Girish Kulkarni is supported by a research fellowship from the Canadian Institutes of Health Research.
Clinical predictors of gleason score upgrading
Implications for patients considering watchful waiting, active surveillance, or brachytherapy
Article first published online: 11 MAY 2007
Copyright © 2007 American Cancer Society
Volume 109, Issue 12, pages 2432–2438, 15 June 2007
How to Cite
Kulkarni, G. S., Lockwood, G., Evans, A., Toi, A., Trachtenberg, J., Jewett, M. A. S., Finelli, A. and Fleshner, N. E. (2007), Clinical predictors of gleason score upgrading. Cancer, 109: 2432–2438. doi: 10.1002/cncr.22712
- Issue published online: 4 JUN 2007
- Article first published online: 11 MAY 2007
- Manuscript Accepted: 14 FEB 2007
- Manuscript Revised: 13 FEB 2007
- Manuscript Received: 27 NOV 2006
Brachytherapy, active surveillance, and watchful waiting are increasingly being offered to men with low-risk prostate cancer. However, many of these men harbor undetected high-grade disease (Gleason pattern ≥4). The ability to identify those individuals with occult high-grade disease may help guide treatment decisions in this patient cohort.
The authors identified 175 cases of low-risk prostate cancer treated with radical prostatectomy. By using logistic regression analysis, 11 a priori-defined preoperative risk factors were evaluated for their ability to predict upgrading from Gleason 6 at biopsy to Gleason ≥7 at radical prostatectomy. An internally validated nomogram using all clinical variables was subsequently created to help physicians identify patients who had undetected high-grade disease.
A total of 60 (34%) patients were upgraded to high-grade disease. On multivariate analyses, both prostate-specific antigen (PSA) level (P = .02) and the level of pathologist expertise (P = .007) were predictive of upgrading. The predictive nomogram contained these variables plus age, digital rectal examination, transrectal ultrasound results, biopsy scheme applied (sextant vs extended), presence of prostatic intraepithelial neoplasia, prostate gland volume, and percentage of cancer in the biopsy. The nomogram provided acceptable discrimination (C statistic 0.71).
The authors identified significant predictors of upgrading for patients diagnosed with low-risk prostate cancer. A nomogram based on these study findings could help physicians further risk-stratify patients with low-risk prostate cancer before embarking on treatment. Caution should be exercised in recommending nonradical therapy to individuals with a high probability of undetected high-grade disease. Cancer 2007. © 2007 American Cancer Society.
In the modern prostate-specific antigen (PSA) era, the number of patients diagnosed with low-risk prostate cancer (biopsy Gleason score ≤6, PSA ≤10 ng/mL and clinical stage T1 or T2a), as defined by D'Amico and colleagues,1 has been rising. In fact, half of the patients registered in the Cancer of the Prostate Strategic Urologic Research Endeavor (CaPSURE) database are now classified as low risk, a marked increase from 11% in 1989.2 Given that long-term data demonstrate that patients with low-risk disease are at low risk for death,3 the provision of curative radical therapy to low-risk patients may be considered overtreatment. To this end, less invasive treatment modalities in the form of watchful waiting, active surveillance with selective delayed intervention, and brachytherapy monotherapy have been pursued for patients with low-risk disease.
Implicit in assuming these patients will benefit from nonradical treatment options is the understanding that they have been accurately risk-stratified. However, numerous prostatectomy series have demonstrated a 30%–40% risk of biopsy Gleason score upgrading,4, 5 indicating that several low-risk patients on the basis of biopsy criteria are, in fact, at higher risk. Higher risk is substantiated by the finding that patients with Gleason 6 cancer who are upgraded at prostatectomy have significantly lower biochemical-free survival compared with their nonupgraded counterparts.6 Thus, low-risk patients who harbor occult high-grade (Gleason pattern 4 or 5) prostate cancer fare worse and may not be ideal candidates for current, less invasive treatment regimens.
The identification of clinical risk factors that predict upgrading from low-risk to intermediate-risk or high-risk prostate cancer would help diminish patient-risk misclassification. Physicians could then recommend therapy without inadvertently prescribing nonradical treatment to patients who may actually benefit from a more radical regimen. On the basis of this reasoning, we attempted to identify significant clinical predictors of upgrading of biopsy-derived low-risk prostate cancer by using radical prostatectomy specimens as the gold standard. To facilitate application of our findings, we also constructed a pragmatic tool, in the form of a nomogram, which could aid in clinical decisions for patients with biopsy-derived low-risk prostate cancer.
MATERIALS AND METHODS
We retrospectively reviewed the University Health Network institutional dataset of TRUS (transrectal ultrasound)-guided prostate biopsies from 2000 to 2004. A contemporary sample of all eligible patients was chosen to mitigate the phenomenon of Gleason grade shift present in prostate cancer cohorts accrued over long time periods.7 Only patients who fulfilled the widely accepted D'Amico low-risk categorization for prostate cancer were selected. Thus, inclusion criteria were as follows: 1) a preoperative PSA ≤10.0 ng/mL, 2) a biopsy Gleason score of ≤6, and 3) both TRUS biopsy and radical prostatectomy at our institution. This latter criterion enabled comparison of biopsy and prostatectomy pathology results from the same patients and helped minimize bias related to pathologist inter-rater reliability.
Indications for biopsy included an abnormal age-specific PSA and/or an abnormal DRE (digital rectal examination). All TRUS biopsies were performed by 1 physician (A.T.) A uniform biopsy technique was used for all patients independent of prostate gland volume. TRUS and biopsy were performed by using an end-fire ultrasound transducer and biopsy gun with an 18-gauge needle (ATL HD 13000 with C9-5ICT transducer; Philips, Amsterdam, Netherlands). Prostatic volumes were assessed by using the prolate ellipse formula. Systematic, laterally directed, sextant biopsies were performed for each patient. Hypoechoic lesions and nodules were sampled in addition to the routine sextant pattern but were not submitted separately for pathological review. Toward the end of the study time frame, our institution adopted an extended-pattern, 10-core, biopsy schema that ultimately was performed on 17% of the patients in this study.
Prostatectomy specimens were evaluated by the same uropathology group responsible for interpreting the initial biopsy results. A total of 11 a priori-defined preoperative risk factors were assessed for their ability to predict upgrading from Gleason score ≤6 to Gleason score ≥7. These included 1) age, 2) prebiopsy PSA, 3) DRE results, 4) presence of hypoechoic lesions on TRUS, 5) type of biopsy (ie, systematic sextant or extended) 6) TRUS volume, 7) pathology expertise, 8) number of cores sampled, 9) total cancer volume in the biopsy specimen (<5%, 10%, 15%–40%, ≥50%), 10) presence of prostatic intraepithelial neoplasia (PIN), and 11) presence of inflammation. We were unable to assess the impact of perineural invasion or the number of positive biopsy cores on upgrading because of incomplete data for these 2 variables.
Statistical analyses were performed by using SAS version 9.1 statistical software (SAS Institute, Cary, NC) and R version 2.2.1 (The R Foundation for Statistical Computing, Vienna, Austria). Medians with ranges or frequencies were calculated for each variable for patients whose biopsies were and were not upgraded to high-grade (Gleason pattern ≥4) disease. Single variable logistic regression was performed to ascertain the significance of each factor. PSA and TRUS volume were log-transformed. Upgrade status was then regressed with the same variables in a multiple logistic regression analysis to create a model containing only significant risk factors. This was performed with a nonautomated selection procedure. Presence of inflammation was excluded from all multivariate modeling because of zero cells, which caused nonconvergence, and number of cores sampled was excluded because of excessive missing data (11%). A 2-tailed P value < .05 was considered significant for these analyses.
Because the inclusion of all candidate variables has been shown to minimize overfitting of predictive models when tested in independent datasets,8 a second predictive/nomogram model, using all variables without excessive missing values despite statistical significance, was then developed. The goodness-of-fit of the predictive model was assessed by using the Hosmer-Lemeshow test, and the predictive value was evaluated by using the C statistic. The C statistic is a measure of concordance that quantifies the model's ability to distinguish patients whose biopsy-derived low-risk prostate cancer is upgraded from those patients whose cancers are not upgraded. In logistic regression, the C statistic is identical to the area under the receiver operating characteristic curve. A C statistic between 0.7 and 0.8 indicates a model with acceptable predictive power.9
The regression equation linking predictor variables to outcome was represented as a nomogram, which can easily be used to calculate the predicted probability of upgrading for an individual patient.10 The nomogram is a rescaling of the regression equation where the number of points assigned to each predictor variable is directly proportional to the size of the corresponding regression coefficient, and the total of points for all predictors is scaled to give the resulting predicted probability. The nomogram allows clinicians to use different thresholds for decisions rather than dictating 1 classification rule for all. Bootstrap resampling and refitting was performed (200 iterations) for internal validation to estimate optimism and overfitting bias in the model.
A total of 1475 patients with a PSA level ≤10.0 ng/mL were diagnosed with prostate cancer at the University Health Network between June 2000 and November 2004 (Fig. 1). Of these patients, 369 elected to undergo radical prostatectomy at our center. Selecting only low-grade tumors yielded 175 patients who ultimately fulfilled our inclusion criteria of low-risk prostate cancer. Sixty (34%) were upgraded upon review of their prostatectomy specimens.
Eleven covariates were measured for each patient (Table 1). These variables were selected a priori because of their potential relation to upgrading of prostate cancer from biopsy to radical prostatectomy. On univariate analyses, patient PSA, biopsy inflammation status, and pathologist expertise achieved statistical significance (Table 2). Patients whose biopsies were upgraded had higher PSA values, had their biopsies interpreted by a nonexpert pathologist more frequently, and had inflammation less often. Only patient PSA and pathologist expertise maintained significance on multivariate analyses (Table 3). Patients with higher PSA values (log) and those who had their biopsy interpreted by nonexpert pathologists were more likely to be upgraded, with corresponding odds ratios (OR) of 3.99 (95% confidence interval [CI], 1.34–11.84) and OR of 6.60 (95% CI, 1.61–27.06), respectively. This model yielded a C statistic of 0.67. Further exploration of the effect of PSA was performed by regressing PSA in tertiles (1st tertile, 2.00–4.88; 2nd tertile, 4.89–6.49; 3rd tertile, 6.50–10.00) (Table 4). By using the lowest tertile as the referent category, the OR of the 2nd tertile was 1.01 (95% CI, 0.43–2.37), and the OR for the highest tertile was 2.77 (95% CI, 1.23–6.23), thus indicating that much of the significant effect of PSA is mediated at the upper end of the low-risk range.
|Variables||N = 175|
|Median (range)||60.0 (39.0–79.0)|
|Median (range)||5.59 (2.0–10.0)|
|Digital rectal exam (DRE) positive*||82 (47.1%)|
|Hypoechoic lesion(s) on transrectal ultrasound (TRUS)*||91 (52.3%)|
|Biopsy scheme (% sextant)||155 (87%)|
|Median (range)||42.7 (14.5–137.1)|
|Uropathology interpretation‡||160 (93.0%)|
|No. of cores†|
|Median (range)||8 (5–24)|
|Cancer volume, % of total tissue‡|
|Median (range)||10 (1.00–95.0)|
|Prostatic intraepithelial neoplasia (PIN)||32 (18.3%)|
|Median (range)||61 (49–76)||60 (39–79)|
|Median (range)||6.4 (3.1–10)||5.3 (2–10)|
|Positive||33 (55.9%)||49 (42.6%)|
|Negative||26 (44.1%)||66 (57.4%)|
|Hypoechoic TRUS lesion||.36|
|Present||28 (47.5%)||63 (54.8%)|
|Absent||31 (52.5%)||52 (45.2%)|
|Sextant||47 (78.3%)||98 (85.2%)|
|Extended||13 (21.7%)||17 (14.8%)|
|TRUS volume, cc||.77|
|Median (range)||44.2 (14.5–123)||42.1 (19.7–137)|
|Yes||50 (84.7%)||110 (97.3%)|
|No||9 (15.3%)||3 (2.7%)|
|No. of cores||0.29|
|Median (range)||8 (5–24)||8 (5–18)|
|Cancer volume, % of total tissue||0.37|
|≤5%||15 (25.4%)||31 (27.4%)|
|10%||14 (23.7%)||29 (25.7%)|
|15–40%||18 (30.5%)||35 (31.0%)|
|≥50%||12 (20.3%)||18 (15.9%)|
|Present||14 (23.3%)||18 (15.7%)|
|Absent||46 (76.7%)||97 (84.3%)|
|Present||0 (0%)||7 (6.1%)|
|Absent||60 (100%)||108 (93.9%)|
|Variables||Odds ratio (95% CI)||P|
|PSA, log||3.99 (1.34–11.84)||.02|
|Uropathology interpretation||6.60 (1.61–27.06)||.007|
|PSA tertile, ng/mL||2.00–4.88||4.89–6.49||6.50–10.00|
|OR (95% CI)||1 (ref)||1.01 (0.43–2.37)||2.77 (1.23–6.23)|
The predictive model that used all candidate variables had a C statistic of 0.71. The Hosmer-Lemeshow goodness-of-fit was not significant (P = .25), thus indicating that our model fit these data well. The nomogram is given in Figure 2. The percentage of patients predicted correctly is maximized (76.5%) using a cut point at predicted probability of .54. The optimism corrected C statistic obtained from the bootstrap validation was 0.65.
A paradigm shift is occurring in diagnosing and treating prostate cancer. With PSA screening, more and more patients are being diagnosed with low-risk prostate cancer. These tumors have a low propensity to cause death and thus nonradical therapies are being offered to these individuals. In light of the push toward noninvasive treatment strategies for low-risk prostate cancer, our study has great relevance. We have identified preoperative clinical predictors that can facilitate risk substratification and a mechanism, in the form of a nomogram, which can aid physicians in distinguishing whether their patients have true low-risk prostate cancer or whether they harbor indolent high-grade disease. This latter population may arguably be served best with radical therapies.
To our knowledge, no study has specifically assessed predictors of upgrading in a low-risk prostate cancer population. Other investigators have studied predictors of upgrading but in the context of relatively unselected patients. For example, D'Amico and colleagues4 assessed risk factors for Gleason patterns 4 or 5 at radical prostatectomy in patients diagnosed with Gleason pattern 3 or lower at biopsy. In this report, however, low-risk patients were not exclusively studied, as 26% of patients had PSA values >10 ng/mL. Both King et al.11 and Chun et al.12 also have recently assessed predictors of significant upgrading whereby “significant” was defined as changes in Gleason scores from ≤6 to >6 or 7 to >7 at biopsy and radical prostatectomy, respectively. Both studies included a significant number of subjects (26%) with elevated PSA values of >10 ng/mL. The study by Chun and investigators also included patients with clinical T3 disease. No studies to date have included purely low-risk patients, and, consequently, it is difficult to draw conclusions concerning risk of underlying high-grade disease in low-risk patients from such analyses. Our study differs from past reports in that it focused exclusively on patients who were considered candidates for less invasive treatment modalities (ie, diagnosed with D'Amico-classified low-risk prostate cancer) and in that it considered many variables potentially related to upgrading that others have not.
We demonstrated that patients with low-risk disease have a higher probability of harboring occult pattern 4/5 disease if their PSA value was elevated (especially >6.5 ng/mL) or if their original biopsy Gleason score was derived by a pathologist without expertise in interpreting prostate biopsies. A high PSA value as an indicator of high-grade disease has been demonstrated by others13 and is biologically intuitive, as one would expect more PSA secretion from more aggressive tumors. The potential for higher upgrade rates attributable to unspecialized pathologists was first recognized by Gleason.14 Pathologists with less experience interpreting prostate biopsies may be reluctant to characterize small amounts of high-grade tumor in an otherwise low-grade background given the treatment implications (radical therapy) associated with a high-grade diagnosis.
Prostate volume was not significantly associated with upgrading. In a previous report, we observed that the presence of high-grade disease (Gleason score >6) in all patients who undergo radical prostatectomy was associated with gland volume at biopsy but not at radical prostatectomy, the latter finding secondary to differential upgrading based on prostate volume.15 Prostate volume was not a significant predictor of upgrading in this study because of our selection of patients with biopsy Gleason 6 disease. Patients with low-grade disease at biopsy tend to have larger prostates. Gland size was, therefore, taken into account simply by selecting patients with low-grade disease. In other words, prostate volume significantly impacts the sensitivity of detecting high-grade prostate cancer at biopsy but does not affect the positive predictive value of high-grade (Gleason pattern ≥4) disease at prostatectomy given low-grade disease at biopsy.
Although our study precludes the ability to provide definitive recommendations for patients with a high probability of upgrading, we suggest the following options. Before embarking on a treatment pathway, low-risk individuals with a high probability of upgrading, who have not had expert pathological review of their biopsy specimens, could undergo a uropathological review to decrease the chance of missed indolent high-grade disease. The observation that general pathologists have low interobserver agreement for Gleason scoring compared with urological pathologists on biopsy cases with established uropathological consensus scores, in part secondary to considerable (47%) undergrading of Gleason 7 tumors by general pathologists, supports our assertion.16 Likewise, low-risk patients with an elevated PSA (>6.5 ng/mL) or those for whom expert pathological review is not practical could opt for a repeat biopsy, given that ample evidence demonstrates improved accuracy of Gleason scores with additional biopsy cores.17–19 More studies are required to determine whether these recommendations improve the accuracy of the biopsy Gleason sum for low-risk prostate cancer patients.
Limitations to this study are as follows: First, we have defined predictors of upgrading in patients opting for radical prostatectomy. It is possible that low-risk patients who are willing to undergo surgery differ from low-risk patients who would opt for less invasive strategies in that the former may possess higher volume disease. However, a median percentage of tumor in biopsy specimens of 10% in our study makes this potential shortcoming unlikely to materially affect our results. Second, we did not have data on PSA doubling time, which is often an important variable that guides therapy for patients on active surveillance or for those considering brachytherapy and thus we cannot comment on the ability of PSA doubling time to predict upgrading of low-risk disease. Third, only 2 patients underwent a repeat biopsy before radical prostatectomy. Given such a small number, we cannot discuss the predictive effect of a repeat biopsy on Gleason sum upgrading. Fourth, although we have presented a nomogram that can serve as a useful adjunct when counseling patients with low-risk disease, it has not been externally validated. Nevertheless, the C statistic of our nomogram falls within an acceptable range based on previously published models,20, 21 and the nomogram variables possess face validity. Fifth, it is unclear whether the modernization of biopsy Gleason grading, particularly for high-grade disease based on 2005 International Society of Urological Pathology (ISUP) consensus,22 will affect the utility of our nomogram. Because ISUP consensus is based on uropathological expertise, updating the Gleason scoring system in this manner will not compromise our nomogram if the consensus is slow to disseminate among general pathologists. Should the modernized Gleason grading scheme be rapidly adopted by the pathological community, however, it still remains uncertain whether generalists will assign biopsy scores in a manner concordant with specialists.
Despite these limitations, we feel our methodology is sound and our results important. We have provided clinicians with a list of significant factors associated with upgrading of low-risk disease and have created a tool which can aid clinical decisions. By facilitating identification of individuals who are at high risk of upgrading from low-risk disease status, clinicians can diminish the risk of undertreatment of patients with undetected high-grade cancer and, thereby, pursue treatment regimens with more confidence.
Patients diagnosed with low-risk prostate cancer at biopsy may harbor undetected high-grade disease. Significant predictors of upgrading in this patient population include the PSA value and level of expertise of the pathologist who interprets the biopsy specimen. These variables, among others, have been incorporated into a nomogram that can be used adjunctively when making clinical decisions and when recommending treatment to patients with low-risk prostate cancer.
- 9Applied logistic regression. 2nd ed. New York: John Wiley; 2000., .
- 10Regression modeling strategies: with applications to linear models, logistic regression and survival analysis. New York: Springer-Verlag; 2001..