464 476
2.4.
Statistical procedures
STATA 12.0 was used for statistical analysis. Multivariable Cox regression
was performed using backward selection from a model starting with all
significant (
p
0.05) variables. The assumptions of proportional hazards
were verified, and variables were tested for collinearity before
multivariable regression analysis. The predictive accuracy of prognostic
models was quantified using Harrell’s concordance index. Data were not
assessed using competing-risks methods as information about death from
other causes was not available in most cases.
3.
Results
In this Europeanmulticentre prospective study of molecular
risk stratification in NMIBC we enrolled a total of
1224 patients from ten hospitals in Denmark, Sweden,
Germany, The Netherlands, Spain, and Serbia. We obtained
sufficient amounts of RNA for RT-qPCR analysis (median
RIN 9, median carcinoma cell percentage 90%) from 750 of
1224 patients who fulfilled all the criteria for testing
(851 tumours; Supplementary Table 1,
Fig. 1A). No signifi-
cant differences in stage and grade distribution were
observed between the excluded and included patients;
however, the frequency of large tumours was, as expected,
significantly higher in the included patient cohort
(
p
= 0.006;
x
2
test). For the included patient cohort, the
median follow-up was 28 mo (range 0–76), during which
37 progressions to MIBC occurred. A progression score was
calculated for each tumour based on the 12-gene RT-qPCR
assay, and previously defined cut-off values were applied
(cut-off
optimal
and cut-off
90% sensitivity
)
[17].
3.1.
12-gene progression score performance (first tumour
analysed)
We analysed the progression scores using the first tumour
included from each patient. A comparison of gene expres-
sion patterns for the 12 genes, clinical and histopathological
information, and progression score distribution is shown in
Fig. 1B. An overview of the clinical and histopathological
characteristics for all cases stratified according to dichot-
omised progression score (cut-off
optimal
) is shown in
Table 1. We observed a difference in tumour classification
based on the centre of origin, especially between the two
largest contributing centres, Frederiksberg and Skejby. This
may reflect a difference in patient enrolment at the different
centres. A high progression score was significantly
(
p
0.001) associated with high age, high stage, high grade,
concomitant CIS, BCG treatment, progression to MIBC, and
high EORTC risk score. Furthermore, univariate Cox regres-
sion analysis showed that stage, grade, growth pattern,
EORTC risk score, and progression score were highly
significantly associated with disease progression to MIBC
( Table 2). Kaplan-Meier estimates of cumulative incidence
as a function of the dichotomised progression scores using
the two predefined cut-off values are shown in
Fig. 2 .The use of continuous progression score and EORTC risk
score as variables increased the prognostic power and
predictive accuracy, so these were included as continuous
Table 1 – Clinical and histopathological information for all
patients stratified by progression score (optimal cut-off)
aCharacteristics
Progression score
p
value
Low risk
(
n
= 390)
High risk
(
n
= 360)
Clinical centre,
n
(%)
0.012
Aalborg
17 (4)
15 (4)
Barcelona
27 (7)
27 (8)
Belgrade
34 (9)
13 (4)
Erlangen
25 (6)
26 (7)
Frederiksberg
101 (26)
75 (21)
Jena
13 (3)
28 (8)
Munich
27 (7)
17 (5)
Rotterdam
15 (4)
15 (4)
Skejby
123 (32)
132 (37)
Uppsala
8 (2)
12 (3)
Median age (yr)
69
72
<
0.001
Gender,
n
(%)
0.663
Male
299 (77)
281 (78)
Female
91 (23)
79 (22)
Smoking status,
n
(%)
0.027
Current
132 (34)
92 (26)
Former
114 (29)
135 (38)
Never
47 (12)
51 (14)
Unknown
97 (25)
82 (23)
Primary tumour,
n
(%)
0.023
Yes
219 (56)
172 (48)
No
171 (44)
188 (52)
Stage,
n
(%)
<
0.001
Ta
331 (85)
201 (56)
T1
26 (7)
55 (15)
T1a
26 (7)
69 (19)
T1b
1 (0)
19 (5)
T1c
1 (0)
8 (2)
CIS
5 (1)
8 (2)
Grade (WHO 2004),
n
(%)
<
0.001
High
95 (24)
215 (60)
Low
271 (70)
136 (38)
PUNLMP
12 (3)
0 (0)
Unknown
12 (3)
9 (2)
CIS
<
0.001
Yes
31 (8)
78 (22)
No
359 (92)
282 (78)
Growth pattern,
n
(%)
0.090
Papillary
345 (88)
296 (82)
Mixed
6 (2)
11 (3)
Solid
6 (2)
11 (3)
Unknown
33 (8)
42 (12)
Size,
n
(%)
0.051
3 cm
48 (12)
65 (18)
<
3cm
230 (59)
210 (58)
Unknown
112 (29)
85 (24)
Multiplicity,
n
(%)
0.333
Single tumour
284 (73)
250 (69)
Multiple tumours
106 (27)
110 (31)
BCG treatment,
n
(%)
0.001
Yes
46 (12)
77 (21)
No
344 (88)
283 (79)
Progression to MIBC,
n
(%)
<
0.001
Yes
7 (2)
30 (8)
No
383 (98)
330 (92)
EORTC risk score,
n
(%)
<
0.001
Low (0)
146 (38)
51 (14)
Intermediate (1–6)
161 (41)
115 (32)
High (
>
6)
83 (21)
194 (54)
BCG = bacillus Calmette-Gue´rin; Cis = carcinoma in situ diagnosed any
time in the disease course; PUNLMP = papillary urothelial neoplasm of low
malignant potential;
MIBC = muscle-invasive bladder cancer;
EORTC = European Organisation for Research and Treatment of Cancer.
a
Calculations are based on the predefined optimal cut-off value for the
progression score, and based on the first tumour included for each patient in
the study. Fischer’s exact test was used except for calculation of differences
in patient age, for which a
t
test was used.
E U R O P E A N U R O L O G Y 7 2 ( 2 0 1 7 ) 4 6 1 – 4 6 9
464