Correspondence: Siamak Daneshmand, University of Southern California Institute of Urology, Keck School of Medicine and Norris Comprehensive Cancer Center, 1441 Eastlake Avenue, NOR 7416, Los Angeles, CA 90033, USA.
What's known on the subject? and What does the study add?
Neoadjuvant chemotherapy is advocated for most patients with carcinoma invading bladder muscle. An improved risk-stratification of clinical stage T2 (cT2) patients can potentially identify candidates who may derive maximal benefit from this approach. cT2 patients who are pathologically upstaged at cystectomy have significantly worse prognosis than their counterparts who are not upstaged. The identification of such candidates who may be subsequently upstaged represents a strategy for selecting those patients who may benefit the most from neoadjuvant chemotherapy, whereas other patients can undergo early cystectomy.
The present study describes a unique cross-validated decision tree generated using precystectomy variables aiming to stratify patients with cT2 tumours based on the risk of pathological upstaging and adverse oncological outcomes. This model can be potentially employed as a tool for making clinical decisions with respect to neoadjuvant chemotherapy in these patients.
To categorize patients with clinical stage T2 bladder cancer into risk groups based on their potential for pathological upstaging and eventual oncological outcomes at cystectomy.
To pre-emptively identify such patients who will be upstaged and have poor outcomes after cystectomy, aiming to better determine the ideal candidates for neoadjuvant chemotherapy.
Patients and Methods
A retrospective review was conducted of 1964 patients who underwent radical cystectomy for bladder cancer with intent to cure at the University of Southern California between 1971 and 2008.
Neoadjuvant chemotherapy-naïve patients with clinically organ-confined urothelial carcinoma invading bladder muscle (cT2N0M0) were included.
Univariate analysis and multivariable decision tree modelling with cross-validation were employed to identify precystectomy variables that could predict pathological upstaging and poor oncological outcomes.
A total of 948 patients met the inclusion criteria, of whom 512 (54%) patients were upstaged at cystectomy; upstaging was associated with a worse recurrence-free and overall survival (both P < 0.001).
Age, presence of hydronephrosis, evidence of deep muscularis propria invasion and lymphovascular invasion on transurethral resection specimen, as well as tumour growth pattern and count, were significantly associated with upstaging.
When these factors were included in a decision tree model, 70.6% of patients with hydronephrosis experienced upstaging and had the worst outcome (P < 0.001).
In patients without hydronephrosis, tumour growth pattern was a second-tier discriminator (P < 0.001); in patients with non-papillary tumours, 71.7% of cases with evidence of deep muscularis propria involvement experienced upstaging compared to 53.8% of cases with no deep muscle involvement (P = 0.012), whereas, among patients with combined papillary and non-papillary features, 33% of cases aged ≤65 years were upstaged compared to 47% of cases aged >65 years (P = 0.036).
The cross-validated decision tree resulted in three risk groups with significantly varying probabilities of recurrence-free and overall survival (both with overall P < 0.001).
Hydronephrosis, tumour growth pattern, deep muscle involvement and age can collectively identify patients with cT2N0M0 bladder cancer who have varying risks of pathological upstaging.
Such categorization using a visually intuitive model can facilitate clinical decision-making with respect to neoadjuvant therapy in these patients.
Pathological upstaging at cystectomy is associated with poor outcomes in urothelial cell carcinoma of the bladder (UCB) [1, 2]. Neoadjuvant chemotherapy administration has been associated with pathological downstaging, decreased odds of positive surgical margins, and early treatment of any micrometastases in UCB [3, 4]. Phase III studies have shown its beneficial role [5, 6] but its adoption across centres has been slow. Although neoadjuvant chemotherapy is intended for patients with carcinoma invading bladder muscle, its benefit in ‘average risk’ patients with clinical stage T2 (cT2) tumours is modest at best . However, patients with high-risk disease, such as those with cT3b cancers, may derive a more substantial benefit.
Patients with cT2 UCB who are subsequently upstaged at cystectomy can potentially benefit the most from neoadjuvant chemotherapy. However, the inability to identify such patients at the outset dilutes its potential advantage, as documented by a modest 5% absolute 5-year overall survival benefit . The present study was designed to identify patients with cT2 UCB who are at risk for pathological upstaging. Such patients would potentially derive maximal benefit from neoadjuvant chemotherapy, whereas other patients with modest or no further benefit may elect for early cystectomy. Although routine prognostic risk stratification for pathological T2 disease after cystectomy has been described , the present study is unique in terms of presenting a visually interpretive decision model for identifying patients with cT2 UCB who have varying risks of upstaging at cystectomy.
Patients and Methods
In total, 1964 patients who underwent radical cystectomy for UCB with intent to cure at the University of Southern California between 1971 and 2008 were identified through an Institutional Review Board-approved database. Patients with clinically organ-confined UCB invading bladder muscle (cT2N0M0) who did not receive neoadjuvant chemotherapy were included in this analysis. Muscle invasion was confirmed by pathological review of transurethral resection of bladder tumour (TURBT) specimens; clinical absence of extravesical disease was confirmed by imaging and no palpable tumour on physical examination. Exclusion criteria were: (i) presence of non-UCB histology in ≥80% of the primary tumour; (ii) presence of urethral or upper tract primaries at diagnosis; (iii) nodal and/or distant metastasis at initial diagnosis; (iv) receipt of neoadjuvant chemotherapy; and (v) patients undergoing salvage surgical procedure after failure of definitive radiotherapy (≥50 Gy). A total of 948 patients met the above criteria and were included in the analysis.
Patient Management and Precystectomy Variables
Patients underwent a routine precystectomy assessment, including physical examination, cystoscopic and upper tract evaluation, TURBT, and visual assessment of uninvolved bladder and prostatic urethra . Based on body mass index, 480 (50.6%) patients were categorized as being underweight/normal weight (<25 kg/m2), overweight (25 to <30 kg/m2) and obese (≥30 kg/m2). The type of tumour growth (papillary and/or non-papillary) and the number of tumours visible on cystoscopy were documented in 732 (77.2%) and 694 (73.2%) patients, respectively. Lymphovascular invasion in TURBT specimens was noted in 663 (69.9%) patients, as described previously . Evidence of tumour invasion into deep muscularis propria was assessed in original and repeat TURBT specimens where possible. Routine radiographical evaluations, including that of the upper urinary tract, were performed. Evaluation for the presence of hydronephrosis, if any, was performed in 794 (83.8%) patients, as described previously .
All patients underwent radical cystectomy, extended pelvic lymphadenectomy and urinary diversion . Tumour staging and grading were standardized to 2002 American Joint Committee on Cancer and 1973 WHO recommendations, respectively [14, 15]. Initial clinical and final pathological stages were compared to determine upstaging; a patient was considered pathologically upstaged at cystectomy if the tumour was staged ≥pT3 and/or nodal metastasis was detected. Postoperative follow-up was carried out at 4-month intervals in year 1, 6-month intervals in year 2, and annually thereafter, with laboratory and imaging studies unless otherwise clinically indicated.
Primary outcomes of interest were recurrence-free survival (RFS) and overall survival (OS). RFS duration was calculated from the date of cystectomy to the date of first clinical recurrence; patients who were recurrence-free at end of the present study were censored at death or last follow-up. OS duration was calculated from the date of cystectomy to death from any cause; surviving patients were censored at last follow-up.
Analyses were performed using SPSS, version 17.0 (SPSS Inc., Chicago, IL, USA). P ≤ 0.05 (two-sided) was considered statistically significant. Pearson's chi-squared test was used to examine associations between categorical variables. The log-rank test and Kaplan–Meier plots were used to compare univariate differences in outcomes.
For multivariable modelling, an exhaustive chi-squared automatic interaction detection-based decision tree technique was employed to assess the hierarchical interactions of variables that were univariately associated with pathological upstaging and, in addition, could predict RFS and/or OS . The tree-growing algorithm examined all possible splits for each variable that was independently predictive of pathological upstaging, aiming to identify the strongest association at each step. Tree-growing was restricted to at least 50 patients per terminal node to limit complexity. Tree-pruning was then performed to produce simpler subtrees by assessing the misclassification error associated with a particular subtree. Categories were split based on Pearson's chi-squared at P < 0.05; significance values were adjusted using the Bonferroni method. The tree model was cross-validated over 10 sample folds to ensure robustness and reproducibility. The accuracy of the final model in predicting pathological upstaging, as determined by area under a receiver operating characteristic curve, was determined using logistic regression based on variables included in the decision tree.
Table 1 shows the overall patient characteristics. The median (range) age was 67.2 (23.4–93.0) years. In total, 898 (94.7%) patients had high-grade carcinomas. A total of 161 (17%) patients underwent radical cystectomy after failing one or more courses of intravesical bacille Calmette-Guérin, thiotepa or mitomycin C instillation. A median (range) of 36 (3–177) lymph nodes was dissected during cystectomy. Urinary diversion via ileal conduit, continent cutaneous and orthotopic neobladder was performed in 198 (20.9%), 181 (19.1%) and 569 (60%) patients, respectively. Adjuvant chemotherapy was administered to 249 (26.3%) patients after radical cystectomy. The median follow-up was 14.2 years (range, 1.1–35.6 years), during which 322 (34%) patients experienced recurrence and 624 (65.8%) patients died.
Table 1. Patient demographics and clinicopathological characteristics, and their association with pathological upstaging at cystectomy.
To assess any potential variability in patient characteristics, treatment and outcome metrics across the 38-year cohort, patients were categorized into two eras: cystectomy performed in or before 1997 (n = 558; 59%) and in or after 1998 (n = 388; 41%). There were no differences with respect to sex, the presence of hydronephrosis (both P = 0.14) and tumour count (P = 0.48) in patients across both eras. However, the latter era had greater proportion of older (54.3% vs 66.5%; P < 0.001) and more obese (15% vs 24.6%; P = 0.023) patients. After 1997, more patients received orthotopic neobladders (43.2% vs 84.3%; P < 0.001). There was no change in the frequency of the receipt of adjuvant chemotherapy between both eras (P = 0.90). Despite the few changes in baseline characteristics, there were no differences in RFS (log-rank P = 0.56) and OS (log-rank P = 0.88) between the two eras. This suggested that the aggregate findings within this cohort could be considered as representative of the prognostic course of cT2N0M0 disease independent of the era when a patient underwent cystectomy.
Univariate Associations with Pathological Upstaging
In this cohort, 512 (54%) patients were pathologically upstaged at cystectomy. Of these patients, 244, 77 and 191 patients, respectively, experienced tumour, nodal, and both tumour and nodal upstaging. Pathological upstaging was associated with a worse RFS and OS (both log-ranks P < 0.001; Fig. 1). Pathological upstaging was therefore considered as an indicator of poor prognosis in cT2N0M0 cases. By univariate analysis, increasing age (P = 0.010), the presence of hydronephrosis (P < 0.001) and evidence of deep muscularis propria (P < 0.001) and lymphovascular invasion (P = 0.010) on TURBT specimen were independently associated with pathological upstaging (Table 1). Pathological upstaging was also associated with non-papillary lesions (P < 0.001) and also the number of tumours observed on cystoscopy (P = 0.004).
Univariate Associations with Clinical Outcome
The associations of patient characteristics with RFS and OS are shown in Table 2. Women patients had a lower RFS probability in this cohort (P = 0.021). As expected from their significant associations with pathological upstaging, the presence of hydronephrosis and evidence of deep muscularis propria invasion were associated with a poor RFS (both P < 0.001). The presence of lymphovascular invasion on TURBT specimen also showed a trend towards a lower RFS (P = 0.052). Older patients had a lower OS probability (P < 0.001). Similar to their trends with pathological upstaging and RFS, evidence of deep muscularis propria invasion and the presence ofhydronephrosis were associated with a poor OS (both P < 0.001).
Table 2. Univariate associations of demographic and clinicopathological characteristics with clinical outcome.
Variables that were univariately associated with pathological upstaging and, in addition, may predict outcome, were used to construct a multivariable decision tree. The aim of this exercise was to construct a clinically relevant algorithm aiming to identify precystectomy determinants of subsequent pathological upstaging in the order of their relative statistical significance. The variables input into this model included age, presence of hydronephrosis, evidence of deep muscularis propria invasion, lymphovascular invasion on TURBT, type of growth and tumour count.
The final decision tree had three tiers of branching using four categorical variables that resulted in six terminal nodes (Fig. 2). The presence of hydronephrosis was the most important discriminator (P < 0.001); 70.6% of patients with hydronephrosis experienced upstaging and also had the worst RFS and OS probabilities. In patients without hydronephrosis, the tumour growth pattern was a second-tier discriminator (P < 0.001). Among patients with papillary tumours, only 36.6% experienced upstaging. In patients with non-papillary tumours, evidence of deep muscularis propria invasion was a third-tier discriminator (P = 0.012); 71.7% of patients with deep muscle invasion were upstaged compared to 53.8% of patients where deep muscle invasion was not identified. For patients with combined papillary and non-papillary features, age was a third-tier discriminator (P = 0.036); 33% of patients aged ≤65 years were upstaged compared to 47% of patients aged >65 years.
After categorization of patients into specific nodes based on their potential for pathological upstaging, oncological outcomes of individual nodes were compared aiming to determine risk groups for prognosis. Consequently, nodes were combined into groups based on their risks for RFS and OS. Nodes 1–3 comprised the low–moderate risk group, where 47.5% of patients were upstaged with 5-year RFS and OS probabilities of 66% and 48%, respectively (Table 3). Nodes 4 and 5 comprised the moderate–high risk group, where 54.8% of patients experienced upstaging with 5-year RFS and OS probabilities of 61% and 34%, respectively. Patients in node 6 comprised the high-risk group, where 5-year RFS and OS probabilities of 46% and 27%, respectively, were the lowest, and 70.6% of patients were upstaged. Patients in these risk groups therefore had significantly different RFS and OS (both with overall P < 0.001; Fig. 3 and Table 3). The mean (se) accuracy of the final model in predicting pathological upstaging was 65.7% (2.1%).
Table 3. Associations of risk groups with clinical outcome.
aCalculated by the log-rank test. L-M, low–moderate; M-H, moderate–high; H, high.
L-M risk group
M-H risk group
L-M risk group
M-H risk group
Reproducibility of the decision tree was confirmed by cross-validation over 10 sample folds. To ensure that the identified metrics were prognostically consistent over the cohort accrual period, patients were sub-stratified into the two aforementioned eras before being assigned to their respective terminal nodes. Even with such sub-stratification, the decision tree was able to predict RFS and OS in patients who underwent cystectomy in or before 1997 (P = 0.007 and P < 0.001, respectively) and in or after 1998 (P = 0.004 and P = 0.001, respectively; data not shown). These results can thus be generalized across the entire cohort, and the tree could be used as a robust clinical decision tool.
Patients with cT2 UCB represent a heterogeneous population where neoadjuvant chemotherapy has modest benefits . They also represent the largest sub-population of patients with UCB invading bladder muscle . In our cohort, 54% of patients with cT2 UCB experienced pathological upstaging with poor outcomes, which is an association that has also been documented in other studies [1, 2]. Given the potential ability of neoadjuvant chemotherapy to effect pathological downstaging, patients with clinically organ-confined UCB who are destined to be upstaged at cystectomy may be the best candidates for such therapy. However, the ability to pre-emptively predict which patients with cT2 UCB will subsequently be upstaged is currently limited. The present study therefore aimed to identify the precystectomy factors predictive of pathological upstaging and oncological outcomes, as well as their relative importance in systemic chemotherapy-naïve cT2 UCB patients. This resulted in a multivariable clinical decision tool for cT2NoMo UCB patients that identified candidate subgroups with varying risks of upstaging, recurrence and death.
Prediction models have been suggested previously in UCB [19-22]. Our multivariable decision tree model has the advantage of enabling a surgeon to readily identify those patients with cT2 UCB who have the greatest probability of being upstaged at cystectomy and have poor outcomes without employing complex nomograms. Such a decision-analytic approach may be possible by including hydronephrosis, type of tumour growth, evidence of deep muscularis propria invasion and patient age as defining variables. Cross-validation of the decision tree over 10 sample folds ensured the robustness and reproducibility of the final model.
Hydronephrosis was the first-tier discriminator in predicting upstaging. Hydronephrosis has been associated with aggressive UCB and poor outcomes [12, 23]. Tumour growth pattern was a second-tier discriminator; UCB appearance has been associated with their biological behaviour . Age and deep muscularis propria invasion were third-tier discriminators. Advanced age can adversely affect UCB outcome . Although assessing the depth of tumour invasion into muscularis propria may be pathologically challenging, this often depends on the quality of the TURBT specimen. The American Joint Committee on Cancer recommendations incorporate T2 substaging in their guidelines , although its prognostic value in pathological stage T2 patients is controversial [26-28]. However, a more recent report indicates that such sub-stratification may indeed be prognostic in patients with pathological stage T2 UCB . Furthermore, the analysis in the present study shows that this assessment in TURBT specimens may be particularly beneficial for clinical stage T2 patients with non-papillary tumours without hydronephrosis. The significance of these factors in predicting upstaging indicates the importance of imaging, cystoscopy and the pathological assessment of TURBT specimens, particularly with respect to the depth of muscularis propria invasion, which may not always be documented at all centres. It should also be noted that, although these clinicopathological parameters can improve the accuracy of predicting pathological upstaging beyond random chance, there is room for improvement using other novel variables and biomarkers that were not included in the analysis performed in the present study.
The present study has some limitations. Despite being one of the largest cT2 UCB cohorts treated subsequently by radical cystectomy to be analyzed so far with respect to predicting pathological upstaging, the present study nevertheless comprises a retrospective, single-institution investigation that requires further validation. However, consistencies in surgical and patient management within this cohort allowed the generation of a thoroughly cross-validated multivariable decision model. Tumour volume was not a part of this analysis, although other precystectomy surrogates, including the extent of muscularis propria invasion and the number of tumours, were analyzed. However, it is possible that patients with a lower tumour burden may have a lower probability of upstaging and, consequently, a higher likelihood of responding to the receipt of neoadjuvant chemotherapy than those with a relatively higher tumour load, although they may share the same risk group. We also recognize that, although patients with hydronephrosis have the highest risk of upstaging, severe hydronephrosis and accompanying renal insufficiency may render it difficult to deliver cisplatin-based chemotherapy to these candidates . However, patients with obstruction can be managed with percutaneous nephrostomy or internal stents if they are otherwise good candidates for chemotherapy. These patients may also be optimal candidates for targeted agent-based therapies in the future.
In conclusion, the identification of patients with cT2 UCB who have disparate risks of pathological upstaging and clinical outcome is an important clinical question because it maximizes the advantages of the receipt of neoadjuvant chemotherapy in candidates who need it the most, at the same time as allowing early cystectomy for those patients who may not otherwise derive much gain. The decision tree described in the present study is visually intuitive and can help with clinical decision-making. Although the most effective therapeutic regimens and targets remain to be defined and may vary for different patients, the present study identifies subpopulations of patients with cT2 UCB who have varying risks of upstaging and outcome.