 and 21 years of experience as genitourinary pathologists, respectively, performed the histopathological re-review according to the 2016 WHO criteria.⁴ Each case was reviewed by one pathologist. The pathologists were blinded to the diagnostic GS and case–control status. In total, 8982 cores were reviewed, of which 3713 cores contained cancer.

For each core, we circled the core area, recorded the core length and the presence of cancer. For each core with cancer, we circled the tumor area, recorded the tumor length, primary and secondary Gleason pattern, GGs, poorly formed glands, cribriform pattern, comedonecrosis, perineural invasion, intraductal, ductal and mucinous carcinoma. For each core with GS 7, we circled the area of Gleason pattern 4. The percentage of Gleason pattern 4 was calculated as: (area of Gleason pattern 4/tumor area) × 100. Finally, for each slide (but not for each core), we recorded the presence of acute, chronic, periglandular, intraglandular and stromal inflammation, high-grade prostatic intraepithelial neoplasia (HGPIN) and post-atrophic hyperplasia. Case level summaries were calculated as the sum across all cores for continuous features, and as presence in at least one of the cores/slides for binary features. The highest core-level ISUP 2014 GS and GGs was used as an overall score for a case. Case level percentage of Gleason pattern 4 was calculated as the average percentage of Gleason pattern 4 in all cores with GG2-3 for the subjects with a case level GG2-3.

Pre-2005 Gleason Grading

We used two approaches to compare the ISUP 2014 GS with the pre-2005 GS. First, we used information on cribriform pattern and poorly formed glands to approximate the pre-2005 Gleason patterns. Gleason pattern 4 (primary or secondary) was downgraded to Gleason pattern 3 whenever Gleason pattern 4 was assigned based on the i) cribriform pattern only, ii) poorly formed glands only and iii) cribriform pattern and/or poorly formed glands. The pre-2005 GS was calculated as the sum of the back-transformed primary and secondary Gleason pattern. Second, we used the diagnostic GS recorded in the NPCR and in the medical charts as the pre-2005 GS. This approach was restricted to men diagnosed with PCa before 2006.

Statistical Analyses

For the main analysis, we used a complete-case approach (N = 369 cases and N = 369 controls, Figure 1). We used Firth logistic regression adjusted for the matching variables to estimate odds ratios (ORs) and 95% confidence intervals (CIs) of the association between pre-2005 GS, ISUP 2014 GS and GGs, and all other histopathological features and PCa death.³³ As a sensitivity analysis, we repeated all the analyses using conditional logistic regression.

We fit two models for the comparison of the different Gleason grading systems and three models for other histopathological features. Model 1, the “crude” model, was adjusted only for primary treatment. For the comparison of the different Gleason grading systems, Model 2 included primary treatment, age, PSA and clinical tumor stage at diagnosis. For other histopathological characteristics, Model 2 included primary treatment and GGs at diagnosis, and Model 3 included primary treatment, GGs, age, PSA and clinical tumor stage at diagnosis. Model performance was compared by calculating bootstrap corrected AUC.^34,35 Two correlated AUCs were compared using DeLong’s method.³⁶

All analyses were conducted in Stata (version 16.1, StataCorp, College Station, Texas, USA) and R (version 3.5.3, R Foundation for Statistical Computing, Vienna, Austria).

Results

Of 500 cases and 500 controls in ProMort2 (Supplementary Table S1), 369 cases and 369 controls remained for the main analysis; there were no major differences between included and excluded cases/controls (Table 1, Supplementary Table S2). The baseline and re-reviewed histopathological characteristics for the 369 cases and 369 controls are presented in Table 1 and Supplementary Table S3. The distribution of the back-transformed pre-2005 GS is presented in Supplementary Table S4. Compared to controls, cases were on average older (median age: 75.0 vs 68.0 years), had higher PSA (26.0 vs 11.4 ng/mL), clinical tumor stage (≥cT3: 55.6% vs 22.8%) and the diagnostic GS (GS > 7: 46.2% vs 12.6%), and were more often treated without curative intent (73.8% vs 29.8%) (Table 1). Of 707 subjects who had information on both the diagnostic GS and ISUP 2014 GS, 426 (60.3%) were upgraded at re-review, while 46 (6.5%) were downgraded (Figure 2, Supplementary Table S5).

Table 1 Baseline Characteristics Including Centrally Re-Reviewed Histopathological Information Among 369 Cases and 369 Controls Included in the Main Analysis

Figure 2 Gleason score upgrade/downgrade among 707 subjects who had information on both the diagnostic and the centrally re-reviewed ISUP 2014 Gleason score.

The ORs for the associations between pre-2005 GS, ISUP 2014 GS and GGs and PCa death are presented in Table 2. The GGs and ISUP 2014 GS performed equally in discriminating PCa death (Model 1: AUC = 0.807 vs 0.808; Model 2: AUC = 0.818 vs 0.819). The ISUP 2014 GS performed better than the pre-2005 GS back-transformed using only cribriform pattern (Model 1: AUC = 0.785; Model 2: AUC = 0.808) or both cribriform and poorly formed glands (Model 1: AUC = 0.792; Model 2: AUC = 0.814). In contrast, there was no difference in discrimination between the ISUP 2014 GS and the pre-2005 GS back-transformed using only poorly formed glands (Model 1: AUC: 0.800; Model 2: AUC = 0.818) (Table 2). In the analysis restricted to men diagnosed before 2006, the ISUP 2014 GS discriminated death from PCa better than the diagnostic GS (Model 1: AUC = 0.808 vs 0.781; Model 2: AUC = 0.818 vs 0.802) (Supplementary Table S6).

Table 2 Odds Ratios and 95% Confidence Intervals of the Association Between Different Gleason Grading System Revisions and Death from Prostate Cancer Among 369 Cases and 369 Controls

Almost all evaluated histopathological characteristics were associated with PCa death in Model 1 (Figure 3). Comedonecrosis, HGPIN and number of cores with ≥50% cancer predicted PCa death independently from GGs; however, only comedonecrosis (OR: 5.1, 95% CIs: 1.2–21.0) and HGPIN (OR: 0.6, 95% CIs: 0.4–1.0) remained individual predictors in Model 3 (Figure 3). Adding comedonecrosis and HGPIN to the models with GGs, clinical tumor stage, PSA and age changed the discrimination only slightly (AUC = 0.822 and 0.821, respectively) (Supplementary Table S7). Using different cut-offs for percentage of cancer in individual cores had a minor influence on the discrimination (Supplementary Tables S8 and S9).

Figure 3 Odds ratios and 95% confidence intervals of the association between different histopathological features and death from prostate cancer among 369 cases and 369 controls. Notes: aAdjusted for the matching variables (year and county of diagnosis and follow-up time) and primary treatment (deferred treatment (ie active surveillance or watchful waiting), curative treatment (ie radical prostatectomy or radiotherapy) and primary androgen deprivation therapy). bAdjusted for the matching variables, Gleason Grade Groups and primary treatment. cAdjusted for the matching variables, Gleason Grade Groups, clinical tumor stage (cT1, cT2, cT3 and cT4), age (continuous) and PSA (transformed using restricted cubic splines with knots at 4.5, 16 and 105.65 ng/mL) at diagnosis and primary treatment. dOdds ratios per 10% increase. eCalculated as a number of cores with (tumor length/core length) × 100 ≥50. fCalculated as a number of cores with (tumor area/core area) × 100 ≥50. gEvaluated only for the subjects with a case-level Gleason Grade Group 2 and 3.

The results from sensitivity analysis using conditional logistic regression were similar (Supplementary Tables S10–S14).

Discussion

We found that the ISUP 2005 and 2014 revisions of the Gleason grading system have improved PCa prognostication. Our results suggest that the improvement in discrimination was due to classifying cribriform patterns, rather than poorly formed glands, as Gleason pattern 4. We further found that the number of cores with ≥50% cancer, comedonecrosis and HGPIN predicted PCa death independently of the GGs, while comedonecrosis and HGPIN predicted PCa death independently of the GGs, clinical tumor stage, PSA and age at diagnosis. Among men with GG2 and GG3, the percentage of Gleason pattern 4 did not predict PCa death.

The main motivation behind the ISUP 2005 and 2014 revisions was to better define the Gleason patterns, particularly 3 and 4, in order to improve inter-pathologist agreement and, ultimately, prognostication.^1–3 Since ISUP endorsed the GGs in 2014, a plethora of studies have evaluated the ability of the GGs to predict BCR^{10–15,37,38} and/or PCa death.^{8,9,15,16,38–40} However, only four studies had information from centrally re-reviewed diagnostic biopsies,^37–40 and in only one of those studies had the re-review been conducted according to the ISUP 2014 criteria.³⁷ The study modelled the GGs and the diagnostic GS in predicting BCR, but their prognostic performance was not compared.³⁷ Our study is the first study that has centrally re-reviewed diagnostic biopsies according to ISUP 2014/WHO 2016 criteria and has PCa death as the outcome. We approximated pre-2005 Gleason pattern 3 and 4 by using information on cribriform pattern and poorly formed glands. Although imperfect, when using this approximation, the ISUP 2014 GS discriminated PCa death better than the pre-2005 GS. We also confirmed this finding by comparing the ISUP 2014 GS to the diagnostic GS among men diagnosed before 2006. Importantly, when we re-created the pre-2005 GS based on poorly formed glands alone it did not change discrimination whereas when we re-created it based on cribriform patterns it did, indicating that the improvement in discrimination was due to grading all cribriform patterns as Gleason pattern 4.

Different measures of tumor extent, such as the percentages of cores with cancer or total length of cancer in mm, have been evaluated as potential prognostic factors in PCa.^22–25,41 The results for the percentage of cores with cancer are most consistent,^22–24 though the contribution to discrimination seems to be minimal.^25,41 Furthermore, it is included in several risk grouping systems such as the CAPRA score⁴² and NCCN.⁴³ In our study, the percentage of cores with cancer predicted PCa death in the baseline model, but not after adjustment for the GGs. Of all evaluated measures of tumor extent, only the number of cores with ≥50% cancer predicted PCa death independently from the GGs. However, after adjustment for clinical tumor stage, PSA and age at diagnosis, the association disappeared. Taken together with previous studies, our results point to the limited prognostic utility of measures of tumor extent when GS and other standard prognostic factors are available. This has implications for the evolving era of MRI-guided biopsies, as it suggests that the prognostic information lost, in terms of measures of tumor extent, when using MRI-guided biopsies instead of systematic biopsies is not too large.

This is the first study to show that comedonecrosis predicts PCa death independently of the GGs. In one previous study, comedonecrosis in radical prostatectomy samples was associated with BCR.⁴⁴ Given that comedonecrosis is a Gleason pattern 5 feature and men with Gleason pattern 5 are already recommended treatment, the clinical utility of this finding is questionable. However, it is possible that comedonecrosis captures a biological entity with strong prognostic value beyond GS. Finally, in our study, HGPIN was inversely associated with PCa death. To the best of our knowledge, only one previous study has evaluated HGPIN as a predictor of lethal PCa with opposite findings.⁴⁵ We see no obvious explanation for different findings and our results regarding HGPIN should thus be interpreted with care.

Several other histopathological features, such as cribriform pattern and/or intraductal cancer,^17,18,46 perineural invasion,^26,27 and the percentage of Gleason pattern 4,^19–21 have been extensively studied as prognostic factors in PCa. The effect estimates we observed for these features were in line with previous studies; however, they did not predict PCa death independently from the GGs. It is possible that these factors are not predictors when using contemporary ISUP Gleason grading. It is also unclear how histopathological features reported on the core level, such as percentage of Gleason pattern 4, should be used for predictions on the case/patient level. We defined the case level percentage of Gleason pattern 4 as the average proportion of Gleason pattern 4 area in all the cores with GG2 or GG3 cancer for the subjects with the case-level GG2 or GG3. While our definition corresponds to the overall percentage of Gleason pattern 4 described in other studies,^19–21 in all these studies, the percentage of Gleason pattern 4 for subjects with case-level GG4 or GG5 is completely ignored, regardless of how many cores with GG2 or GG3 these subjects have had. By using a single-summary case-level measure, we are thus losing a lot of potentially prognostic information. One of the next steps in improving prognostic models in PCa should thus be finding optimal ways of using the core-level information.

This study has limitations. Our back-transformation to the pre-2005 GS was suboptimal. Gleason pattern 1 and 2 were not assigned at review, and the secondary Gleason pattern was defined as the highest less represented pattern in the core and not the second most common pattern as was done before the ISUP 2005 revision. However, tertiary pattern is uncommon in biopsy cores and the highest Gleason pattern often corresponds to the second most common Gleason pattern. In addition, we also confirmed our results using the diagnostic GS as another proxy of the pre-2005 GS. Another potential limitation of our study is the sample size, and we could have lacked power to identify all independent predictors especially if their effect is small or they are rare.

Conclusion

This is the first study to use information from re-reviewed diagnostic biopsies and PCa death as the outcome. We found that the GGs and ISUP 2014 GS discriminate death from PCa better than pre-2005 GS, and our results suggest that the improvement in discrimination is mostly due to classifying all cribriform patterns as Gleason pattern 4. The number of cores with ≥50% cancer, comedonecrosis and HGPIN predict PCa death independently from the GGs, and comedonecrosis and HGPIN also independently from other clinical variables.