- Top of page
- MATERIALS AND METHODS
- CONFLICT OF INTEREST
If metastatic recurrence occurs in patients with prostate cancer, biochemical recurrence (BCR) can be detected ≈8 years earlier . Moreover, the probability of BCR and metastatic recurrence varies according to baseline characteristics, such as pretreatment PSA levels, clinical stage, biopsy Gleason sum and various biopsy core characteristics. Based on these clinical characteristics, numerous prediction tools have been developed to assess cancer progression risk and mortality . However, many of these tools are restricted to a specific single treatment modality and are designed to predict only one clinical outcome (e.g. BCR or metastatic recurrence) . Moreover, most of these tools do not include thresholds to stratify patients into risk groups for research purposes [3,4]. To circumvent these limitations, Cooperberg et al.  developed the Cancer of the Prostate Risk Assessment (CAPRA) score, which is based on data of men treated in primarily community-based hospitals in the USA. Up to now, the ability of the CAPRA score to predict BCR at the time of diagnosis has been shown in different studies carried out in the USA and Europe [5–7]. Additionally, analyses in US patients with prostate cancer reported a good accuracy for predicting metastasis and mortality .
However, it was suggested that important population differences may exist between patients with prostate cancer from the USA and those from Europe [9–11]. For example, microsimulation analyses, which compared the PSA screening performance in US patients and those European patients who participated in the European Randomized Study of Screening for Prostate Cancer trial, reported differences between both cohorts . Moreover, different selection criteria and aspects, such as the type of hospital (community-based vs tertiary referral centre) and surgical volume of the treating physician (high-volume vs low-volume) may also contribute to differences in outcomes achieved in different geographical regions [11,13,14].
Because the discriminatory power of prediction tools is also affected by these aspects, we used a large cohort of European radical prostatectomy (RP) patients who were treated in a tertiary care referral centre for external validation of the ability of the CAPRA score to predict BCR and metastatic recurrence .
Compared to the development cohort, the European patients harboured more aggressive tumour characteristics. Because the CAPRA score does not account for the positive surgical margin status, which may affect the occurrence of BCR and MR, the present study hypothesis stated that the CAPRA score may perform differently in the validation cohort from Europe compared to the initial development cohort.
- Top of page
- MATERIALS AND METHODS
- CONFLICT OF INTEREST
The clinical and pathological characteristics of the 2937 men in the final study cohort and the distribution of the summed CAPRA scores are listed in Table 1. In the final cohort, the mean (median, range) patient age at diagnosis was 63.2 (64, 48–74) years and the mean (range) CAPRA score was 3.2 (0–10). Overall, 2236 (76.1%) patients had a PSA level ≤ 10 ng/mL or less and 1627 (55.4%) patients had a biopsy Gleason score of 6 (Table 1). Of the patients in the present study, 505 (17.2%) had positive surgical margins, 869 (29.6%) had extracapsular extension, 270 (9.2%) had seminal vesicle invasion and 119 (4.1%) had positive lymph nodes on pathological analyses (Table 2). Overall, BCR was detected in 530 (18.0%) patients and a total of 58 (2.0%) patients developed metastatic recurrence. Mean (median, range) follow-up among men not recurring was 56 (49, 12–220) months.
Table 2. Pathological and clinical outcomes by 8-group University of California Prostate Cancer Risk Assessment (UCSFA–CAPRA) Scoring System and odds ratios (OR) of different pathological and clinical outcomes for each Cancer of the Prostate Risk Assessment (CAPRA) score one-point increase
|CAPRA score (n)|| n ||Positive margins, n (%)||Extracapsular extension, n (%)||Seminal vesicle invasion, n (%)||Lymph node invasion, n (%)||Biochemical recurrence, n (%)||Metastatic recurrence, n (%)|
|0–1||569||47 (8.3)||35 (6.2)||10 (1.8)||1 (0.9)||39 (6.9)||1 (0.2)|
|2||711||107 (15.0)||112 (15.8)||20 (2.8)||7 (3.1)||58 (8.2)||3 (0.4)|
|3||595||94 (15.8)||156 (26.2)||33 (5.5)||5 (1.6)||99 (16.6)||10 (1.7)|
|4||401||72 (18.0)||161 (40.1)||42 (10.5)||11 (3.9)||92 (22.9)||10 (2.4)|
|5||274||71 (25.9)||151 (55.1)||57 (20.8)||26 (11.0)||86 (32.8)||10 (3.7)|
|6||201||44 (21.9)||112 (55.7)||44 (21.9)||21 (10.8)||67 (33.3)||9 (4.5)|
|7||121||42 (34.7)||86 (71.1)||33 (27.3)||26 (23.0)||52 (43.0)||8 (6.6)|
|8–10||65||28 (43.1)||56 (86.2)||31 (47.7)||22 (34.4)||37 (56.9)||7 (10.8)|
|OR (CI)|| ||1.27 (1.21–1.34)*||1.77 (1.68–1.87)*||1.71 (1.60–1.84)*||1.77 (1.57–1.98)*||1.44 (1.35–1.54)*||1.61 (1.42–1.86)*|
When pathological outcomes were tabulated according to the CAPRA scores, steady statistically significant increases in rates of each pathological outcome (positive margins, extracapsular extension, seminal vesicle involvement and lymph node involvement) were detected (Table 2). Accordingly, the rate of BCR and metastatic recurrence also increased from 6.9% and 0.2%, respectively, for patients within the lowest CAPRA score group (0–1) compared to 56.9% and 10.8%, respectively, for patients within the highest CAPRA score group (8–10) (Table 2). When treated as a continuous variable across the full spectrum of CAPRA scores, the odds ratios for each one-point increase for the different pathological outcomes ranged from 1.27 (95% CI, 1.21–1.34) for positive surgical margins to 1.77 (95% CI, 1.57–1.98) for positive lymph node invasion (all P < 0.001). Similarly, the hazard ratios for BCR and metastatic recurrence were 1.44 (95% CI, 1.35–1.54) and 1.61 (95% CI, 1.42–1.86), respectively, for each one-point increase (all P < 0.001). Interestingly, sensitivity analyses for predicting BCR and metastatic recurrence showed that no statistically significant increase between patients within CAPRA scores of 0–1 and 2 could be detected (Table 2).
The hazard ratios for BCR and metastatic recurrence at each categorical CAPRA score are presented in Table 3. Relative to CAPRA scores of 0–1, hazard ratios for BCR and metastatic recurrence ranged from 2.3 (95% CI, 1.6–3.4) to 12.8 (95% CI, 8.2–20.0) and 3.8 (95% CI, 1.2–16.5) to 16.9 (95% CI, 4.8–77.2), respectively, for CAPRA scores of 3 and 8–10 (all P < 0.05) (Table 3). Accordingly, the 3- and 5-year actuarial BCR estimates for CAPRA scores of 0–1 and 8–10 fell steadily from 95.3% to 50.6% and 90.8% to 29.2%, respectively, for the same categories. Similarly, the 3- and 5-year metastases-free survival estimates for CAPRA scores of 0–1 and 8–10 fell from 99.3% to 92.1% and 99.3% to 83.6%, respectively, for the same categories (Table 3).
Table 3. Results of Cox regression model and Kaplan–Meier analyses
|CAPRA score||BCR||Metastatic recurrence-free survival||Recurrence-free survival||Metastasis-free survival|
|HR||95% CI|| P ||HR||95% CI|| P || Probability || Probability |
|3 years (95% CI)||5 years (95% CI)||3 years (95% CI)||5 years (95% CI)|
|0–1||1.0||Reference||–||1.0||Reference||–||95.3 (92.8–96.9)||90.8 (87.2–93.5)||99.3 (97.7–99.8)||99.3 (97.7–99.8)|
|2||1.1||0.7–1.7||0.606||1.0||0.2–4.9||0.955||95.1 (93.1–96.6)||90.7 (87.7–93.1)||100 (–)||100 (–)|
|3||2.3||1.6–3.4||<0.001||3.8||1.2–16.5||0.018||87.3 (84.0–90.0)||83.3 (79.4–86.6)||100 (–)||99.3 (97.3–99.8)|
|4||3.3||2.3–4.8||<0.001||4.4||1.4–19.6||0.009||82.3 (77.8–86.1)||73.9 (68.3–78.8)||99.2 (97.4–99.7)||98.2 (95.5–99.3)|
|5||5.5||3.9–8.0||<0.001||6.7||2.1–29.5||0.001||69.7 (63.1–75.5)||60.9 (53.6–67.8)||98.6 (95.5–99.5)||97.6 (93.4–99.2)|
|6||5.5||3.7–8.1||<0.001||7.8||2.3–35.4||0.001||72.3 (64.8–78.6)||62.1 (53.3–70.0)||97.1 (93.1–98.8)||95.8 (90.5–98.2)|
|7||9.0||6.0–13.6||<0.001||13.4||3.9–61.4||<0.001||54.0 (43.9–63.7)||48.4 (36.9–60.0)||94.2 (86.6–97.6)||92.2 (83.3–96.6)|
|8–10||12.8||8.2–20.0||<0.001||16.9||4.8–77.2||<0.001||50.6 (38.9–64.2)||29.2 (16.9–45.5)||92.1 (80.5–97.1)||83.6 (68.7–92.2)|
|0–2||1.0||Reference||–||1.0||Reference||–||95.4 (94.0–96.6)||91.5 (89.4–93.2)||99.8 (99.2–100.0)||99.7 (99.0–99.9)|
|3–5||3.1||2.4–3.9||<0.001||4.7||2.2–11.6||<0.001||82.0 (79.5–84.3)||75.6 (72.5–78.4)||99.4 (98.7–99.7)||98.6 (97.4–99.2)|
|6–10||7.0||5.5–9.0||<0.001||11.5||5.3–28.9||<0.001||63.1 (57.5–68.2)||52.7 (46.3–58.9)||95.3 (92.2–97.2)||92.2 (87.8–95.1)|
When patients were stratified into three groups according to low (CAPRA scores of 0–2), intermediate (CAPRA scores of 3–5) and high risk (CAPRA scores of 6–10), 1280 (43.6%), 1270 (43.2%) and 387 (13.2%) of patients were within each group (Table 3). Use of the Kaplan–Meier method for the low-, intermediate- and high-risk CAPRA score groups showed a 3-year BCR-free survival rate of 95.4%, 82.0% and 63.1% and a 5-year BCR-free survival rate of 91.5%, 75.6% and 52.7%, respectively (P < 0.001) (Fig. 1 and Table 3). Similarly, for the CAPRA low-, intermediate- and high-risk groups, 3-year metastasis-free survival was 99.8%, 99.4% and 95.3% and 5-year metastasis-free survival was 99.7%, 98.6% and 92.2%, respectively (P < 0.001) (Fig. 2 and Table 3).
Figure 1. A, Actuarial survival curves by Cancer of the Prostate Risk Assessment (CAPRA) score for biochemical recurrence (BCR)-free survival. B, Actuarial survival curves by CAPRA score for BCR-free survival among patients with CAPRA scores categorized by risk level as low risk (0–2 points), intermediate risk (3–5 points) and high risk (6–10 points).
Download figure to PowerPoint
Figure 2. A, Actuarial survival curves by Cancer of the Prostate Risk Assessment (CAPRA) score for metastatic recurrence-free survival. B, Actuarial survival curves by CAPRA score for metastatic-free survival among patients with CAPRA scores categorized by risk level as as low risk (0–2 points), intermediate risk (3–5 points) and high risk (6–10 points).
Download figure to PowerPoint
Finally, the 8-group CAPRA score was accurate in predicting BCR and metastatic recurrence and discriminating with a c-index of 76.2% and 78.5%, respectively. Similarly, the 3-group CAPRA score features a c-index of 71.4% and 75.0% for predictions of BCR and metastatic recurrence. Finally, the calibration plot of the 8-group CAPRA score shows graphically that the predictions closely approximated the outcomes, and only slightly underestimated the risk of metastatic recurrence within the high probability range of recurrence (Fig. 3.)
Figure 3. Calibration plot for 8-group Cancer of the Prostate Risk Assessment score predicting metastatic recurrence.
Download figure to PowerPoint
- Top of page
- MATERIALS AND METHODS
- CONFLICT OF INTEREST
Subsequent to publication of the original preoperative nomogram by Kattan et al. , the number of prediction tools for counselling patients with prostate cancer has grown rapidly in the last decade. The usefulness and generalizability of several instruments has been tested in external validation studies, which are mandatory before recommending such instruments for daily clinical use . Recently, the widespread and commonly used nomogram by Stephenson et al.  and the three-level classification by D`Amico et al.  were enriched by the CAPRA score . The CAPRA score can be calculated without paper nomograms, look-up tables or computer software, and also allows the prediction of clinical outcomes in patients with prostate cancer after different treatment modalities . It was originally developed based on data of 1439 US patients who were treated in community-based hospitals . A multi-institutional validation, including 1346 patients treated in US veterans affairs and military hospitals, confirmed the validity of the CAPRA score for predicting BCR and pathological outcomes . In these studies, the accuracy of the CAPRA score was consistently comparable with the Kattan nomogram (c-index range, 0.68–0.78) [20,22,25,26].
Because the discriminatory power and generalizability of prediction tools can best be validated in cohorts that provide different distribution patterns of risk factors, we used a cohort of European RP patients with important differences compared to the initial development cohort. First, the patients evaluated in the present study stem from Europe, where the performance of PSA screening is different compared to those patients from the USA, as recently suggested by Wever et al. . For example, after stratification of both patient cohorts according to the 3-group CAPRA scores, a direct comparison between the present study cohort and the development cohort showed worse tumour characteristics in the European cohort. Specifically, the original development cohort from the USA includes a higher proportion of patients within the low-risk group (57.8% vs 43.8%). Moreover, the proportion of high-risk patients was only 4.9% compared to 13.2% in the European cohort . Similarly, at final pathology, a higher proportion of European patients harboured unfavourable pathological characteristics (e.g. extracapsular extension and seminal vesicle invasion).
Second, despite a higher proportion of men with unfavourable tumour characteristics among patients in the present study, a smaller proportion of patients was diagnosed with positive surgical margins at final pathology. Similarly, comparison of 3- and 5-year BCR-free and metastasis-free recurrence rates showed a lower BCR rate among European patients. Both findings may be a result of most patients in the European cohort being operated on by high-volume surgeons, contributing to lower surgical margin rates [13,14].
Similar to the previous analyses of Cooperberg et al. , the results obtained in the present study indicate a good discrimination of patients at different risks for BCR and metastatic recurrence when using the 8-group CAPRA scores. Similarly, the 3-group CAPRA score also allows easy discrimination between low-, intermediate- and high-risk patients with a slightly lower predictive accuracy. A deviation of the otherwise high consistency between the data presented by Cooperberg et al.  and the data of the present study was identified for a subgroup of patients when studying the risk of BCR and metastatic recurrence using Cox regression models. Specifically, no statistically significant difference can be detected between patients in the group with a CAPRA score of 2 and the group with a CAPRA score of 0–1. These findings may be explained by the overall low rate of BCR and metastatic recurrence among low-risk patients. Sub-analyses of those two patients who unexpectedly experienced metastatic recurrence after being classified after surgery into the CAPRA low-risk category showed that both patients harboured seminal vesicle invasion and high-risk Gleason grade at final pathology.
The overall discriminative power of the CAPRA score was shown by a concordance index of 76.2, which was substantially higher compared to the concordance index of 68.0 initially reported by Cooperberg et al. . Moreover, the graphical representation of the calibration plot showed that the predictions closely approximate the outcomes. Interestingly, the accuracy for prediction of metastatic recurrence (c= 78.5) was markedly superior to the findings for BCR. Both findings may be explained by the previously described better performance of the CAPRA score in a tertiary care setting and the higher validity of metastatic recurrence as a clinical endpoint compared to BCR [6,20].
However, similar to the findings of the present study, external validation of the CAPRA score for prediction of BCR in US patients showed a higher concordance index in series from high-volume institutions compared to the initial development cohort. Specifically, the accuracy was 76.4% and 76% in a cohort from the Johns Hopkins and Northwestern compared to 66% of the initial CaPSURE (Cancer of the Prostate Strategic Urologic Research Endeavor) dataset. Based on the known difficulties of a reliable Gleason score assignment, the experience of the involved high-volume uropathologist may also contribute to the differences detected.
Despite its strength, the present study does have limitations. The first limitation hinges on the relatively short follow-up because the initial aim of the present study was to examine the ability to predict further clinical endpoints such as cancer-specific survival. Another limitation hinges on the lack of information on timing of salvage radiotherapy and androgen deprivation therapy. Both treatments were administered at the discretion of the treating urologists and information on timing was available only in a few patients. However, sub-analyses of those patients in whom all information were available for analyses showed that androgen deprivation therapy and salvage radiotherapy were always administered in accordance with current guidelines. Moreover, the number of patients experiencing metastatic recurrence was relatively small, based on the natural history of prostate cancer. Therefore, a longer follow-up is necessary and should provide more events, including further endpoints such as cancer-specific mortality. However, the validation study by Cooperberg et al.  also reported on a median follow-up of 34 months and another validation study from the USA assessed the biochemical recurrence rate after a median follow up of 42.6 months .
Second, the definition of BCR (0.2 mg/dL) may also have influenced the results obtained in the present study because the threshold of 0.4 mg/dL emerged in some series as the optimal PSA level threshold when defining BCR . However, the calibration plot for BCR also shows a fairly good calibration of the 8-group CAPRA score in the present study cohort, where BCR was diagnosed based on a threshold of 0.2 mg/dL.
Third, we recently detected a shift towards less favourable tumour characteristics in our centre . Therefore, a larger proportion of patients included in the present series may harbour less favourable tumour characteristics compared to other institutions.
Finally, being operated on by high-volume surgeons influences the surgical margin and overall complication rate and therefore also affects the endpoints examined. Interestingly, despite not accounting for the surgical margin status, the CAPRA score performed better in series from tertiary care institutions, with fewer clinicians and more consistent treatment patterns compared to findings from previous series from the CaPSURE and The Shared Equal Access Regional Cancer Hospital (SEARCH) databases [6,20]. Unfortunately, the database in the present study cannot directly address any potential differences of the discriminative power of the CAPRA score between patients treated in a European, non-academic and academic setting.
More precise predictions in patients with prostate cancer can be achieved based on postoperative data by using nomograms and the newly developed CAPRA-S score. For example, the recently developed CAPRA-S score includes information on PSA levels, margin status, seminal vesicle involvement, specimen Gleason score, extracapsular extension and lymph node involvement after prostatectomy . When applying the CAPRA-S score to the present study cohort, survival analyses showed a slightly higher predictive accuracy of the CAPRA-S score for the prediction of biochemical and metastatic recurrence (78% and 85%) compared to the CAPRA score, which relies on preoperative data. However, only a few pairwise comparisons among adjacent CAPRA-S scores (CAPRA-S score 1 vs 0; 2 vs 1, etc.) were statistically significant.
In conclusion, the European patients included in the present analyses harbour worse tumour characteristics compared to those US patients from the original development cohort of the CAPRA score. However, a lower surgical margin rate, BCR and metastatic recurrence rate were detected in European patients. Despite these differences in patient characteristics and outcomes achieved, the present analyses showed a good discriminatory power of the CAPRA score for predicting BCR and metastatic recurrence. Therefore, the CAPRA score can be used with high accuracy for predicting these endpoints in European patients who are treated in tertiary referral centres.