- To externally validate currently available bladder cancer nomograms for prediction of all-cause survival (ACS), cancer-specific survival (CSS), other-cause mortality (OCM) and progression-free survival (PFS).
In 2012, an estimated 73 510 new cases of bladder cancer were diagnosed in USA . At initial diagnosis, 20–30% of tumours are muscle-invasive and hence fatal for most patients within 2 years, if left untreated. Radical cystectomy (RC) and pelvic lymphadenectomy (PLND) provide durable long-term survival and are therefore considered as the standard of care for clinically localised muscle-invasive bladder cancer and high-grade recurrent non-muscle-invasive bladder cancer . Nevertheless, the procedure still has a 50% 2-year risk of distant metastasis and 60% 5-year risk of death in the setting of muscle-invasive disease . Oncological outcomes after RC depend on specific variables, including tumour stage, lymph node yield and positivity, lymphovascular invasion (LVI), surgical margin status, and neoadjuvant chemotherapy .
Prognostic-categorisation using preoperative data, such as, the clinical stage has limited accuracy, with a 40–60% rate of upstaging at RC [3-5]. In addition, comorbidities, advanced age and competing causes of mortality make the process of formulating treatment plan challenging.
Numerous prognostic variables must be considered to formulate appropriate individual treatment plans for patients with bladder cancer. Nomograms integrating several variables to predict disease outcome have been used as a tool for risk stratification. Clinical judgment has been shown to be less accurate than nomograms in prostate cancer literature and can lead to recall bias [6, 7]. Furthermore, nomogram-derived data may guide the patients through the decision-making process. Applicability to different patient populations through external validation is mandatory for a prediction tool to prove useful. We compared most of the currently available preoperative and postoperative bladder cancer nomograms for prediction of all-cause survival (ACS), cancer-specific survival (CSS), other-cause mortality (OCM), progression-free survival (PFS), advanced pT stage, and risk of lymph of node metastasis using the Bladder Cancer Research Consortium (BCRC), International Bladder Cancer Nomogram Consortium (IBCNC) and Lughezzani nomograms.
Data from a prospectively maintained quality assurance database for 282 consecutive patients who underwent robot-assisted RC (RARC) from 2005 to 2012 by single surgeon (K.A.G.) were used. Clinical and pathological data were analysed for demographics (age, gender), preoperative disease characteristics (preoperative chemotherapy, radiation, clinical stage, clinical grade, presence of LVI and carcinoma in situ), pathological characteristics (tumour stage, lymph node yield, and number of positive nodes) and postoperative data (recurrence, ACS and CSS). Patients with non-TCC and patients with missing preoperative, pathological and postoperative variables were excluded. Data required for the nomograms was available for 236 patients. The technique for RARC and PLND has been described previously . All pathological specimens were reviewed by the dedicated genitourinary pathologists at Roswell Park Cancer Institute, Buffalo, NY, USA. Tumour grade and pathological staging was determined based on the 2002 American Joint Committee on Cancer (AJCC) TNM staging system and the 2004 WHO/International Society of Urologic Pathologists classification of papillary urothelial neoplasms, respectively . Postoperative surveillance was performed based on the National Comprehensive Cancer Network (NCCN) guidelines, and included evaluation of urine cytology, liver function tests, serum creatinine, and electrolytes every 3–6 months for 2 years and then as indicated. Chest, abdomen and pelvis imaging were conducted every 3–12 months for 2 years based on the risk of recurrence and then as clinically indicated. Outcome measures consisted of CSS, ACS, PFS, pathological tumour stage, and pathological lymph node metastasis. Causes of death were identified from death certificates or physician correspondence. Bladder cancer was considered as a cause of death for patients with progressive and highly symptomatic metastases at the time of death.
Descriptive statistics, such as frequencies and relative frequencies, were computed for all categorical variables. Numeric variables were summarised using simple descriptive statistics, such as the mean, standard deviation, range, etc. For survival outcome, probabilities of 2- and 5-year overall, recurrence free, and bladder cancer-specific survivals were estimated using standard Kaplan–Meier methods. For pathological outcome, probabilities of pathological T3–T4 and N1–N3 stages were estimated using sample proportions. All the probabilities were estimated in each predicted probability group (0–0.2, 0.21–0.4, 0.41–0.6, 0.61–0.8 and 0.81–1.0) for each nomogram. For the purpose of external validation, the concordance between the predicted probabilities (nomogram) and the actual results in the Roswell Park Cancer Institute population were graphically assessed using calibration plots and formally analysed using the c-index with corresponding 95% CI. A c-index of 0.5 indicates a poor model, while values closer to 1 (with 1 being perfect) indicate better predictive models. All tests were two-sided and a P < 0.05 was considered to indicate statistical significance.
The median (range) patient age was 70 (36–90) years. Most patients (77%) had muscle-invasive disease. Lymph node metastasis was identified in 24% of patients (Table 1).
|Median (range) age, years||70 (36–90)|
|Lymphovascular invasion||108 (39)|
|Carcinoma in situ||52 (18)|
|Clinical T stage:|
|High grade||272 (97)|
|Pathological T stage:|
|Pathological N stage:|
|Neoadjuvant chemotherapy||36 (13)|
|Preoperative radiation||5 (2)|
|Adjuvant chemotherapy||59 (21)|
|Postoperative radiation||18 (7)|
|Bladder cancer-specific mortality||47 (17)|
|Median (range)||15 (0–85)|
At a mean follow-up of 20 months, local or distant disease recurrence developed in 30% of patients, 33% overall mortality and 17% died from bladder cancer. The actuarial 2- and 5-year PFS after RARC was 62% (95% CI 54–68) and 55% (95% CI 46–63), respectively. The actuarial 2- and 5-year ACS was 66% (95% CI 59–72) and 47% (95% CI 37–55), respectively. The 2- and 5-year CSS was 81% (95% CI 74–86) and 67% (95% CI 57–76), respectively (Fig. 1).
The ability of three different nomograms, BCRC, IBCNC and Lughezzani to predict survival outcomes of our patients is summarised in Table 2. For PFS, c-indices were determined for the IBCNC (5-year) and BCRC (2- and 5-year) nomograms. The PFS c-index for IBCNC was 0.70 at 5 years, and for BCRC was 0.77 at both the 2 and 5 years. The accuracy of ACS and CSS prediction was evaluated using the BCRC and Lughezzani nomograms. Using the BCRC nomogram, c-indices for 2- and 5-year ACS were each 0.73, and c-indices for 2- and 5-year CSS were 0.70 each. The performance of Lughezzani nomogram for 5-year ACS, cancer-specific mortality (CSM) and OCM were 0.73, 0.72 and 0.40, respectively.
|Predicted outcomes||Nomograms performance (95% CI)|
|2-year RFS||–||0.77 (0.72–0.83)||–|
|5-year RFS||0.70 (0.6–0.71)||0.77 (0.72–0.83)||–|
|2-year ACS||–||0.73 (0.68–0.78)||–|
|5-year ACS||–||0.73 (0.67–0.78)||0.73 (0.68–0.80)|
|2-year CSS||–||0.70 (0.65–0.76)||–|
|5-year CSS||–||0.70 (0.65–0.76)||0.72 (0.63–0.80)|
|5-year OCM||–||–||0.40 (0.30–0.53)|
|Pathological stage||–||0.66 (0.60–0.73)||–|
|Lymph node metastasis||–||0.61 (0.44–0.78)||–|
In addition to survival outcomes, we also evaluated the ability of the BCRC nomogram to predict advanced pathological stage and lymph node metastasis. The accuracy of predicting these two variables was modest, with lower c-indices of 0.66 and 0.61, respectively (Table 2).
Figure 2 graphically shows 2- and 5-year PFS, risk of advanced tumour stage and lymph node metastasis derived from each nomogram relative to actual outcomes. It shows that all nomograms tend to predict 2- and 5-year PFS with good accuracy. The prediction for lymph node metastasis was hampered by the narrow range of predicted probabilities and clearly represents an overestimation of that risk. Calibration plots for ACS, CSS and OCM (Fig. 3) show that, ACS predictions at 2 and 5 years were achieved with good concordance especially at 2 years by the BCRC nomogram. Meanwhile, CSS prediction tends to be overestimated for patients at lower risk of death from bladder cancer by BCRC nomogram but with pronounced improvement in performance for patients with a high risk of death from bladder cancer. The Lughezzani nomogram performance for OCM was indiscriminative, with a c-index of 0.4.
Several bladder cancer nomograms have been proposed for risk stratification to formulate individualised treatment plans [10-13], However, an ideal nomogram has not yet been developed and currently available instruments have limitations. In general, nomogram performance tends to be lower when external validation is performed. Additionally, extreme variables could impact a nomogram negatively. Thus, external validation before clinical use becomes essential to address shortcomings of any such predictive instrument.
Several nomograms are available for clinical use in bladder cancer to predict oncological outcomes after RC (e.g. ACS, CSS, and PFS). We evaluated the accuracy of three published nomograms for predicting oncological outcomes using data from a single high-volume RARC surgeon.
The published nomograms evaluated in the present study were constructed based on high-volume open RC series with variable follow-up intervals. IBCNC developed a postoperative nomogram to predict 5-year PFS using a dataset of 900 patients from 12 worldwide centres with a median follow-up of 30 months. The nomogram performance was tested at each participating institution after removing their contributing data, with c-indices ranging from 0.65 to 0.78 . The data from 731 patients who underwent RC and PLND for TCC of the bladder between 1984 and 2003 in three different high-volume USA institutions was used to construct BCRC nomogram. This tool predicted 2-, 5- and 8-year ACS, CSS, PFS, advanced tumour stage and lymph node metastasis with median follow-up of 25 months. Internal validation revealed predictive accuracy of 0.73, 0.79, 0.78, 0.61 and 0.63 for ACS, CSS, PFS, advanced tumor stage and risk of pelvic lymph node metastasis, respectively [10-12]. In 2010, Lughezzani et al.  reported a model to predict ACS, CSM and OCM based on SEER (Surveillance, Epidemiology and End Result) registries, which included 11 260 patients who underwent RC and PLND between 1988 and 2006 with median follow-up of 38 months.
Few prior studies have attempted to externally validate these nomograms. Using data from 2500 patients treated by RC-PLND at eight European centres between 1989 and 2008, Nuhn et al.  found the BCRC nomogram prediction accuracy of ACS and CSS at 2, 5, and 8 years after RC was 71.0, 69.1, and 68.2, and 74.9, 73.1, and 72.4, respectively. PFS prediction accuracy at the same time period was 76.5, 75.3, and 74.9, respectively. Similarly, Zaak et al.  externally validated the BCRC and IBCNC nomograms based on 246 patients treated with RC-PLND at two German centres between 1992 and 2007. The predictive accuracies of BCRC nomograms for PFS, ACS and CSS were 0.84, 0.78 and 0.85, respectively. However the predictive accuracy of IBCRC nomogram was 0.86. This high rate of concordance across different populations indicates the stability and robustness of the nomograms. The BCRC nomogram for prediction of locally advanced tumour stage and lymph node metastasis was tested by May et al.  who found the predictive accuracies of 67.5 and 54.5, respectively. To our knowledge, IBCNC and BCRC nomograms have not been externally validated in a USA population and the Lughezzani nomogram has never been validated.
In the present study, nomograms predicting ACS and CSS achieved adequate accuracy with c-indices ranging from 0.7 to 0.73. Additionally, good concordance was present for PFS prediction using both IBCNC and BCRC nomograms, with c-indices 0.70 to 0.77, respectively. However, the Lughezzani nomogram prediction for OCM failed to show any discrimination abilities.
To prove useful for clinical use, the predictive tool should be generalizable . Notably, multi-institutional, international and high-volume centres datasets of the original cohorts used to generate the nomograms and most external validation cohorts create heterogeneity that could impact the nomogram performance negatively by lowering the predictive accuracy. However, heterogeneity makes the nomogram performance stable across different populations. Thus, external validations of IBCNC and BCRC nomograms performed adequately through eight European centres. In the present study, PFS, ACS and CSS prediction of the nomograms was adequate and consistent with the above external and internal validation studies. The present study represents a single surgeon and high-volume centre experience of RARC with a recent cohort (2005–2012). Thus, homogeneity of population, follow-up, and surgical technique across the cohort make the present population different from the nomogram cohorts, nevertheless, the performance of the nomograms was adequate.
RARC has been emerging as an equivalent approach to open RC with potentially equivalent operative and oncological outcomes [19-21]. However, little data on long-term oncological outcomes after RARC makes physician-patient decision-making, using open RC data, anecdotal. Therefore, clinical use of a validated tool provides the patient with therequired data to make an informed decision about proceeding with surgery or not. A nomogram, validated by using the RARC population, will help patients make informed choices about the use of a robot-assisted surgical technique for the management of bladder cancer, based on the risks and benefits as highlighted by such tools. Although, RC-PLND has been established for muscle-invasive bladder cancer, the extent of lymph node dissection has been changing with wide variation amongst surgeons. This affects survival outcomes, which could in turn impact nomogram performance . We found that the BCRC nomogram performance for predicting advanced tumour stage and lymph node metastasis was modest and consistent with the internal and external validations reported by May et al. . Originally, the pre-RC nomogram used common variables (age, clinical stage, grade, and carcinoma in situ). Although these are strong predictors, others could contribute to prediction such as lymphovascular invasion and repeat transurethral resection of tumour effect on changing the final stage assignment. In addition, the high discordance between the clinical and pathological staging makes the predictive power of any tool difficult [3-5].
Although the performance of the Lughezzani nomogram prediction for ACS and CSS was adequate, it failed to discriminate for OCM. Variation in reporting the cause of death, especially in bladder cancer where the population is usually significantly morbid and can succumb to many competing causes of death, could negatively impact the nomogram. In author's practice the IBCNC and BCRC nomograms have been used for clinical use. The use is typically employed to counsel patients with advanced age and disease or multiple co-morbidities, so that they can reach an informed decision about the surgery.
The present study is primarily limited by its relatively short follow-up at median (range) of 15 (0–85) months, whereas the IBCNC and BCRC nomograms data were constructed with median follow-up intervals of 30 and 25 months, respectively. It is possible that accuracy of these nomograms may be different, better or worse, with additional follow-up in the present cohort. It should also be noted that the present findings apply to patients treated by an experienced minimally invasive surgeon, at a high-volume tertiary care centre. This may limit the generalizability of these findings to low-volume community based practices. The future implication of the present study would be to develop a predictive tool, based upon the long-term oncological outcomes of the larger patient cohort of RARC, such as from the International Robotic Cystectomy Consortium. Such a predictive model would not only cover the outcomes from institutions across the globe, but also provide information pertinent to this emerging technology for the surgical management of patients with bladder cancer.
In conclusion, currently available bladder cancer nomograms adequately predict ACS, CSS and PFS. However, prediction of advanced bladder tumour stage and lymph node metastasis was modest and the Lughezzani nomogram failed to predict OCM.
K.A. Guru is a Board Member for Simulated Surgical Systems, outside the submitted work. No other conflicts declared.
Bladder Cancer Research Consortium
International Bladder Cancer Nomogram Consortium
(robot-assisted) radical cystectomy