Patients with Muscle-Invasive Bladder Cancer with Nonluminal Subtype Derive Greatest Benefit from Platinum Based Neoadjuvant Chemotherapy

Abstract

Purpose:

Neoadjuvant chemotherapy (NAC) prior to radical cystectomy (RC) in patients with nonmetastatic muscle-invasive bladder cancer (MIBC) confers an absolute survival benefit of 5%–10%. There is evidence that molecular differences between tumors may impact response to therapy, highlighting a need for clinically validated biomarkers to predict response to NAC.

Materials and Methods:

Four bladder cancer cohorts were included. Inverse probability weighting was used to make baseline characteristics (age, sex and clinical tumor stage) between NAC-treated and untreated groups more comparable. Molecular subtypes were determined using a commercial genomic subtyping classifier. Survival rates were estimated using weighted Kaplan-Meier curves. Cox proportional hazards models were used to evaluate the primary and secondary study end points of overall survival (OS) and cancer-specific survival, respectively.

Results:

A total of 601 patients with MIBC were included, of whom 247 had been treated with NAC and RC, and 354 underwent RC without NAC. With NAC, the overall net benefit to OS and cancer-specific survival at 3 years was 7% and 5%, respectively. After controlling for clinicopathological variables, nonluminal tumors had greatest benefit from NAC, with 10% greater OS at 3 years (71% vs 61%), while luminal tumors had minimal benefit (63% vs 65%) for NAC vs non-NAC.

Conclusions:

In patients with MIBC, a commercially available molecular subtyping assay revealed nonluminal tumors received the greatest benefit from NAC, while patients with luminal tumors experienced a minimal survival benefit. A genomic classifier may help identify patients with MIBC who would benefit most from NAC.

Abbreviations and Acronyms

CSS

cancer-specific survival

GSC

genomic subtyping classifier

IPW

inverse probability treatment weighting

MIBC

muscle-invasive bladder cancer

MVA

multivariable analysis

NAC

neoadjuvant chemotherapy

OC

organ-confined

OS

overall survival

RC

radical cystectomy

UVA

univariable analysis

Neoadjuvant cisplatin-based chemotherapy (NAC) prior to radical cystectomy (RC) in patients with nonmetastatic muscle-invasive bladder cancer (MIBC) confers an absolute overall survival (OS) benefit of 5%–10%.^1–3 Despite level 1 evidence and guideline panel recommendations, the uptake of NAC in clinical practice is inconsistent, with utilization rates of approximately 20%–30% in the U.S.^4,5 There are several factors that impact decision making regarding the use of NAC, including patient comorbidities, potential chemotherapy-related toxicity, the risk of progression in patients who do not respond, the delay in time to radical local treatment and the option to administer adjuvant chemotherapy. A major element of this decision involves balancing the modest survival benefit with the potential risks, especially in patients with organ-confined (OC) disease (cT2N0).

While the majority of patients with MIBC are diagnosed with clinical (c)T2N0 disease, up to 50% of these will be upstaged to pathological (p)T3/4 or node-positive disease at RC.⁶ As a consequence, NAC is recommended for all patients with cT2-T4aN0M0 bladder cancer in a nonselective fashion, even though the risk of harboring nonorgan-confined disease and the likelihood of benefit from therapy varies widely. If one knew who had OC disease or would be unlikely to respond to NAC, then avoiding NAC and proceeding with RC would likely be the preferred option. Patients with pT1-T2N0 disease will be cured with RC alone in 80%–90% of cases.^6,7 Furthermore, patients who receive NAC and do not experience a pathological response have actually been found to have worse outcomes at RC than those who never received NAC.⁸ Conversely, patients with nonorgan-confined disease are likely to derive the greatest relative benefit from NAC.³ Similarly, there is evidence that molecular differences between tumors may impact response to therapy. Hence, there is a need for clinically validated biomarkers at the moment of diagnosis to predict response to NAC. Such biomarkers would allow likely nonresponders to avoid the toxicity of NAC and expedite local therapy, while simultaneously enriching for patients likely to benefit from NAC.

Molecular subtyping has demonstrated that subtypes of MIBC have distinct clinical and biological characteristics that correlate with differential responses to NAC and survival.^9–15 For example, a recent study using a genomic classifier based on molecular subtyping found that basal tumors were associated with the greatest improvements in survival with NAC, while luminal tumors appeared to have similar survival with and without NAC.¹¹ Furthermore, luminal cT1-2 tumors are more likely to be OC at the time of cystectomy than nonluminal tumors, which are more likely to be nonorgan-confined.⁶ These studies, however, did not adequately compare patients of similar stage and subtype who received NAC prior to RC vs those who underwent RC alone.

Herein, we compiled a multi-institutional cohort of over 800 bladder cancer patients to evaluate the impact of molecular subtyping on survival, specifically comparing patients who received NAC followed by RC to those who had RC alone. To better account for differences in the clinical characteristics of the NAC and non-NAC treated patients, we used inverse probability treatment weighting (IPW) to make covariates comparable between treatment groups. Our goal was to determine if molecular subtyping might improve selection of MIBC patients most likely to benefit from NAC compared to unselected utilization of NAC in all patients with MIBC.

Materials and Methods

Study Patient Populations

We combined 4 cohorts of patients with bladder cancer who underwent RC and had genomic data available from transurethral resection of bladder tumor samples: 1) NAC I—neoadjuvant chemotherapy cohort I (82), which consisted of patients with clinical T2-4aN0-3M0 bladder cancer treated with NAC followed by RC;¹¹ only the validation cohort was used in this study, as the model being tested here was trained on the discovery cohort; 2) NAC II—neoadjuvant chemotherapy cohort II (244), which consisted of patients with clinical T2-4N0-3M0 or TanyN1-3M0 bladder cancer treated with NAC followed by RC;¹⁶ 3) CirGuidance study—bladder cancer cohort (292), which consisted of patients with clinical T2-4aN0-1M0 MIBC treated with NAC and RC or RC alone;¹⁷ and 4) MOL—molecular upstaging cohort (210), which consisted of patients with clinical T1-2N0 bladder cancer who underwent RC without NAC.⁶ Only tumors with muscle-invasive disease were included (see below).

The final analytical cohort was generated a priori based on the following. All tumor samples had pure urothelial carcinoma histology, and we excluded tumors with secondary variant histology (nested, micropapillary, sarcomatoid, plasmacytoid and small cell) and those classified as neuroendocrine-like based on gene expression¹⁸ (see below). There was no centralized pathology review for this study. Patients were excluded if they received noncisplatin-based chemotherapy, did not undergo RC or were missing followup data. As this is a retrospective analysis and not a study of intent-to-treat, patients with cT1 tumors were included if they ultimately were found to have MIBC (pT2-4N0-3) at RC in the non-NAC group. There were no cT1 cases in the NAC group.

Specimen Collection and Processing

Specimen collection and sample processing were conducted as described previously.^11,19 Decipher®, a clinical-grade whole-transcriptome assay, was used to generate genomic subtyping classifier (GSC) scores for all specimens based on 2 validated signatures.¹⁹

Molecular Subtyping

The GSC subtypes were assigned by first identifying neuroendocrine-like cases,¹⁸ which were excluded, and then classifying the remaining tumors using the Seiler 2017 model.¹¹ The intent of this study was the clinical validation of a commercially available biomarker, and therefore other molecular subtyping models were not evaluated.

Statistical Analysis

The primary and secondary study end points were OS and cancer-specific survival (CSS), respectively. Missing CSS status (77 patients) was assumed to be random and was imputed from respective OS status (7 died, 70 alive). Descriptive statistics were reported by medians and interquartile range or frequencies and proportions, where appropriate.

IPW was used to make baseline clinical characteristics between NAC-treated and untreated groups more comparable.²⁰ The weights were the inverse of treatment propensity scores constructed using age group (divided at the cohort’s median age: ≤68.1 vs >68.1), sex and clinical stage (cT) group (cT1-2 vs cT3-4). Molecular subtyping was not considered for weighting. Survival rates were estimated using weighted Kaplan-Meier curves. The variables of continuous age, sex and clinical T stage group were included in weighted univariable analysis (UVA) and multivariable analysis (MVA) Cox proportional hazards models for study end points.

In post hoc secondary analyses, Cox proportional hazards models were constructed for different subcohorts with more stringent inclusion criteria, such as excluding patients with cT1 with and without positive nodes or patients who received adjuvant chemotherapy. In prespecified sensitivity analyses, more conservative missing CSS imputations were run for robustness, and cumulative incidence curves were constructed to estimate risks of CSS to account for death from other causes as a competing risk. The analyses were repeated using unweighted models and other alternative approaches (pseudo-replication, 1-to-1 matching and IPW with the addition of GSC) to account for baseline differences between the NAC and non-NAC groups.

Interaction term analysis was not included as part of the study design and was limited by the sample size but was performed to study the interaction between molecular subtypes and receipt of NAC. All statistical tests were 2-sided and the reported significance level was 0.05. Analyses were performed in R v3.4.1 (R Foundation for Statistical Computing, Vienna, Austria).

Results

Cohort Patient Demographics and Impact of Neoadjuvant Chemotherapy

A cohort of 828 patients with bladder cancer who underwent RC with or without NAC was assembled, which after filtering for inclusion criteria resulted in a total of 601 eligible patients (fig. 1 and table 1). The NAC group was comprised of 247 patients with a median age of 65.0 years and the non-NAC group consisted of 354 patients with a median age of 70.2 years (p <0.001). Tumors were stage cT2 in 76% of NAC and 66% of non-NAC cases (p <0.001), with 31 cT1 (pT2-4N0-3) tumors included in the non-NAC group (supplementary table 1, https://www.jurology.com). The median followup for the NAC and non-NAC groups was 2.5 and 2.9 years, respectively (table 1).

Figure 1.

Combined study cohort of patients with bladder cancer filtered by exclusion criteria and stratified by receipt of NAC (NAC cohorts I and II, CirGuidance study cohort and MOLecular upstaging cohort). Reasons for exclusion are listed in box (nonunique patients). Patients in non-NAC group were excluded if diagnosed with cT1 disease and were ≤pT1 at time of RC (see Materials and Methods).

Table 1.

Characteristics of patients with MIBC who underwent RC with and without NAC before and after IPW

To account for differences in baseline clinical characteristics between the NAC and non-NAC groups, we applied IPW. After IPW, patients in both the NAC and non-NAC groups had similar median age, percentage of female patients and clinical stage (table 1). The NAC cohort had a greater degree of pathological downstaging overall, with ypT0 rates of 21.7% vs pT0 rates of 10% for RC without NAC (p <0.001; supplementary table 1, https://www.jurology.com). Stratifying the meta-cohort by NAC treatment revealed the net benefit to OS and CSS at 2 years was 6% and 5%, respectively (fig. 2).

Figure 2.

Patient outcomes for IPW cohort stratified by treatment with NAC. A, OS. B, CSS. As estimates are weighted, proportions of patients at risk (as opposed to number at risk) are reported. CoxPH, Cox proportional hazards.

Luminal Patients Had No Significant Survival Benefit with Neoadjuvant Chemotherapy

Molecular subtypes were generated for the cohort using the GSC.¹¹ To evaluate whether response to NAC differed between GSC luminal and nonluminal tumors, the cohort was separated into luminal and nonluminal (basal, claudin-low, luminal infiltrated) groups for further analysis (table 1). Unless stated otherwise, all results were reported using the IPW cohort.

Interestingly, we found that the outcomes for patients with luminal tumors were comparable regardless of receipt of NAC. The OS of patients with luminal tumors receiving NAC was 82% at 2 years and 63% at 3 years (fig. 3, A). The rates were similar in patients treated without NAC, with OS rates of 78% and 65% at 2 and 3 years, respectively (fig. 3, A). Conversely, patients with nonluminal tumors who received NAC had higher 2 and 3-year OS of 78% and 71%, respectively, compared to 69% and 61% for patients with nonluminal tumors who underwent RC without NAC (fig. 3, B). On UVA, we found luminal tumors had no significant benefit to OS with receipt of NAC (HR 0.9, 95% CI 0.52–1.55, p=0.7), while nonluminal tumors had significant benefit (HR 0.62, 95% CI 0.42–0.91, p=0.02; table 2). The results of MVA were similar, with receipt of NAC associated with significantly improved OS among patients with nonluminal tumors (HR 0.66, 95% CI 0.44–0.98, p=0.04) but not among patients with luminal tumors (HR 0.91, 95% CI 0.50–1.63, p=0.74; table 2). There was no statistically significant interaction between luminal status and receipt of NAC (p=0.3; fig. 4, A and supplementary table 2, https://www.jurology.com).

Figure 3.

Patient outcomes for IPW cohort stratified by subtype and treatment with NAC. A, OS for luminal patients. B, OS for nonluminal patients. C, CSS for luminal patients. D, CSS for nonluminal patients. As estimates are weighted, proportions of patients at risk (as opposed to number at risk) are reported. CoxPH, Cox proportional hazards.

Table 2.

Inverse probability treatment weighted Cox UVA and MVA for OS of NAC treatment effect within GSC luminal and nonluminal subgroups

Figure 4.

Predicted rates at 2 years from IPW Cox analysis of interaction between receipt of NAC and GSC luminal/nonluminal for OS (A) and CSS (B).

Similar results were obtained for CSS, wherein receipt of NAC was not associated with improved outcomes among the luminal subset of patients (fig. 3, C). In contrast, patients with nonluminal tumors were noted to have an 11% survival advantage at 3 years with NAC (77% vs 66%; fig. 3, D). The results for UVA found no significant benefit for luminal tumors (HR 1.10, 95% CI 0.6–2.02, p=0.76), but nonluminal tumors had significant benefit (HR 0.57, 95% CI 0.37–0.88, p=0.01; table 3). Likewise, on MVA there was no evidence of a benefit of NAC for luminal tumors (HR 1.11, 95% CI 0.57–2.15, p=0.76), but nonluminal tumors benefited significantly (HR 0.59, 95% CI 0.38–0.93, p=0.02; table 3). As with OS, a significant interaction between luminal status and receipt of NAC was not determined for CSS (p=0.1; fig. 4, B and supplementary table 3, https://www.jurology.com). For both OS and CSS, secondary analyses of subcohorts where cT1, node positive cases or patients who received adjuvant chemotherapy were removed showed similar point estimates and confidence intervals with p values differing by 0.05 or less (supplementary table 4, https://www.jurology.com).

Table 3.

Inverse probability treatment weighted Cox UVA and MVA for CSSof NAC treatment effect within GSC luminal and nonluminal subgroups

Discussion

MIBC is a heterogeneous disease with noted highly variable responses in patients to NAC, including patients who experience little or no benefit to potentially toxic therapy. Indeed, 60% of patients have ≥ypT2 disease following NAC, and the outcome of patients with residual invasive disease is very poor.^14,15,21 Similar to prior randomized trials that found a survival benefit of NAC in patients with MIBC of 5%–10%, the OS benefit for NAC in our cohort was 6% and 7% at 2 and 3 years, respectively.^1–3 Therefore, many patients are not benefiting from standard of care therapy while being exposed to toxic side effects, further highlighting the need for predictive biomarkers of response to NAC at the moment MIBC is diagnosed.

Prior nonrandomized studies have suggested that patients with basal tumors had the greatest improvements in survival with NAC, while patients with luminal tumors did not benefit from treatment.^11,13,22 To confirm these findings, we first evaluated patient outcomes stratified by subtypes and found that patients with nonluminal tumors had an OS benefit with NAC prior to RC compared to RC alone of 9% and 10% at 2 and 3 years, respectively. Conversely, patients with luminal tumors have little or no survival benefit with NAC, which was corroborated on UVA and MVA.

Our results support the empirical notion that the modest survival benefit of NAC is due to a dilution of the large benefit that some patients obtain with the absence of benefit that other patients experience, including those with nonresponsive tumors, those with OC disease who have an excellent outcome without systemic therapy and those who are harmed by the morbidity of chemotherapy or delay in definitive local therapy. The ability to stratify patients based on a probability of NAC response would ensure patients who may benefit the most are encouraged to receive this therapy, while allowing patients unlikely to benefit to avoid the toxicity of therapy and reduce time to curative local therapy.

The data presented here strongly suggest that classification of patients based on molecular subtypes could inform patient and clinician decision making regarding the benefit of NAC, which could offer an improvement over the current uniform recommendation of NAC for all patients with MIBC. While the cohort in our study was assembled based on available data and is not derived from a clinical trial, it still suggests that if a genomic classifier were used to select patients for NAC in a population with similar subtype distribution, then 60% of patients selected to receive NAC (nonluminal tumors) may have a survival benefit if provided this therapy.

Further, these data reveal an intersection between the biological and clinical characteristics of MIBC, providing new insights that may partly explain the reduced benefit from NAC for patients for luminal tumors. Approximately 80%–90% of patients with pathological stage T1 or T2 tumors who have negative lymph nodes will be cured with cystectomy alone.^7,23 As such, if luminal tumors are more likely to have truly OC disease, then NAC might be less likely to demonstrate a survival advantage. The higher rate of upstaging for nonluminal tumors at time of RC may explain some of the survival advantage observed with NAC.

Several other biomarkers have been evaluated as predictive markers for their ability to predict response to NAC, including mutations in DNA repair genes ATM, RB1 and FANCC,^24,25 mutations in excision repair cross-complementation group 2 (ERCC2) gene^26,27 and protein biomarkers.²⁸ Utilizing biomarkers to enrich the response of patients undergoing NAC has the potential to increase the utilization of NAC in those patients most likely to respond and spare those unlikely to respond from potential toxicities associated with treatment. However, in the recent SWOG S1314 trial, the gene expression-based COXEN (co-expression extrapolation) scores for gemcitabine/cisplatin and dose-dense MVAC (methotrexate, vinblastine, doxorubicin, cisplatin) were not predictive of response to the corresponding therapy.²⁹ Other ongoing clinical trials (RETAIN: NCT02710734; A031701: NCT03609216; HCRN GU16-257: NCT03558087) are using DNA repair gene mutation status to stratify patients to bladder sparing therapy if the patient achieves a good response to treatment. The data presented in this study set the stage for biomarker selected randomized clinical trials, comparing NAC for all-comers to subtype-selected treatment.

There are limitations to this analysis, including the retrospective study design with risk of selection bias. Even though this was an MIBC cohort, including patients who were cT1 but were found to be pT2 (or higher) constitutes potential selection of less aggressive disease in the non-NAC group. However, when we excluded cT1 and cTanyN1-3 patients, similar results were obtained. The cohorts in this study were also combined in several instances, despite having similar tissue types, platforms and normalization methods, and we cannot eliminate the potential influence of cohort-level batch effects. These types of analyses of patients who underwent RC with or without NAC are not replacements for randomized trials as there are inherent biases regarding which patients undergo NAC or RC alone. Three of the 4 cohorts in our study were treated with either NAC and RC or RC alone, which has a potential treatment-effect bias in the study and is a further limitation. The interaction term analysis between subtypes and receipt of NAC was limited by the sample size and therefore not included as part of the study design. Finally, the groups in the study were not uniformly assembled, and some patients were excluded due to inadequate data or followup.

Conclusions

To conclude, in this multicenter cohort of 601 MIBC patients treated with RC with or without NAC, we found that, based on whole transcriptome profiling, nonluminal tumors received the greatest benefit from NAC, while patients with luminal tumors experienced little or no survival benefit from NAC. Utilizing a genomic classifier to select MIBC patients for NAC could enrich for patients most likely to benefit while identifying patients who may not benefit from additional treatments.

Supplementary Material

juro-207-541-s001.pdf ^{(149.2KB, pdf)}