EURURO Vol. 72 No. 4

552 656

In multivariable analysis adjusting for clinical tumor
stage, age, and gender in the non-NAC cohort, patients with
a GSC basal subtype had a hazard ratio of 2.22 (p = 0.002;
for OS compared with the luminal subtype. In
contrast, in the NAC cohort, GSC basal subtype patients did
not fare differently from patients with luminal tumors
(hazard ratio: 0.84, p = 0.61; ). Similar results were
observed when considering only those patients from the
independent NAC dataset (n = 69) not used for GSC model
training ). Overall, this analysis suggested that the
outcome of patients with basal tumors may improve the
most with NAC.
3.6. NAC survival benefit in GSC basal tumors is independent of
pathological response
As in our analysis using the previously described subtyping
methods, GSC was not significantly associated with a major
pathological response (ie, ypT < 2N0) in the NAC cohort
(Supplementary Tables 3 and 5), even though the major
pathological response was associated with improved OS
(p < 0.001; Supplementary Fig. 7 and Supplementary
Table 6). In an exploratory analysis, we further compared
NAC responders (n = 108) and nonresponders (n = 143) in
each GSC subtype B). Patients with luminal tumors
who experienced a major response (n = [21_TD$DIFF]48) had a 3-yr OS of
95% (95% CI 89–100%) compared with 58% (95% CI 44–76%)
in nonresponders (n = [22_TD$DIFF]54, p = 0.002; B, lower right). In
stark contrast, patients with GSC basal tumors did not show
any significant differences in OS between major responders
and nonresponders , upper right and Supplementary
Table 7). The findings in the luminal-infiltrated and claudin-
low tumors were more similar to the luminal tumors,
although the sample sizes of these subgroups were too small
and the duration of follow-up too short to allow definitive
conclusions B, left panels and Supplementary Table 8).
4. Discussion
The recent molecular characterization of MIBC, including the
description of subtypes based on gene expression ,
provides a framework for further study of this frequently
lethal malignancy and offers potential biological insight into
different clinical phenotypes. One potential immediate
impact of the molecular subtyping is to guide selection of
optimal therapy. This concept was first suggested by Choi
et al in the context of cisplatin-based NAC, and
subsequently by two trials in the context of systemic
checkpoint inhibition Here, we have provided more
evidence in a large patient cohort that outcome after NAC
varies by molecular subtype. Furthermore, we have devel-
oped the methodology for single patient subtyping in a CLIA-
certified laboratory , which represents a significant step
toward potential clinical application of molecular subtyping.
The most important finding of our study was the relative
shift in outcome by subtype in patients with and without
NAC, albeit based on a comparison across studies. Our
findings support the clinical utility of the four subtypes.
Patients with basal tumors appear to derive the most
benefit from NAC. These highly proliferative basal tumors
demonstrated a poor prognosis when treated with surgery
alone but their prognosis was dramatically improved
in the NAC cohort. Patients with luminal nonimmune-
infiltrated tumors had the best prognosis, irrespective of the
treatment strategy, implying that these patients do not
appear to derive benefit from NAC. The pathological stage
and prognosis of patients with luminal immune-infiltrated
tumors were significantly worse than those with luminal
noninfiltrated tumors. Patients with luminal-infiltrated
tumors appear to have poor prognosis with and without
NAC.
Of a potentially high clinical impact, the patients with
luminal-infiltrated tumors (corresponding to TCGA cluster
II) seemed to benefit most from checkpoint inhibition with
azetolizumab in the IMvigor 210 trial . However, in the
recently published CheckMate 275 trial, patients with
basal tumors appeared to benefit from nivolumab, another
checkpoint inhibitor Unfortunately, the gene expres-
sion data from neither of these trials have been made
publicly available and the methodology for assignment of
patients to TCGA clusters has not been revealed. Therefore,
it is not currently possible to draw conclusions on the
impact of subtyping on response to checkpoint blockade.
Clinical trials will need to address the relative merit of
NAC and perioperative checkpoint inhibition in these
patients.
Table 2 – Results of Cox proportional hazard analysis of GSC’s ability to predict overall survival (NAC and non-NAC)
Variable MVA MV MV
Non-NAC set (n = 476) All NAC set (n = 269) NAC validation set (n = 69)
HR 95% CI p value HR 95% CI p value HR 95% CI p value
Age 1.02 1–1.05 0.066 1.01 0.99–1.04 0.235 0.98 0.93–1.03 0.433
Female (Ref) 1 – 1.000 1 – 1.000 1 – 1.000
Male 0.81 0.53–1.24 0.330 1.12 0.67–1.86 0.661 4.38 0.7–27.52 0.115
Luminal (Ref) 1 – 1.000 1 – 1.000 1 – 1.000
Inf-luminal 2.38 1.33–4.28 0.004 2.46 1.29–4.7 0.006 5.68 0.4–81.3 0.201
Basal 2.22 1.34–3.68 0.002 0.84 0.42–1.68 0.614 0.88 0.16–4.94 0.881
Claudin-low 3.06 1.71–5.47 <0.001 2.16 1.22–3.81 0.008 3.73 0.81–17.25 0.092
CI = confidence interval; GSC = genomic subtyping classifier; HR = hazard ratio; Inf = infiltrated; NAC = neoadjuvant chemotherapy; Ref reference.
a MVA models adjusted for institution and clinical stage.
E U R O P E A N U R O L O G Y 7 2 ( 2 0 1 7 ) 5 4 4 – 5 5 4 552