1 INTRODUCTION

Due to the high intertumoral heterogeneity of urothelial carcinoma, a significant percentage of patients treated with radical cystectomy (RC) for urothelial carcinoma of the bladder (UCB) still experience disease progression.1-3 Accurate identification of patients who could benefit from intensified perioperative systemic therapy remains an unmet clinical need.4 Current prognostic models are mostly based on clinicopathologic features.5-9 Nevertheless, preoperative patient selection for individualized treatment and follow-up scheduling remains challenging as we lack clinically reliable biomarkers for outcome prediction 6, 9-18 To accurately predict biologically and clinically aggressive disease as well as poor survival in UCB patients, novel biomarkers need to improve the current outcome prediction by a prognostically and clinically significant margin.19

Angiogenesis has been proposed as a critical event in the initiation and progression of solid malignancies.20 Endoglin is highly expressed by human vascular endothelial cells and has been reported as a marker of angiogenesis.21 Elevated preoperative plasma levels of endoglin have been associated with worse oncologic outcomes in various malignancies.22-24 Among urological malignancies, higher blood levels of endoglin have been found to be associated with higher preoperative serum prostate-specific antigen, adverse pathologic features, as well as biochemical progression in prostate cancer patients.25-27 The association of circulating levels of endoglin with bladder cancer remains, however, uninvestigated.

We hypothesized that elevated preoperative endoglin plasma levels would be associated with features of biologically and clinically aggressive UCB as well as poor survival outcomes. To test this hypothesis, we studied the predictive and prognostic values of blood levels of endoglin in a large consecutive cohort of patients with nonmetastatic UCB treated with RC and pelvic lymphadenectomy.

2 MATERIALS AND METHODS 2.1 Data source and patient cohort

All procedures described in the present study were undertaken with the approval and oversight of the Institutional Review Board for the Protection of Human Subjects (IRB: 1011011386, 069826900). This study is a retrospective analysis of a prospectively collected consecutive cohort of patients who were treated with RC for nonmetastatic UCB at two medical centers. Sample collection took place between 2003 and 2015. The exclusion criteria were the following: patients with any concomitant secondary malignancies, including upper urinary tract carcinoma, as well as patients with missing data. The extent of lymphadenectomy and choice of urinary diversion was at the surgeon's discretion. No patient received neoadjuvant chemotherapy or radiotherapy. Adjuvant chemotherapy was administered to 167 patients (16.1%) at the clinicians' discretion based on tumor stage and overall health status. No patient received adjuvant radiotherapy.

2.2 Biomarker measurements

Plasma samples were collected after a preoperative overnight fast on the morning of the day of surgery. Specimen collection and measurement have been described in detail elsewhere.28 Briefly, blood was collected into Vacutainer CPT 8-ml tubes containing 0.1 ml of molar sodium citrate (Becton Dickinson) and centrifuged at room temperature for 20 min at 1500g. The top layer corresponding to plasma was decanted using sterile transfer pipettes. The plasma was immediately frozen and stored at −80°C in polypropylene cryopreservation vials (NalgeNunc). For quantitative measurements of endoglin level, we used commercially available quantitative immunoassays (R&D Systems). Every sample was run in duplicate, and the mean was calculated for data analyses. The coefficient of variation was less than 10%.

2.3 Pathological review

All surgical specimens were processed according to standard pathological procedures. Genitourinary pathologists assigned tumor grades according to the 1973 WHO grading system. Pathological stage was reassigned according to the 2002 American Joint Committee on Cancer TNM staging system. The presence of concomitant carcinoma in situ (CIS) was defined as the presence of CIS in conjunction with another tumor other than CIS.29 Pelvic lymph nodes were examined grossly, and all lymphoid tissue was submitted for histological examination. Positive soft tissue surgical margin was defined as the presence of tumor at inked areas of soft tissue on the RC specimen.30 Urethral or ureteral margins were not considered soft tissue surgical margins. Lymphovascular invasion was defined as the unequivocal presence of tumor cells within an endothelium-lined space without underlying muscular walls.31 Any nonorgan confined disease (NOCD) was defined as both ≥pT3 disease and lymph node metastasis.

2.4 Follow-up

Clinical and radiological follow-up was performed in accordance with institutional protocols and current guidelines. Routine follow-up usually included physical examination, radiological imaging (CT of the thorax and abdomen), and urine cytology every 3 months for 2 years. Between the second and the fifth year, follow-up was performed every 6 months. Afterward, in most cases, an annual follow-up was performed. Tumor recurrence was defined as the occurrence of locoregional recurrence or distant metastasis on radiological imaging. Recurrence-free survival (RFS) time was calculated from the date of RC to tumor recurrence or last follow-up. Cause of death was abstracted from medical charts and/or from death certificates.32 Overall survival (OS) time was calculated from the date of RC to death or last follow-up. Cancer-specific survival (CSS) time was calculated from the date of RC to death from disease or last follow-up.

2.5 Statistical analysis

Report of categorical variables included frequencies and proportions. Continuous variables were reported as medians and interquartile ranges (IQR). The median value of endoglin was calculated as 3.142 μg/ml, and this value was used as an independent endoglin cut-off for analysis requiring a categorical variable such as Kaplan–Meier curve analysis and 5-year survival comparisons. For all logistic and Cox regression models as well as for decision curve analysis (DCA), endoglin was considered as a continuous variable. With respect to preoperative plasma level of endoglin, group comparisons were performed using the Mann–Whitney U, Kruskal–Wallis, Wilcoxon rank-sum, Pearson's χ2, or Fisher's exact t tests, and subsequent significance testing, as appropriate.

Binominal logistic regression analysis was performed using preoperative available variables to evaluate the association of preoperative plasma level of endoglin with lymph node metastasis, ≥pT3 disease, or any NOCD. The risk of events was expressed as odds ratios (ORs) and 95% confidence intervals (95% CIs). The area under the curve (AUC) of receiver operating characteristics (ROC) curves was calculated to determine the predictive accuracy of multiple logistic regression models. DeLong's test was used to assess the additional predictive value of preoperative endoglin after adding it to a reference model by comparing the AUCs of the models.

Association between preoperative endoglin with RFS, CSS, and OS was assessed in univariable and multivariable Cox regression models. The risk of survival was expressed as hazard ratios (HRs) and 95% CI. Kaplan–Meier survival curves were used to depict the association between endoglin level and survival. The log-rank test was used to determine the statistical difference between the endoglin (<3.142 and ≥3.142) groups with respect to recurrence or death. Two separate Cox regression models that featured either preoperative clinical variables or postoperative histopathological variables were created. Clinical and pathological tumor grade was excluded as a variable for all predictive models as virtually all RC patients had high-grade UCB. The discriminative ability of the models before and after the inclusion of endoglin was tested and compared using Harrel's concordance indices (C-index) to assess the additional prognostic value of endoglin. The additional clinical net-benefit of endoglin was evaluated using DCA.33 All reported p values were two-sided, and statistical significance was set at 0.05. All statistical analyses were performed using R Version 4.0.4.

3 RESULTS 3.1 Association of preoperative plasma endoglin level with clinicopathologic features

A total of 1036 patients were included in the analysis. The median age of the entire cohort was 67 years (IQR: 60–73). Patient characteristics are shown in Table 1. Median plasma levels of endoglin were significantly higher among patients with adverse pathologic features such as lymphovascular invasion (p < 0.001), lymph node metastasis (p < 0.001), contaminant CIS (p < 0.01), and advanced pathologic tumor stage (p < 0.001).

Table 1. Association of median preoperative plasma level of endoglin with clinicopathologic characteristics in 1036 patients treated with radical cystectomy for urothelial carcinoma of the bladder Variable Median plasma endoglin level, ng/ml (IQR) Stratified by median preoperative plasma level of endoglin

Overall (N = 1036)

31.4 (20.0–39.2) p value

Low ≥ 3.142 μg/ml (N = 515)

High < 3.142 μg/ml (N = 521)

p value Age 67 (60, 73) - - 67 (61, 73) 66 (59, 72) 0.2 Gender 0.71 0.8 Male 814 (79%) 31.4 (19.6–39.1) 406 (79%) 408 (78%) Female 222 (21%) 31.5 (22.2–40.2) 109 (21%) 113 (22%) Blood transfusion 0.26 >0.9 No 768 (74%) 31.5 (19.7–39.1) 381 (74%) 387 (74%) Yes 268 (26%) 31.4 (20.4–39.9) 134 (26%) 134 (26%) Thrombocytosis 0.047 0.2 No 923 (89%) 31.2 (19.6–39.0) 465 (90.3%) 458 (88%) Yes 923 (89%) 33.5 (23.7–41.7) 50 (9.7%) 63 (12%) Hypoalbuminemia 0.06 0.2 No 891 (86%) 31.1 (19.9–39.0) 450 (87%) 441 (85%) Yes 145 (14%) 33.9 (21.8–40.0) 65 (13%) 80 (15%) Clinical tumor grade >0.9 G2 6 (0.6%) 3 (0.6%) 3 (0.6%) G3 1022 (99%) - - 510 (99%) 512 (99%) Unknown 8 2 6 Clinical tumor stage 0.01 0.006 cTa 23 (2.2%) 31.7 (13.3–39.6) 11 (2.1%) 12 (2.3%) cTis 105 (10%) 29.2 (19.4–39.1) 54 (11%) 51 (9.9%) cT1 336 (33%) 29.9 (20.2–37.9) 184 (36%) 152 (29%) cT2 498 (48%) 32.7 (19.5–39.7) 233 (45%) 265 (51%) cT3 38 (3.7%) 37.3 (30.2–43.4) 11 (2.1%) 27 (5.2%) cT4 29 (2.8%) 23.5 (21.0–40.1) 20 (3.9%) 9 (1.7%) Unknown 7 40.4 (27.1–47.7) 2 5 Pathological tumor grade 0.05 0.2 G1 62 (6.0%) 35.3 (24.1–41.5) 25 (4.9%) 37 (7.1%) G2 11 (1.1%) 19.4 (13.9–38.5) 7 (1.4%) 4 (0.8%) G3 963 (93%) 31.3 (19.6–39.1) 483 (94%) 480 (92%) Pathological tumor stage <0.001 <0.001 pT0 62 (6.0%) 35.3 (24.1–41.5) 25 (4.9%) 37 (7.1%) pTa 22 (2.1%) 43.1 (14.1–48.6) 9 (1.7%) 13 (2.5%) pTis 131 (13%) 24.6 (11.1–36.9) 72 (14%) 59 (11%) pT1 162 (16%) 27.1 (12.7–37.4) 96 (19%) 66 (13%) pT2 248 (24%) 27.3 (17.7–36.3) 164 (32%) 84 (16%) pT3 281 (27%) 33.3 (26.0–39.1) 103 (20%) 178 (34%) pT4 130 (13%) 38.6 (23.2–46.1) 46 (8.9%) 84 (16%) Positive soft tissue surgical margins <0.001 0.001 No 941 (91%) 30.8 (19.3–38.8) 483 (93.8%) 458 (88%) Yes 95 (9.2%) 35.6 (25.7–43.7) 32 (6.2%) 63 (12%) Lymphovascular invasion <0.001 0.004 No 741 (72%) 30.1 (18.0–39.0) 389 (76%) 352 (58%) Yes 295 (28%) 33.3 (23.7–40.4) 126 (24%) 169 (32%) Concomitant CIS <0.01 0.5 No 464 (45%) 32.0 (22.2–39.8) 225 (44%) 239 (46%) Yes 572 (55%) 31.1 (18.0–39.1) 290 (56%) 282 (54%) Lymph node involvement <0.001 <0.001 No 773 (75%) 27.3 (16.6–38.5) 439 (85%) 334 (64%) Yes 263 (25%) 34.9 (30.2–41.6) 76 (15%) 187 (36%) Adjuvant chemotherapy <0.001 0.01 No 869 (84%) 30.5 (18.9–39.1) 447 (87%) 422 (81%) Yes 167 (16%) 33.3 (26.5–40.6) 68 (13%) 99 (19%) Note: Median (IQR); n (%). Bold p values are statistically significant.

On multivariable logistic regression modeling, elevated preoperative plasma levels of endoglin were significantly associated with an increased risk of lymph node metastasis, ≥pT3 disease, and any NOCD (all p < 0.001) (Table 2). ROC curve analyses showed that the addition of preoperative plasma levels of endoglin to a reference model comprising age, sex, and clinical tumor stage improved the discriminatory ability for the prediction of lymph node metastasis (10%, p < 0.001), ≥pT3 disease (5%, p < 0.001), and any NOCD (5.8%, p < 0.001).

Table 2. Multivariable logistic regression models for the prediction of lymph node metastasis, ≥pT3 disease, and any nonorgan confined disease in 1029 patients treated with radical cystectomy for urothelial carcinoma of the bladder Variable Lymph node involvement ≥pT3 disease Any nonorgan confined disease OR 95% CI p value OR 95% CI p value OR 95% CI p value Endoglin 1.72 1.52, 1.97 <0.001 1.53 1.37, 1.72 <0.001 1.61 1.44, 1.81 <0.001 Age 1.00 0.98, 1.01 0.9 1.03 1.01, 1.04 <0.001 1.02 1.01, 1.04 0.001 Gender (female) 1.40 0.98, 1.99 0.06 1.02 0.73, 1.42 0.9 1.14 0.82, 1.59 0.4 Clinical tumor stage cTa/cTis/cT1 Ref Ref Ref Ref Ref Ref Ref Ref Ref cT2 2.38 1.72, 3.30 <0.001 2.64 1.99, 3.51 <0.001 2.96 2.25, 3.91 <0.001 ≥cT3 3.13 1.74, 5.55 <0.001 8.83 4.89, 16.7 <0.001 8.25 4.44, 16.2 <0.001 AUC with endoglin 0.733 0.725 0.734 AUC without endoglin 0.629 0.675 0.676 DeLong's test p value <0.001 <0.001 <0.001 Note: Bold p values are statistically significant. Abbreviations: AUC, area under the curve; CI, confidence interval; OR, odds ratio. CI, confidence interval.

On DCA for prediction of lymph node metastasis, the addition of preoperative endoglin plasma levels to the preoperative standard model resulted in the improved clinical net-benefit between a threshold probability of 30%–60% (Figure 1A); 67.8% of patients would benefit from the endoglin model for prediction of lymph node metastasis. DCA for prediction of both ≥pT3 disease and any NOCD revealed that the addition of preoperative plasma level of endoglin to the preoperative standard model resulted only in a slight improvement of the clinical net-benefit (Figure 1B,C). Only 28.1% of patients would benefit from the endoglin model for prediction of ≥pT3 disease, and 27.4% of patients would benefit from the endoglin model for prediction of any NOCD.

Decision curve analyses (DCA) for the evaluation of the clinical net-benefit using the log models for the prediction of (A) lymph node metastasis, (B) ≥ pT3 disease, and (C) any nonorgan confined disease

3.2 Association of survival outcomes within a preoperative model

Median follow-up of patients alive was 37 months (IQR: 14.5–108.5). Overall, the 5-year estimates for RFS, CSS, and OS were 62.5% (95% CI: 59.2%–66%), 66% (95% CI: 63.3%−70%), and 57% (95% CI: 53.6%–60.5%), respectively. In patients with low versus high median levels of preoperative endoglin, the 5-year RFS, CSS, and OS were 71% (95% CI: 67%–76%) versus 53.8% (95% CI: 49%–59%), 77% (95% CI: 72.8%–81.5%) versus 56.5% (95% CI: 51.9%–61.7%), and 67.6% (95% CI: 63%–72.5%) versus 47% (95% CI: 42.4%–52%), respectively. Higher preoperative plasma level of endoglin was associated with worse RFS (HR: 1.85, 95% CI: 1.49–2.31, p < 0.001), CSS (HR: 2.02, 95% CI: 1.60–2.55, p < 0.001), and OS (HR 1.63, 95% CI: 1.38–1.92, p < 0.001) (Figure 2).