Inter
national
J
our
nal
of
Adv
ances
in
A
pplied
Sciences
(IJ
AAS)
V
ol.
15,
No.
3,
September
2026,
pp.
1193
∼
1202
ISSN:
2252-8814,
DOI:
10.11591/ijaas.v15.i3.pp1193-1202
❒
1193
Mapping
signicant
clinical
factors
to
r
elapse-fr
ee
sur
vi
v
al
pr
ognosis
in
br
east
cancer
patients
Boby
Al
Qurthuby
,
I
Made
Murwantara
Master
of
Informatics,
F
aculty
of
Articial
Intelligence
and
Data
Science,
Uni
v
ersitas
Pelita
Harapan,
Jakarta,
Indonesia
Article
Inf
o
Article
history:
Recei
v
ed
May
27,
2025
Re
vised
Jun
14,
2026
Accepted
Aug
13,
2026
K
eyw
ords:
Breast
cancer
Cox
re
gression
Kaplan-Meier
Relapse-free
survi
v
al
Survi
v
al
analysis
ABSTRA
CT
This
study
e
xplores
the
role
of
clinical
f
actors
in
inuencing
relapse-free
survi
v
al
(RFS)
in
breast
cancer
patients
through
comprehensi
v
e
survi
v
al
analysis
(SA).
The
data
were
deri
v
ed
from
a
lar
ge
cohort
of
breast
cancer
patients,
encompassing
clinical
v
ariables
such
as
age
at
diagnosis,
tumor
size,
number
of
positi
v
e
lymph
nodes,
tumor
grade,
histological
subtype,
and
receipt
of
chemotherap
y
and
radiotherap
y
.
Analytical
methods
included
Kaplan-Meier
plots,
log-rank
tests,
Cox
proportional
hazards
modeling,
and
feature
ranking
based
on
uni
v
ariate
and
multi
v
ariate
log-rank
and
Cox
tests.
The
results
sho
w
that
patients
who
recei
v
ed
chemotherap
y
demonstrated
a
shorter
median
RFS
compared
to
those
who
did
not
(102.7
vs.
145.0
months,
p
=
0.03),
lik
ely
due
to
high-risk
patie
nts
being
selected
for
chemotherap
y
.
Meanwhile,
radiotherap
y
w
as
associated
wit
h
impro
v
ed
prognosis
(hazard
ratio
(HR)
<
1
in
Cox
model,
p
<
10
−
5
),
in
line
with
meta-analyses
sho
wing
reduced
breast
cancer
mortality
post-radiotherap
y
.
The
contrib
uti
on
maps
the
k
e
y
clinical
determinants
of
RFS
in
breast
cancer
and
highlights
the
practicality
of
performing
SA.
This
is
an
open
access
article
under
the
CC
BY
-SA
license
.
Corresponding
A
uthor:
I
Made
Murw
antara
Master
of
Informatics,
F
aculty
of
Articial
Intelligence
and
Data
Science,
Uni
v
ersitas
Pelita
Harapan
Jakarta,
Indonesia
Email:
made.murw
antara@uph.edu
1.
INTR
ODUCTION
Breast
cancer
remains
one
of
the
most
pre
v
alent
oncological
diseases
among
w
omen
globally
[1],
[2],
contrib
uting
substantially
to
morbidity
and
mortality
rates.
Despite
adv
ancements
in
diagnostic
and
therapeutic
modalities,
the
risk
of
relapse
continues
to
pose
a
signicant
clinical
challenge.
Relapse-free
survi
v
al
(RFS)
[3],
dened
as
the
time
from
initial
treatment
to
the
recurrence
of
disease,
is
a
crucial
endpoint
for
e
v
aluating
treatment
ef
cac
y
and
tumor
characteristics.
Pre
vious
studies
ha
v
e
demonstrated
that
traditional
clinical
f
actors
such
as
tumor
grade,
tumor
size,
and
lymph
node
in
v
olv
ement
play
a
signicant
role
in
determining
RFS.
Ka
w
ase
et
al
.
[4]
found
that
higher
-grade
tumors
and
positi
v
e
nodal
status
were
independently
associated
with
shorter
RFS,
e
v
en
after
mult
i
v
ariate
analysis.
Lar
ge
cohort
studies
ha
v
e
consistently
sho
wn
that
increases
in
tumor
size
and
nodal
spread
signal
a
w
orsening
prognosis
[5].
In
addition
to
that,
biological
mark
ers
such
as
hormone
receptor
status
and
histological
subtypes
[6]–[8]
are
kno
wn
to
present
dif
fering
prognoses
[9].
Where
hormone
receptor
status
[10]
refers
to
whether
a
tumor’
s
cells
e
xpress
receptors
for
the
female
se
x
hormones—estrogen
(ER)
[11]
and
progesterone
(PR)
[12]—which
dri
v
e
cell
gro
wth
and
dif
ferentiation
in
hormone-sensiti
v
e
tissues.
Adjuv
ant
interv
entions
[13],
including
chemotherap
y
[14]
and
radiotherap
y
[15],
play
a
s
ignicant
role
in
impro
ving
RFS
outcomes
in
cancer
patients.
These
treatments,
administered
after
the
primary
J
ournal
homepage:
http://ijaas.iaescor
e
.com
Evaluation Warning : The document was created with Spire.PDF for Python.
1194
❒
ISSN:
2252-8814
interv
ention
such
as
sur
gery
,
are
designed
to
eliminate
residual
disease
and
reduce
the
lik
elihood
of
recurrence,
thereby
e
xtending
the
duration
of
remission
and
impro
ving
long-term
survi
v
al
rates.
Among
the
v
arious
chemotherap
y
re
gimens,
anthrac
ycline-based
[16]
adjuv
ant
therap
y
has
been
e
xtensi
v
ely
studied
and
widely
adopted
in
clinical
practice,
particularly
in
the
management
of
breast
cancer
.
Evidence
from
lar
ge-scale
clinical
trials
and
meta-analyses
indicates
that
this
class
of
drugs
can
signicantly
reduce
breast
cancer
mortality
.
The
e
xtent
of
the
benet,
ho
we
v
er
,
can
v
ary
based
on
se
v
eral
patient-related
f
actors,
including
age
and
tumor
characteristics.
Specically
,
anthrac
ycline-based
re
gimens
ha
v
e
been
associated
[17],
[18],
with
a
20%
to
38%
reduction
in
mortality
,
demonstrating
a
substantial
impro
v
ement
in
prognosis
when
used
as
part
of
a
comprehensi
v
e
treatment
strate
gy
.
Radiotherap
y
[19],
particularly
after
sur
gical
resection
of
the
primary
tumor
,
is
another
cri
tical
component
of
adjuv
ant
treatment.
Its
primary
objecti
v
e
is
to
eradicate
an
y
microscopic
residual
tumor
cells
at
the
site
of
sur
gery
,
which
are
often
responsible
for
local
recurrence.
Numerous
studies
ha
v
e
sho
wn
that
postsur
gical
radiotherap
y
not
only
lo
wers
the
risk
of
local
recurrence
b
ut
also
contrib
utes
to
a
meaningful
decrease
in
long-term
mortality
.
F
or
patients
under
going
breast-conserving
sur
gery
,
the
addition
of
radiotherap
y
has
become
a
standard
of
care
due
to
its
pro
v
en
ability
to
enhance
local
control
and
survi
v
al
outcomes.
The
inte
gration
of
these
adjuv
ant
modalities
into
treatment
protocols
has
led
to
notable
impro
v
ements
in
RFS
curv
es
in
v
arious
types
of
cancer
,
especially
early
stage
breas
t
cancer
.
The
benets
are
particularly
e
vident
in
the
subgroups
of
patients
with
high-risk
pathological
characteristics,
where
the
risk
of
recurrence
without
adjuv
ant
therap
y
w
ould
otherwise
be
substanti
al.
Importantly
,
the
ef
fecti
v
eness
of
adjuv
ant
therap
y
must
be
balanced
ag
ainst
potential
toxicities
and
patient
comorbidities,
which
can
inuence
treatment
tolerance
and
results.
This
study
contrib
utes
by
enhancing
the
analytical
capability
to
in
v
estig
ate
the
relationship
between
v
arious
clinical
v
ariables
and
RFS
in
breast
cancer
patients.
By
applying
survi
v
al
analysis
(SA)
model,
it
of
fers
deeper
insights
into
ho
w
specic
clinical
f
actors
inuence
patient
outcomes,
potentially
guiding
more
ef
fecti
v
e
treatment
and
follo
w-up
strate
gies.
2.
METHOD
2.1.
Data
collection
Dataset
for
this
study
w
as
from
the
molecular
taxonomy
of
breast
cancer
international
consortium
(MET
ABRIC)
[20]–[22].
MET
ABRIC
analyzed
genomic
and
transcriptomic
data
from
approximately
2,000
primary
breast
tumors,
inte
grating
this
information
with
long-term
clinical
outcomes.
This
comprehensi
v
e
approach
enabled
the
identication
of
ten
distinct
m
olecular
subtypes
of
breast
cancer
,
each
characterized
by
unique
genomic
dri
v
ers
and
associated
with
specic
clinical
prognoses.
The
MET
ABRIC
dataset
a
v
ailable
via
cBioPortal
and
the
European
genome-phenome
archi
v
e.
K
e
y
features
in
the
dataset
include
age
at
diagnosis,
which
serv
es
as
an
important
demographic
f
actor
kno
wn
to
inuence
disease
progression
and
treatment
response.
T
umor
size,
measured
in
millimeters,
and
the
number
of
positi
v
e
lymph
nodes
are
included
as
indicators
of
cancer
stage
and
spread—both
strong
prognostic
mark
ers
for
SA.
T
umor
grade,
classied
from
I
to
III,
reects
the
le
v
el
of
abnormality
of
the
cancer
cells,
with
higher
grades
typically
associated
with
more
aggressi
v
e
disease.
Histological
subtypes
[23],
such
as
ductal,
lob
ular
,
tub
ular
,
and
mucinous
carcinomas,
is
also
recorded
to
account
for
biological
heterogeneity
within
breast
cancer
cases.
These
subtypes
dif
fer
in
clinical
beha
vior
and
may
respond
dif
ferently
to
therap
y
.
Additionally
,
treatment
v
ariables
such
as
the
admini
stration
of
chemotherap
y
(yes/no)
and
the
receipt
of
radiotherap
y
(yes/no)
are
included.
These
v
ariables
are
particularly
rele
v
ant
as
the
y
are
kno
wn
to
inuence
RFS
outcomes
by
reducing
the
risk
of
recurrence.
2.2.
Data
pr
epr
ocessing
The
data
preprocessing
in
this
study
w
as
conducted
using
the
Orange3
platform
and
comprised
se
v
eral
essential
steps.
The
MET
ABRIC
dataset,
which
includes
1,904
breast
cancer
patients
and
o
v
er
24,000
clinical
and
genetic
features,
serv
ed
as
the
primary
data
source.
Cate
gorical
v
ariables
such
as
tumor
grade,
chemotherap
y
,
and
radiotherap
y
were
con
v
erted
into
numerical
format
using
the
“First
v
alue
as
base”
method
to
f
acilitate
statistical
analysis.
Numerical
features
including
tumor
size,
age
at
diagnosis,
and
lymph
nodes
e
xamined
positi
v
e
were
normalized
to
a
[0,
1]
range
and
subsequently
discretized
into
three
interv
als
using
the
equal
frequenc
y
discretization
method
for
Kaplan-M
eier
analysis.
Rele
v
ant
features
were
selected,
consisting
of
six
clinical
predictors
and
RFS
time
as
the
tar
get
v
ariable.
Missing
v
alues
were
imputed
using
the
a
v
erage
or
most
frequent
method
to
ensure
a
complete
and
analyzable
dataset.
Finally
,
the
data
were
Int
J
Adv
Appl
Sci,
V
ol.
15,
No.
3,
September
2026:
1193–1202
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Adv
Appl
Sci
ISSN:
2252-8814
❒
1195
transformed
into
a
survi
v
al
format
by
dening
RFS
time
as
the
time
v
ariable
and
survi
v
al
e
v
ent
as
the
censoring
indicator
.
2.3.
Sur
vi
v
al
analysis
This
w
ork
implement
SA
that
is
a
statistical
approach
used
to
analyze
and
interpret
time-to-e
v
ent
dat
a,
commonly
applied
in
clinical
and
epidemiological
studies.
It
focuses
on
estimat
ing
the
probability
of
an
e
v
ent
occurring
o
v
er
time
while
accounting
for
censored
observ
ations.
K
e
y
components
include
the
survi
v
al
function,
hazard
function,
and
models
such
as
Kapl
an-Meier
estimator
and
Cox
proportional
hazards
re
gression.
These
methods
allo
ws
us
to
identify
f
actors
inuencing
survi
v
al
time
and
assess
dif
ferences
between
groups.
The
SA
is
essentia
l
in
longitudinal
studies
where
the
timing
of
e
v
ents
is
critical
and
traditional
statistical
met
hod
s
may
be
inadequate.
The
anal
ysis
w
orko
w
w
as
structured
as
follo
ws:
the
dataset
w
as
rst
loaded
and
then
the
non-clini
cal
columns
were
e
xcluded
to
reduce
noise.
The
initial
step
in
v
olv
ed
uni
v
ariate
analysis
via
the
Kaplan–Meier
plot,
where
users
can
partition
the
dataset
by
cate
gorical
v
ariables
and
visualize
survi
v
al
curv
es
along
with
median
RFS
for
each
subgroup.
Subsequently
,
tw
o
parallel
analytical
procedures
were
e
x
ecuted:
rank
survi
v
al
features
[24]
w
as
used
to
rank
clinical
v
ariables
based
on
uni
v
ariate
Cox
tests
or
multi
v
ariate
log-rank
tests
and
Cox
proportional
hazards
[25]
w
as
applied
to
model
simultaneous
ef
fects
of
multiple
clinical
v
ariables
on
RFS.
The
Cox
model
w
as
implemented
using
all
independent
v
ariables
concurrently
(multi
v
ariate
analysi
s)
to
estimate
the
hazard
ratio
(HR
)
for
each
co
v
ariate.
In
addition,
feature
ranking
w
as
conducted
using
both
the
Cox
log-lik
elihood
ratio
test
(uni
v
ariate)
and
the
multi
v
ariate
log-rank
test.
These
rankings
generated
a
prioritized
list
of
statistically
signicant
features
contrib
uting
to
RFS
outcomes.
The
interpretation
of
the
results
w
as
conducted
using
both
p-v
alues
and
ef
fect
sizes
to
ensure
a
comprehensi
v
e
understanding
of
the
statistical
signicance
and
practical
rele
v
ance
of
the
ndings.
p-v
alues
were
utilized
to
determine
whether
the
observ
ed
a
ssociations
or
dif
ferences
were
statistically
signicant,
typically
e
v
aluated
ag
ainst
a
predened
threshold
(e.g.,
p
<
0.05).
Ho
we
v
er
,
recognizing
that
statistical
signicance
alone
does
not
imply
meaningful
impact,
ef
fect
sizes
were
also
calculated
to
quantify
the
magnit
ud
e
of
the
observ
ed
ef
fects.
This
dual
approach
enabled
a
more
nuanced
interpretation,
balancing
statistical
rigor
wi
th
the
real-w
orld
implications
of
the
results
across
the
v
ariables
e
xamined.
3.
RESUL
TS
AND
DISCUSSION
This
section
presents
the
results
of
the
SA
conducted
using
the
MET
ABRIC
dataset
through
the
Orange3
data
mining
platform.
The
analysis
aimed
to
identify
clinical
v
ariables
signicantly
as
sociated
with
RFS
in
patients
with
breast
cancer
.
A
total
of
1,904
patient
records
were
processed,
and
six
clinical
v
ariables
were
in
v
estig
ated:
tumor
grade,
lymph
nodes
e
xamined
positi
v
e,
chemotherap
y
,
radiotherap
y
,
age
at
diagnosis,
and
tumor
size.
3.1.
Kaplan–Meier
sur
vi
v
al
estimates
Kaplan–Meier
survi
v
al
curv
es
were
generated
to
e
xamine
the
RFS
probability
across
stratied
groups
for
each
clinical
v
ariable.
Each
curv
e
represents
the
survi
v
al
function
o
v
er
time,
and
the
dif
ferences
between
groups
were
statistically
tes
ted
using
the
log-rank
test.
These
curv
es
pro
vide
visual
and
statistical
insights
into
the
heterogeneity
of
clinical
outcomes
based
on
k
e
y
patient
characteristics
and
therapeutic
interv
entions.
Figure
1
s
h
o
ws
Kaplan–Meier
curv
es
of
RFS
stratied
by
tumor
size.
P
atients
with
the
small
est
primary
tumors
(
<
19.5
mm,
blue
curv
e)
e
xhibit
substantially
higher
survi
v
al
than
those
with
intermediate
or
lar
ge
tumors.
In
particular
,
the
blue
curv
e
(small
tumors)
has
a
median
RFS
of
199.1
months,
whereas
the
green
curv
e
(tumors
≥
26.5
mm)
has
a
median
of
only
86.6
mont
hs.
The
condence
bands
(shaded
re
gions)
indicate
minimal
o
v
erlap
between
the
e
xtreme
groups.
F
or
e
xample,
by
150
months
only
about
30%
of
the
lar
ge-tumor
group
remain
relapse-free,
compared
to
roughly
60%
of
the
small-tumor
group.
The
log-rank
test
yields
p
<
0.005,
conrming
that
these
dif
ferences
are
highly
signicant.
Notably
,
all
curv
es
decline
from
near
100%
survi
v
al
at
time
zero,
b
ut
the
lar
ge-tumor
curv
e
drops
much
more
steeply
during
early
follo
w-up,
indicating
earlier
relapses
in
that
group.
Throughout
the
follo
w-up,
the
separation
persists:
at
100
months
roughly
50%
of
the
small-tumor
patients
remain
e
v
ent-free,
whereas
the
majority
of
patients
with
lar
ge
tumors
ha
v
e
already
relapsed.
These
observ
ations
demonstrat
e
that
increasing
tumor
size
is
strongly
associated
with
higher
hazard
of
relapse
in
this
cohort.
T
umor
size
thus
emer
ges
as
a
po
werful
prognostic
f
actor
,
since
e
v
en
Mapping
signicant
clinical
factor
s
to
r
elapse-fr
ee
survival
pr
o
gnosis
in
br
east
cancer
...
(Boby
Al
Qurthuby)
Evaluation Warning : The document was created with Spire.PDF for Python.
1196
❒
ISSN:
2252-8814
the
moderate-sized
group
(red
curv
e)
lies
between
the
e
xtremes
in
both
median
survi
v
al
(148.1
months)
and
curv
e
trajectory
.
Figure
1.
Kaplan-Meier
curv
es
of
RFS
stratied
by
tumor
size
Figure
2
presents
Kaplan–Meier
curv
es
stratied
by
age
at
diagnosis.
P
atients
diagnosed
at
a
younger
age
(
<
55.6
years,
blue
curv
e)
ha
v
e
mark
edly
better
survi
v
al
than
older
patients
(
≥
67.9
years,
green
curv
e),
with
the
middle-age
group
(55.6–67.9
years,
red)
in
between.
The
median
RFS
is
246.2
months
for
the
youngest
group
and
only
101.6
months
for
the
oldest
group.
The
condence
interv
als
for
the
youngest
and
oldest
groups
barely
o
v
erlap,
and
the
log-rank
test
yields
p
<
0.005.
This
indicates
a
highly
signicant
trend
to
w
ard
w
orse
outcomes
with
increasing
age.
Earl
y
in
follo
w-up,
the
dif
ference
is
already
apparent:
at
80
months
about
60%
of
the
youngest
group
are
still
relapse-free
compared
to
roughly
40%
of
the
oldest
group.
By
150
months
the
survi
v
al
g
ap
has
widened
further
.
The
consistent
ordering
and
parallel
decline
of
the
curv
es
suggest
that
the
HR
for
age
remains
roughly
constant
o
v
er
time
(no
crossing
of
curv
es).
These
results
imply
that
older
age
at
diagnosis
confers
a
substantially
higher
relapse
risk.
In
practical
terms,
younger
patients
in
this
dataset
enjo
y
signicantly
longer
RFS.
This
trend
may
reect
age-rel
ated
dif
ferences
in
tumor
biology
and
patient
resilience,
b
ut
re
g
ardless
of
cause,
age
is
clearly
an
important
prognostic
co
v
ariate
in
the
MET
ABRIC
cohort.
Figure
3
sho
ws
RFS
stratied
by
chemotherap
y
status.
In
this
case,
patients
who
recei
v
ed
chemotherap
y
(red
curv
e)
ha
v
e
w
orse
survi
v
al
than
those
who
did
not
(blue
curv
e).
The
median
RFS
for
the
chemotherap
y
group
is
102.7
months,
compared
to
145.0
months
for
the
no-chemotherap
y
group.
The
condence
bands
o
v
erlap
some
what,
and
the
log-rank
test
yields
p
=
0.03,
i
ndicating
a
statistically
signicant
(albeit
weak
er)
dif
ference.
Notably
,
at
early
times
(e.g.
50
months)
the
no-chemotherap
y
curv
e
is
around
90%
survi
v
al
v
ersus
about
80%
for
the
chemotherap
y
curv
e.
Ov
er
the
rst
100
months
the
g
ap
persists
such
that
by
100
months
roughly
70%
of
the
no-chemotherap
y
patients
are
e
v
ent-free
compared
to
only
50%
of
the
chemotherap
y
group.
Be
yond
150
months
the
curv
es
tend
to
con
v
er
ge
due
to
small
numbers.
These
ndings
suggest
that
patients
who
recei
v
ed
chemotherap
y
had
a
higher
hazard
of
relapse.
Ho
we
v
er
,
this
result
lik
ely
reects
confounding
by
indication:
patients
selected
for
chemotherap
y
often
had
more
aggressi
v
e
tumor
features
at
baseline.
W
ithout
adjustment,
the
Kaplan–Meier
plot
cannot
distinguish
between
a
true
harmful
ef
fect
of
chemotherap
y
and
the
f
act
that
the
chemo
group
w
as
inherently
higher
-risk.
Therefore,
while
the
Int
J
Adv
Appl
Sci,
V
ol.
15,
No.
3,
September
2026:
1193–1202
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Adv
Appl
Sci
ISSN:
2252-8814
❒
1197
log-rank
test
indicates
a
signicant
survi
v
al
dif
ference
(p
=
0.03),
this
should
not
be
interpreted
as
e
vidence
that
chemotherap
y
reduces
survi
v
al;
rather
,
it
highlights
the
cli
nical
reality
that
treated
patients
in
this
sample
f
ared
w
orse
on
a
v
erage,
presumably
due
to
more
adv
anced
disease.
Figure
2.
Kaplan-Meier
curv
es
stratied
by
age
at
diagnosis
Figure
3.
Kaplan-Meier
chemotherap
y
status
Mapping
signicant
clinical
factor
s
to
r
elapse-fr
ee
survival
pr
o
gnosis
in
br
east
cancer
...
(Boby
Al
Qurthuby)
Evaluation Warning : The document was created with Spire.PDF for Python.
1198
❒
ISSN:
2252-8814
Figure
4
demonstrate
that
patients
with
<
0.5
positi
v
e
lymph
nodes
had
the
highest
median
RFS,
reaching
171.9
months,
whereas
pa
tients
with
≥
2.5
positi
v
e
lymph
nodes
e
xhibited
the
lo
west
median
RFS
of
64.2
months.
This
dif
ference
w
as
statistically
signicant
(p
<
0.005),
pro
viding
further
e
vidence
that
the
e
xtent
of
lymph
node
in
v
olv
ement
is
strongly
associated
with
the
risk
of
disease
recurrence.
An
increasing
number
of
posit
i
v
e
lymph
nodes
w
as
as
sociated
with
a
progressi
v
ely
poorer
prognosis,
as
greater
lymphatic
in
v
olv
ement
may
indicate
more
e
xtensi
v
e
dissemination
of
malignant
cells
and
a
higher
lik
elihood
of
systemic
metastasis.
Consequently
,
patients
in
the
group
with
≥
2.5
positi
v
e
lymph
nodes
demonstrated
substantially
shorter
RFS
and
may
represent
a
subgroup
with
a
higher
risk
of
recurrence
requiring
closer
clinical
surv
eillance
and
potentially
more
intensi
v
e
therapeutic
management.
Figure
4.
Kaplan-Meier
lymph
nodes
Figure
5
compares
RFS
between
patients
who
did
(red
curv
e)
v
ersus
did
not
(blue
curv
e)
recei
v
e
postoperati
v
e
radiotherap
y
.
The
radiotherap
y
group
has
a
noticeably
better
survi
v
al
curv
e,
with
a
median
RFS
of
153.7
months
v
ersus
119.1
months
in
the
no-radiotherap
y
group.
The
shaded
condence
interv
als
for
the
tw
o
groups
remain
lar
gely
separated
after
the
rst
50
months
of
follo
w-up.
The
log-rank
test
yields
p
<
0.005,
indicating
a
statistically
signicant
dif
ference.
F
or
e
xample,
at
100
months
approximately
60%
of
patients
who
recei
v
ed
radiotherap
y
are
still
relapse-free,
whereas
only
about
50%
of
patients
without
radiotherap
y
remain
e
v
ent-free.
This
pattern
implies
a
protecti
v
e
ef
fect
of
radiotherap
y–patients
who
recei
v
ed
radiation
ha
v
e
consistently
lo
wer
hazard
of
relapse.
In
f
act,
by
200
months
the
majority
of
the
radiotherap
y
group
remain
e
v
ent-free,
whereas
the
non-radiotherap
y
group
has
declined
to
a
much
smaller
survi
v
al
fraction.
In
summary
,
the
Kaplan–Meier
analysis
suggests
that
radiotherap
y
is
associated
with
an
appreciable
impro
v
ement
in
RFS
in
this
cohort.
Figure
6
sho
ws
Kaplan–Meier
curv
es
by
histological
tumor
grade
(1,
2,
or
3).
Lo
wer
-grade
tumors
(blue
curv
e,
grade
1)
are
associated
with
the
best
survi
v
al:
median
RFS
is
195.1
months
for
grade
1.
In
contrast,
high-grade
tumors
(green
curv
e,
grade
3)
ha
v
e
a
median
of
on
l
y
115.8
months.
Grade
2
(red)
is
intermediate
at
152.5
months.
The
curv
es
are
well
separated,
and
the
log-rank
test
gi
v
es
p
<
0.005.
Early
on
(up
to
50
months)
o
v
er
90%
of
grade
1
patients
remain
e
v
ent-free
v
ersus
about
80%
of
grade
3.
By
150
months,
only
20%
of
grade
3
patients
are
relapse-free
compared
to
50%
of
grade
1.
The
shaded
condence
re
gions
sho
w
Int
J
Adv
Appl
Sci,
V
ol.
15,
No.
3,
September
2026:
1193–1202
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Adv
Appl
Sci
ISSN:
2252-8814
❒
1199
limited
o
v
erlap
among
the
grades.
These
patterns
reect
the
e
xpected
biology:
higher
-grade
tumors
are
more
aggressi
v
e
and
recur
earlier
.
The
sizable
dif
ferences
in
median
survi
v
al
(nearly
80
months
between
grade
1
and
grade
3)
indicate
that
tumor
grade
is
a
signicant
determinant
of
prognosis
in
this
cohort.
Figure
5.
Kaplan-Meier
curv
es
postoperati
v
e
radiotherap
y
Figure
6.
Kaplan–Meier
curv
es
by
histological
tumor
grade
Mapping
signicant
clinical
factor
s
to
r
elapse-fr
ee
survival
pr
o
gnosis
in
br
east
cancer
...
(Boby
Al
Qurthuby)
Evaluation Warning : The document was created with Spire.PDF for Python.
1200
❒
ISSN:
2252-8814
In
identifying
feature
ranking
based
on
Cox
log-lik
elihood
ratio
test,
each
v
ariable
is
assessed
by
ho
w
much
it
impro
v
es
the
model’
s
t,
with
lar
ger
v
alues
indicating
greater
importance,
as
e
xplicated
in
T
able
1.
Age
at
diagnosis
is
rank
ed
rst
(log-lik
elihood
ratio
(LLR)
130.4,
p
=
3.31
×
10
−
30
,
f
alse
disco
v
ery
rate
(FDR)
1.98
×
10
−
29
),
follo
wed
by
number
of
positi
v
e
lymph
nodes
(102.9,
p
=
3.45
×
10
−
24
,
FDR
1.03
×
10
−
23
)
and
tumor
size
(98.8,
p
=
2.78
×
10
−
23
,
FDR
5.55
×
10
−
23
).
These
three
ha
v
e
e
xtremely
high
signicance–their
FDR-corrected
p-v
alues
remain
<
10
−
22
–
reecting
that
the
y
dominate
the
survi
v
al
model.
T
umor
grade
is
fourth
(LLR-meaning
the
recurrence
of
cancer
within
the
breast
and
surrounding
areas
after
initia
l
treatment
-
22.7,
p
=
1.92
×
10
−
6
,
FDR
≈
2.89
×
10
−
5
),
a
much
smaller
b
ut
still
signicant
contrib
ution.
Radiotherap
y
(LLR
9.71,
p
=
0.002,
FDR
0.002)
and
chemotherap
y
(LLR
4.546,
p
=
0.033,
FDR
0.033)
ha
v
e
t
he
lo
west
scores.
These
correspond
to
weak
er
prognostic
information
in
a
uni
v
ariate
sense.
The
drop
from
LLR
∼
100
for
the
top
features
to
<
10
for
treatments
is
notable:
it
indicates
that
age,
nodal
status,
and
tumor
b
urden
e
xplain
most
of
the
v
ariability
in
outcome.
In
ef
fect,
age
alone
accounts
for
a
130-point
increase
in
log-lik
elihood–an
enormous
ef
fect
(by
comparison,
a
lik
elihood
ratio
of
3.84
corresponds
to
p
=
0.05
for
one
de
gree
of
freedom).
The
chemotherap
y
v
ariable’
s
LLR
of
only
∼
4.55
suggests
a
mar
ginal
uni
v
ariate
ef
fect
(just
under
p
=
0.05).
The
FDR
column
conrms
that
e
v
en
after
correcting
for
multiple
tests,
all
six
f
actors
are
statistically
signicant
(the
lar
gest
FDR
being
0.033
for
chemotherap
y).
Thus,
the
ranking
reinforces
the
Cox
re
gression
and
Kaplan–Meier
results:
patient
age,
lymph
node
in
v
olv
ement,
and
tumor
size
carry
the
greatest
prognostic
weight,
while
tumor
grade
is
moderately
important
and
treatment
v
ariables,
though
statistically
rele
v
ant,
contrib
ute
less
to
prediction.
T
able
1.
Rank
survi
v
al
features
Features
Log-lik
elihood
ratio
test
p
FDR
Age
at
diagnosis
[years]
130.4
3.31x10
−
30
1.98x10
−
29
L
ymph
nodes
e
xamined
positi
v
e
102.9
3.45x10
−
24
1.03x10
−
23
T
umor
size
[mm]
98.8
2.78x10
−
23
5.55x10
−
23
T
umor
grade
22.7
1.92x10
−
6
2.89x10
−
23
Radiotherap
y
9.713
0.002
0.002
Chemotherap
y
4.546
0.033
0.033
4.
CONCLUSION
This
comprehensi
v
e
SA
reinforces
the
critical
role
of
traditional
clinical
f
actors—age,
tumor
size,
lymph
node
in
v
olv
ement,
and
tumor
grade—as
signicant
predictors
of
RFS
in
breast
cancer
patients.
Findings
from
Kaplan–Meier
curv
es
and
Cox
proportional
hazards
models
conrm
and
quantify
their
prognostic
rele
v
ance,
while
rare
histological
subtypes
and
adjuv
ant
radiotherap
y
sho
w
strong
protecti
v
e
associations.
The
observ
ed
correlation
between
chemotherap
y
and
reduced
RFS
highlights
the
impact
of
clinical
selection
bias
and
cautions
ag
ainst
o
v
erinterpretation
of
observ
ational
outcomes.
Clinically
,
these
insights
aid
in
precise
risk
stratication,
enabling
more
t
ailored
treatment
decisions.
Ho
we
v
er
,
this
retrospecti
v
e
analysis
has
inherent
limitations,
such
as
the
absence
of
molecular
biomark
ers
and
genetic
data.
Thus,
while
the
study
pro
vides
rob
ust
v
alidation
of
established
kno
wledge,
future
research
inte
grating
multi-omics
data
and
adv
anced
modeling
techniques
is
essential
to
rene
predicti
v
e
accurac
y
and
unra
v
el
the
comple
xity
of
breast
cancer
prognosis.
A
CKNO
WLEDGMENTS
This
w
ork
has
been
supported
by
the
Master
of
Informat
ics,
F
aculty
of
Articial
Intelligence
and
Data
Science,
Uni
v
ersitas
Pelita
Harapan,
Jakarta,
Indonesia.
FUNDING
INFORMA
TION
Authors
state
no
funding
in
v
olv
ed.
A
UTHOR
CONTRIB
UTIONS
ST
A
TEMENT
This
journal
uses
the
Contrib
utor
Roles
T
axonomy
(CRediT)
to
recognize
indi
vidual
author
contrib
utions,
reduce
authorship
disputes,
and
f
acilitate
collaboration.
Int
J
Adv
Appl
Sci,
V
ol.
15,
No.
3,
September
2026:
1193–1202
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Adv
Appl
Sci
ISSN:
2252-8814
❒
1201
Name
of
A
uthor
C
M
So
V
a
F
o
I
R
D
O
E
V
i
Su
P
Fu
Boby
Al
Qurthuby
✓
✓
✓
✓
✓
✓
✓
I
Made
Murw
antara
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
C
:
C
onceptualization
I
:
I
n
v
estig
ation
V
i
:
V
i
sualization
M
:
M
ethodology
R
:
R
esources
Su
:
Su
pervision
So
:
So
ftw
are
D
:
D
ata
Curation
P
:
P
roject
Administration
V
a
:
V
a
lidation
O
:
Writing
-
O
riginal
Draft
Fu
:
Fu
nding
Acquisition
F
o
:
F
o
rmal
Analysis
E
:
Writing
-
Re
vie
w
&
E
diting
CONFLICT
OF
INTEREST
ST
A
TEMENT
Authors
state
no
conict
of
interest.
INFORMED
CONSENT
This
section
not
applicable.
This
study
did
not
in
v
olv
e
human
participants,
human
data,
or
an
y
personally
identiable
informati
on.
All
data
used
were
either
publicly
a
v
ail
able,
fully
anon
ymized,
or
deri
v
ed
from
non-human
sources,
and
therefore
no
informed
consent
w
as
required
from
indi
viduals.
ETHICAL
APPR
O
V
AL
This
section
applicable.
This
research
did
not
in
v
olv
e
human
subjects,
human
biological
materi
als,
or
e
xperimental
procedures
on
animals.
The
w
ork
w
as
conducted
solely
on
computational
models,
publicly
a
v
ailable
datasets,
or
non-sensiti
v
e
data
that
did
not
re
q
ui
re
interv
ention
with
li
ving
or
g
anisms.
Therefore,
ethical
appro
v
al
from
an
institutional
re
vie
w
board
or
animal
ethics
committee
w
as
not
necessary
for
this
study
.
D
A
T
A
A
V
AILABILITY
The
supporting
data
of
this
study
are
openly
a
v
ailable
in
cBioPortal
at
https://www
.cbioportal.or
g/study/
summary?id=brca
metabric.
REFERENCES
[1]
P
.
K
e
and
C.
Li,
“Space-time
dynamic
v
ariation
trend
of
female
breast
cancer
incidence
and
mortality:
a
global
perspecti
v
e
(1990–2021),
”
Cancer
Epidemiolo
gy
,
v
ol.
98,
Oct.
2025,
doi:
10.1016/j.canep.2025.102898.
[2]
P
.
Sun
et
al
.,
“Global,
re
gional
,
and
national
b
urden
of
female
cancers
in
w
omen
of
child-bearing
age,
1990–2021:
analysis
of
data
from
the
global
b
urden
of
disease
study
2021,
”
eClinicalMedicine
,
v
ol.
74,
Aug.
2024,
doi:
10.1016/j.eclinm.2024.102713.
[3]
J.
P
.
A.
-Granados
and
L.
F
.
Ni
˜
no,
“Prediction
of
o
v
erall
and
relapse-free
survi
v
al
in
triple-ne
g
ati
v
e
breast
cancer
patients
through
machine
learning-based
clustering
on
clinical
data,
”
Clinical
Br
east
Cancer
,
v
ol.
25,
no.
7,
pp.
714–719,
Oct.
2025,
doi:
10.1016/j.clbc.2025.07.027.
[4]
M.
Ka
w
ase
et
al
.,
“Impact
of
D3
lymph
node
dissection
on
short-term
and
long-term
outcomes
in
elderly
patients
with
colon
cancer
,
”
T
ec
hniques
in
Colopr
octolo
gy
,
v
ol.
29,
no.
1,
Dec.
2025,
doi:
10.1007/s10151-025-03149-9.
[5]
L.
F
allo
weld,
B.
E.
Kiely
,
L.
T
ra
v
ado,
R.
V
entura,
J.
Dullehan,
and
F
.
Cardoso,
“Best
practices
for
communication
of
prognosis
and
end-of-life
discussions
with
patients
with
adv
anced
breast
cancer:
a
report
from
the
adv
anced
breast
cancer
global
alliance,
”
The
Br
east
,
v
ol.
84,
Dec.
2025,
doi:
10.1016/j.breast.2025.104595.
[6]
O.
Y
ersal,
“Biological
subtypes
of
breast
cancer:
prognostic
and
therapeutic
implications,
”
W
orld
J
ournal
of
Clinical
Oncolo
gy
,
v
ol.
5,
no.
3,
2014,
doi:
10.5306/wjco.v5.i3.412.
[7]
E.
M.
Y
ousef
et
al
.,
“Incidence
and
prognostic
signicance
of
hormonal
receptors
and
HER2
status
con
v
ersion
in
recurrent
breast
cancer:
a
retrospecti
v
e
study
in
a
single
institute,
”
Medicina
,
v
ol.
61,
no.
4,
Mar
.
2025,
doi:
10.3390/medicina61040563.
[8]
E.
Nolan,
G.
J.
Lindeman,
and
J.
E.
V
isv
ader
,
“Deciphering
breast
cancer:
from
biology
to
the
clinic,
”
Cell
,
v
ol.
186,
no.
8,
pp.
1708–1728,
Apr
.
2023,
doi:
10.1016/j.cell.2023.01.040.
[9]
Y
.
Chen,
Z.
Xu,
Y
.
Chen,
Y
.
Dai,
and
J.
Ding,
“Comparison
of
the
prognosis
of
medullary
breast
carcinoma
and
in
v
asi
v
e
ductal
carcinoma:
a
SEER-based
study
,
”
T
r
anslational
Cancer
Resear
c
h
,
v
ol.
13,
no.
1,
pp.
231–248,
Jan.
2024,
doi:
10.21037/tcr
-23-858.
[10]
D.
Jiao
et
al
.,
“Impact
of
hormone
receptor
status
and
HER2
e
xpression
on
neoadjuv
ant
tar
geted
therap
y
response
in
HER2-positi
v
e
breast
cancer:
a
multicenter
retrospecti
v
e
study
,
”
Modern
P
atholo
gy
,
v
ol.
38,
no.
8,
Aug.
2025,
doi:
10.1016/j.modpat.2025.100778.
[11]
G.
Gaggi
et
al
.,
“Interference
of
estrogen
signaling
by
endocrine
disruptors
in
male
and
female
cells:
potential
implications
of
BPS
and
PFOS
in
human
de
v
elopment,
”
Eur
opean
J
ournal
of
Cell
Biolo
gy
,
v
ol.
104,
no.
3,
Sep.
2025,
doi:
10.1016/j.ejcb
.2025.151509.
[12]
C.
Niu,
R.
Zhang,
C.
Zhang,
C.
Liu,
and
Z.
W
u,
“Progesterone
attenuates
depression-lik
e
beha
viors
in
chr
onic
unpredictable
stress
via
suppression
of
neuroinammation,
”
Behaviour
al
Br
ain
Resear
c
h
,
v
ol.
497,
Feb
.
2026,
doi:
10.1016/j.bbr
.2025.115879.
Mapping
signicant
clinical
factor
s
to
r
elapse-fr
ee
survival
pr
o
gnosis
in
br
east
cancer
...
(Boby
Al
Qurthuby)
Evaluation Warning : The document was created with Spire.PDF for Python.
1202
❒
ISSN:
2252-8814
[13]
A.
Lindholm
et
al.
,
“Pro-inammatory
c
ytokines
increase
temporarily
after
adjuv
ant
treatment
for
breast
cancer
in
postmenopausal
w
omen:
a
longitudinal
study
,
”
Br
east
Cancer
Resear
c
h
,
v
ol.
26,
no.
1,
Oct.
2024,
doi:
10.1186/s13058-024-01898-3.
[14]
W
.
Cui,
M.
F
an,
and
L.
Li,
“Morphological
analysis
of
tumor
microen
vironment
in
HER2-positi
v
e
breast
cancer:
predicting
response
to
neoadjuv
ant
chemotherap
y
on
histopathological
images,
”
Br
east
Cancer
Resear
c
h
,
v
ol.
27,
no.
1,
Oct.
2025,
doi:
10.1186/s13058-025-02139-x.
[15]
S
.
Lighto
wlers,
S.
Raby
,
and
R.
Chuter
,
“
A
national
frame
w
ork
for
mo
ving
to
w
ards
more
en
vironmentally
sustainable
radiotherap
y:
the
green
radiotherap
y
frame
w
ork,
”
Radiother
apy
and
Oncolo
gy
,
v
ol.
212,
No
v
.
2025,
doi:
10.1016/j.radonc.2025.111114.
[16]
F
.
Girardi
et
al
.,
“Omitting
anthrac
yclines
for
the
adjuv
ant
treatment
of
patients
with
triple-ne
g
ati
v
e
breast
cancer:
a
non-inferiority
meta-analysis,
”
The
Br
east
,
v
ol.
83,
Oct.
2025,
doi:
10.1016/j.breast.2025.104524.
[17]
D
.
V
aliya
v
eettil,
D.
Joseph,
and
M.
Malik,
“Cardiotoxicity
in
breast
cancer
treatment:
causes
and
mitig
ation,
”
Cancer
T
r
eatment
and
Resear
c
h
Communications
,
v
ol.
37,
2023,
doi:
10.1016/j.ctarc.2023.100760.
[18]
J.
Greene
and
B.
Hennessy
,
“The
role
of
anthrac
yclines
in
the
treatment
of
early
breast
cancer
,
”
J
ournal
of
Oncolo
gy
Pharmacy
Pr
actice
,
v
ol.
21,
no.
3,
pp.
201–212,
Jun.
2015,
doi:
10.1177/1078155214531513.
[19]
S.
T
.
-Bayona
et
al
.,
“Radiotherap
y
v
ersus
observ
ation
after
sur
gical
resection
of
atypical
meningiomas,
”
Inter
disciplinary
Neur
osur
g
ery
,
v
ol.
25,
Sep.
2021,
doi:
10.1016/j.inat.2021.101201.
[20]
C.
Curtis
et
al
.,
“The
genomic
and
transcri
ptomic
architecture
of
2,000
breast
tumours
re
v
eals
no
v
el
subgroups,
”
Natur
e
,
v
ol.
486,
pp.
346–352,
Jun.
2012,
doi:
10.1038/nature10983.
[21]
B.
Pereira
et
al
.,
“The
somatic
mutation
proles
of
2,433
breast
cancers
rene
their
genomic
and
transcriptomic
landscapes,
”
Natur
e
Communications
,
v
ol.
7,
no.
1,
May
2016,
doi:
10.1038/ncomms11479.
[22]
O.
M.
Rueda
et
al
.,
“Dynamics
of
breast-cancer
relapse
re
v
eal
late-recurring
ER-positi
v
e
genomic
subgroups,
”
Natur
e
,
v
ol.
567,
pp.
399–404,
Mar
.
2019,
doi:
10.1038/s41586-019-1007-8.
[23]
M.
I.
Higgins
et
al
.,
“Molecular
classication
of
nonurothelial
histologic
subtypes
of
bladder
cancer
,
”
Ur
olo
gic
Oncolo
gy:
Seminar
s
and
Original
In
vestigations
,
v
ol.
43,
no.
11,
pp.
599–608,
No
v
.
2025,
doi:
10.1016/j.urolonc.2025.07.025
[24]
T
.
De
y
,
A.
Mukherjee,
and
S.
Chakraborty
,
“
A
practical
o
v
ervie
w
and
reporting
strate
gies
for
statisti
cal
analysis
of
survi
v
al
studies,
”
Chest
,
v
ol.
158,
no.
1,
pp.
S39–S48,
Jul.
2020,
doi:
10.1016/j.chest.2020.03.015.
[25]
V
.
Poon
and
D.
Lu,
“Performance
of
Cox
proportional
hazard
models
on
reco
v
ering
the
ground
truth
of
confounded
e
xposure–response
relationships
for
lar
ge-molecule
oncology
drugs,
”
CPT
:
Pharmacometrics
&
Systems
Pharmacolo
gy
,
v
ol.
11,
no.
11,
pp.
1511–1526,
No
v
.
2022,
doi:
10.1002/psp4.12859.
BIOGRAPHIES
OF
A
UTHORS
Boby
Al
Qurthuby
is
a
graduat
e
Informatics
student,
Uni
v
ersitas
Pel
ita
Harapan,
Indonesia.
His
skilled
in
fullstack
de
v
elopment,
data
science,
and
cloud
computing
system.
Strong
education
professional
with
a
bachelor’
s
de
gree
focused
in
Math,
from
Uni
v
ersitas
Udayana,
Indonesia.
He
can
be
contacted
at
email:
01679230016@student.uph.edu.
I
Made
Murwantara
earned
his
Ph.D.
de
gree
in
Computer
Science
from
the
Uni
v
ersity
of
Birmingham,
United
Kingdom
in
2016.
He
currently
teaches
cloud
computing,
adv
anced
softw
are
engineering,
fr
ontier
technology
,
softw
are
engine
ering
application
design
studio,
and
operations
research
at
Uni
v
ersitas
Pelita
Harapan,
Indonesia.
His
research
interest
includes
softw
are
product
line
engineering
for
and
in
cloud
computing,
AI
for
biomedical
analysis,
AI-based
softw
are
automation,
and
cloud
computing
ener
gy
management.
He
can
be
contacted
at
email:
made.murw
antara@uph.edu.
Int
J
Adv
Appl
Sci,
V
ol.
15,
No.
3,
September
2026:
1193–1202
Evaluation Warning : The document was created with Spire.PDF for Python.