Introduction
Population-based colorectal cancer (CRC) screening has led to a stage shift in the
screen-detected CRCs, characterized by a higher proportion of stage I and a lower
proportion of stage IV CRCs compared with non-screen-detected CRCs [1]. Whether this more favorable stage distribution is also observed in screen-detected
submucosal invasive (T1) CRCs specifically is unknown. Given the increasing use of
organ-preserving treatment strategies for T1 CRC, it is essential to determine whether
predictors of metastasis and oncologic outcomes in screen-detected T1 CRCs are comparable
with those in non-screen-detected cases. Especially as most risk prediction models
used to date are based on data from non-screen-detected T1 CRCs from the prescreening
period.
Based on the significant reduction in advanced-stage CRC incidence after the implementation
of CRC screening programs, it might be assumed that stage migration owing to earlier
detection also extends to screen-detected T1 CRCs, and that screen-detected T1 CRCs
are detected at an earlier stage, before metastasis occurs. However, it cannot be
definitely established that T-stage migration owing to screening consistently coincides
with migration in terms of lymph node involvement (N stage) and distant metastasis
(M stage), as recent genetic insights have suggested that the development of lymph
node and distant metastases may be already determined in the earliest phases of cancer
development [2].
Up until now, only a limited number of small-scale studies have examined the risk
of lymph node metastasis (LNM) or CRC recurrence specifically in screen-detected T1
CRC [3]
[4]
[5]
[6], showing a lower or comparable risk of LNM and CRC recurrence compared with T1 CRCs
from the pre-screening era. However, these studies either had a short follow-up, excluded
primary surgical resections, or had small sample sizes with few events. Furthermore,
none of the studies directly compared oncologic outcomes between screen-detected and
non-screen-detected T1 CRC patients.
Therefore, the objectives of the current study were to evaluate the stage distribution
and oncologic outcomes of screen-detected and non-screen-detected T1 CRC patients,
and to study whether the established risk factors for LNM, which are used to guide
clinical decision-making, can be extrapolated to the screening population of T1 CRC
patients.
Methods
Study design and selection of patients
A multicenter retrospective observational cohort study was conducted. All consecutive
patients diagnosed with T1 CRC between 2014 and 2017 from 12 hospitals (including
one academic hospital) were identified through the Netherlands Cancer Registry. The
Dutch National screening program started in 2014, with a stepwise introduction by
age cohorts, until all eligible age cohorts had been invited at least once in 2019.
Men and women aged 55–75 years are invited once every 2 years to send in stool samples
for detection of occult blood using a fecal immunochemical test (FIT). During our
study period, the invitation coverage was 40% of the target population in 2014, rising
to 95% in 2017 [1].
The reason for index colonoscopy (i.e. population-based CRC screening program or other
reason) was retrieved from the individual electronic medical records, which were reviewed
at the participating hospitals. Patients were included when the local pathology report
confirmed the diagnosis of pT1 CRC. Exclusion criteria were: (i) hereditary predisposition
to CRC (Lynch or familial adenomatous polyposis syndrome); (ii) inflammatory bowel
disease; (iii) synchronous CRC at diagnosis or previous CRC ≤5 years before the diagnosis
of the T1 CRC; (iv) non-adenocarcinoma; (v) missing pathology or endoscopy reports;
and (vi) neoadjuvant (chemo-)radiotherapy. Further details on the data collection
and definitions of study parameters can be found in Appendix 1s, see online-only Supplementary material.
This study was exempt from institutional board review by the Medical Ethical Review
Committee of the University Medical Center Utrecht (reference number 15–487/C) and
was carried out in accordance with the Helsinki Declaration. The study confirmed to
the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology)
guideline for cohort studies.
Determinant and end points
The determinant of interest was the indication for colonoscopy (i.e. screen detected
or non-screen detected). Screen-detected T1 CRC was defined as T1 CRC detected through
the population-based screening program after a positive FIT. All clinically detected
T1 CRCs outside the screening program were considered to be non-screen detected.
The primary outcome was the proportion of synchronous LNM. This was defined as tumor-positive
lymph nodes in the surgical resection specimen at the time of diagnosis. In addition,
we studied whether the performance of conventional clinical and histopathologic risk
factors for LNM was similar in screen-detected and non-screen-detected patients.
Secondary outcomes were 3- and 5-year time to recurrence (TTR), metastasis-free survival
(MFS), colorectal cancer-specific survival (CSS), and overall survival. TTR was defined
as the time from diagnosis until the event of CRC recurrence (locoregional and/or
distant recurrence) or CRC-related death. MFS was defined as the time from diagnosis
until the event of distant recurrence (i.e. outside the area intended to be removed
with salvage surgery) or CRC-related death. CSS was defined as the time from diagnosis
until the event of CRC-related or CRC treatment-related death. Overall survival was
defined as the time from diagnosis until the event of death due to any cause.
Data analysis
Baseline characteristics were compared between screen-detected and non-screen-detected
T1 CRC patients using the chi-squared test for categorical variables and the Mann–Whitney
U test for continuous data. Categorical data were expressed as frequencies with percentages,
and continuous data as the median with corresponding interquartile range (IQR).
We evaluated the potential association between the indication (screen detected or
non-screen detected) and LNM with univariable and multivariable logistic regression
analyses in the group of surgically treated patients (both primary and completion
surgery). In the multivariable model, we adjusted for age, sex, polyp location, polyp
morphology, polyp size, differentiation grade, lymphovascular invasion (LVI), and
the number of lymph nodes retrieved during surgery. These factors were chosen as potential
confounders based on the available literature. As a sensitivity analysis, we also
studied the association between the indication for colonoscopy and LNM restricted
to patients aged 55–80 years, which is the age range of the screening population.
The upper age limit of 80 was chosen to take into account screen-detected patients
aged >75 years at diagnosis owing to delays in screening invitation, returning the
FIT, or CRC treatment.
To evaluate whether established conventional clinical and histologic risk factors
for LNM are also predictive in screen-detected T1 CRC patients, we analyzed the performance
of a logistic regression model based on these risk factors in the screen-detected
and non-screen-detected patients separately. The risk factors included were age, location,
morphology, differentiation grade, and LVI [7]
[8]
[9]
[10]. The main performance measures were the calibration and the discriminative ability
of the model. Calibration was assessed graphically using smoothed calibration curves.
Discriminative ability was quantified by the area under the receiver operating curve
(AUROC). The discriminative performance of the model (AUROC) was compared between
both groups using a Z test. Furthermore, a closed testing procedure was performed to assess whether the
model, which was developed in the total cohort, needed updating in the screen-detected
or non-screen-detected groups, or in both (e.g. re-estimation of the intercept and/or
slope, or re-estimation of all coefficients).
The TTR, MFS, CSS, and overall survival of screen-detected and non-screen-detected
patients were described using the Kaplan–Meier method. A Cox proportional hazard regression
model was used to study the association between the indication for colonoscopy (i.e.
screen detected or non-screen detected) and oncologic and survival outcomes, adjusted
for potential confounders. For TTR, CSS, and MFS, we corrected for polyp location,
treatment strategy, differentiation grade, LVI, and resection margin status. For overall
survival, we adjusted for age at diagnosis, sex, and American Society of Anesthesiologists
(ASA) classification. Again, we performed a sensitivity analysis in which we included
only patients aged 55–80 years. Additionally, we repeated all analyses in a subgroup
of patients who were treated first with endoscopic resection.
Because of missing data for several clinicopathologic variables ([Table 1]), we performed multiple imputation before data analysis (using multivariate imputation
by chained equations with 22 variables, 10 imputation data sets, and 21 iterations).
Rubin’s rules were used to pool results across imputed datasets. All statistical analyses
were performed using R version 4.0.3 (RStudio, Inc., Boston, Massachusetts, USA).
A two-sided P value ≤0.05 was considered significant.
Table 1 Clinical and histopathologic characteristics of the 1803 included T1 colorectal cancer
(CRC) patients.
|
|
Baseline characteristics
|
P value*
|
|
Total cohort (n = 1803)
|
T1 CRC patients
|
|
Screen detected (n = 1114)
|
Non-screen detected (n = 689)
|
|
ASA, American Society of Anesthesiology; IQR, interquartile range.
*
P value is derived from descriptive statistics of imputed data.
† The percentage of cases with missing data were: ASA classification, 4.4%; morphology,
1.9%; polyp size, 10.0%; differentiation, 1.1%; lymphovascular invasion, 3.3%; resection
margin, 2.0%. All data were available in the remaining categories. Missing data were
imputed before analyses were performed.
‡ Only patients with an initial local resection.
|
|
Age, median (IQR), years
|
69 (11)
|
67 (10)
|
71 (15)
|
<0.001
|
|
Sex, male, n (%)
|
1124 (62.3)
|
726 (65.2)
|
398 (57.7)
|
0.002
|
|
ASA classification, n (%)†
|
<0.001
|
|
|
424 (24.6)
|
277 (25.7)
|
147 (22.8)
|
|
|
1067 (61.9)
|
702 (65.1)
|
365 (56.6)
|
|
|
233 (13.5)
|
100 (9.3)
|
133 (20.6)
|
|
|
79
|
35
|
44
|
|
Location, n (%)
|
<0.001
|
|
|
321 (17.8)
|
181 (16.2)
|
140 (20.3)
|
|
|
1058 (58.7)
|
698 (62.7)
|
360 (52.2)
|
|
|
424 (23.5)
|
235 (21.1)
|
189 (27.4)
|
|
Morphology, n (%)†
|
0.99
|
|
|
1164 (65.8)
|
724 (65.9)
|
440 (65.7)
|
|
|
604 (34.2)
|
374 (34.1)
|
230 (34.3)
|
|
|
35
|
16
|
19
|
|
Polyp size, median (IQR)†
|
<0.001
|
|
|
20 (15)
|
20 (17)
|
20 (15)
|
|
|
181
|
83
|
98
|
|
Differentiation, n (%)†
|
0.99
|
|
|
1693 (94.9)
|
1044 (94.8)
|
639 (95.0)
|
|
|
91 (5.1)
|
57 (5.2)
|
34 (5.0)
|
|
|
19
|
13
|
6
|
|
Lymphovascular invasion, n (%)†
|
0.72
|
|
|
1470 (84.3)
|
905 (83.8)
|
565 (85.2)
|
|
|
273 (15.7)
|
175 (16.2)
|
98 (14.8)
|
|
|
60
|
34
|
26
|
|
Resection margin, n (%)† ‡
|
0.54
|
|
|
787 (63.9)
|
497 (62.9)
|
290 (65.6)
|
|
|
187 (15.2)
|
126 (15.9)
|
61 (13.8)
|
|
|
258 (20.9)
|
167 (21.1)
|
91 (20.6)
|
|
|
36
|
26
|
10
|
|
Treatment, n (%)
|
<0.001
|
|
|
847 (47.0)
|
529 (47.5)
|
318 (46.2)
|
|
|
535 (29.7)
|
298 (26.8)
|
237 (34.4)
|
|
|
421 (23.3)
|
287 (25.7)
|
134 (19.4)
|
Results
Patient characteristics and treatment strategies
A total of 2258 T1 CRC patients were identified in the participating hospitals, of
whom 455 were excluded for meeting one or more of the exclusion criteria ([Fig. 1]). Of the remaining 1803 patients, 1114 (61.8%) were diagnosed through population-based
screening and 689 (38.2%) were non-screen detected. In the non-screen-detected group,
the indications for the colonoscopy where the T1 CRC was detected were as follows:
symptoms, 579/689 (84.0%); incidental finding on imaging, 54/689 (7.8%); surveillance
for history of adenomatous polyp(s), 37/689 (5.4%); surveillance for history of CRC
(>5 years prior to diagnosis of T1 CRC), 10/689 (1.5%); family history without genetic
predisposition, 7/689 (1.0%); reason unknown, 2/689 (0.3%). Screen-detected patients
were younger (67 vs. 71 years; P < 0.001), more often male (65.2% vs. 57.7%; P = 0.002), and had fewer co-morbidities (ASA III/IV 9.3% vs. 20.6%; P < 0.001) ([Table 1]). Left-sided T1 CRCs were more likely to be found in screen-detected patients, while
the percentages of right-sided and rectal T1 CRCs were higher in the non-screen-detected
patients (P < 0.001). No differences were observed regarding morphology and histopathologic features.
Fig. 1 Flowchart of included T1 colorectal cancer (CRC) patients.
While the overall surgical resection rate was comparable between the two groups (52.5%
and 53.8% in the screen-detected and non-screen-detected groups, respectively), primary
surgery was more often performed in non-screen-detected patients (26.8% vs. 34.4%;
P < 0.001). Of the 1268 patients who were treated with endoscopic resection first,
397/816 screen-detected T1 CRCs (48.7%) exhibited one or more histopathologic high
risk features, compared with 223/452 non-screen-detected T1 CRCs (49.3%; P = 0.97). The treatment strategy (completion surgery or surveillance) of these patients
with endoscopically resected high risk T1 CRCs did not significantly differ between
the screen-detected and non-screen-detected patients (P = 0.09) ([Fig. 2]).
Fig. 2 Treatment strategies and oncologic outcomes within the subgroup of 1268 endoscopically
resected screen-detected and non-screen-detected T1 colorectal cancers (CRCs), subdivided
according to histopathologic risk group.
* High risk was defined as having one or more of the following histopathologic risk
factors: lymphovascular invasion, poor differentiation, positive resection margin
(R1) or non-assessable resection margin (Rx). In the absence of all of these risk
factors, a T1 CRC was classified as low risk. When all known risk factors were absent,
but one or more risk factors were missing, a T1 CRC was classified as being of undetermined
risk.
LNM in surgically treated patients
LNM was detected in 107/956 surgically treated patients (11.2%) (Table 1s). LNM was more often observed in screen-detected patients compared with non-screen-detected
patients (12.6% vs. 8.9%), although the difference in LNM was not significant in univariable
analysis (unadjusted odds ratio [OR] 1.49, 95%CI 0.96–2.30). After adjustment for
clinical and histopathologic characteristics, detection through screening was not
an independent risk factor for LNM (OR 1.41, 95%CI 0.89–2.23). The findings of the
sensitivity analysis, which was restricted to patients diagnosed between the ages
of 55 and 80, also did not show a significant difference in the risk of LNM between
screen-detected and non-screen-detected patients (OR 1.52, 95%CI 0.91–2.53).
Prediction of LNM
The performance of a model based on a combination of conventional clinical and histologic
risk factors for the presence of LNM (age, location, pedunculated vs. nonpedunculated
morphology, grade of differentiation, and LVI) was tested in both screen-detected
and non-screen-detected surgically treated patients. Calibration curves showed that
the calibration of the model was better in the screen-detected T1 CRC group (Fig. 1s). The AUROCs were not significantly different between screen-detected (0.66, 95%CI
0.59–0.73) and non-screen-detected T1 CRCs (0.60, 95%CI 0.48–0.71; P = 0.37), indicating no significant difference in the discriminative ability of the
model between the two groups (Fig. 2s). Finally, a closed testing procedure showed that there was no statistical evidence
that the model, which was developed in the total cohort, needed updating in either
screen-detected or non-screen-detected patients.
TTR, MFS, CSS, and overall survival
After a median follow-up of 51 months (IQR 30 months), CRC recurrence occurred in
38/1114 screen-detected patients (3.4%) and 24/689 non-screen-detected patients (3.5%)
(Table 2s). Of the screen-detected patients, 19 (1.7%) died of a CRC-related cause: cancer
recurrence during follow-up (n = 13); distant metastasis at the time of diagnosis
(n = 3); CRC treatment-related cause (n = 3). In the non-screen-detected group, CRC-related
death was observed in 11 patients (1.6%), of whom eight died after cancer recurrence
and three owing to a CRC treatment-related cause. The estimated 3- and 5-year TTRs
among screen-detected T1 CRC patients were similar to those of non-screen-detected
patients (adjusted hazard ratio [HR] 1.05, 95%CI 0.62–1.79) ([Fig. 3]
a; [Table 2] and [Table 3]).
Fig. 3 Outcomes of screen-detected and non-screen-detected T1 colorectal cancer (CRC) patients
in terms of: a time to recurrence; b metastasis-free survival; c cancer-specific survival; d overall survival.
Table 2 Oncologic and survival outcomes of screen-detected and non-screen-detected T1 colorectal
cancer (CRC) patients after 3 and 5 years.
|
|
Screen detected (n = 1114)
|
Non-screen detected (n = 689)
|
Hazard ratio (95%CI)b
|
P value†
|
|
|
Event-free* at:
|
Event-free* at:
|
|
|
3 years, % (95%CI)
|
5 years, % (95%CI)
|
3 years, % (95%CI)a
|
5 years, % (95%CI)
|
|
* Kaplan–Meier method estimates of the percentage of patients without the event of
interest at 3 and 5 years, with corresponding 95%CI.
† Hazard ratio and P value from multivariable Cox proportional hazards regression model, adjusted for polyp
location, treatment strategy, differentiation grade, lymphovascular invasion, and
resection margin (for TTR, MFS, and CSS) or age at diagnosis, sex and American Society
of Anesthesiology classification (for overall survival).
|
|
Time to recurrence (TTS)
|
98.4 (97.6–99.2)
|
95.4 (93.7–97.0)
|
98.0 (96.8–99.1)
|
95.3 (93.3–97.4)
|
1.05 (0.62–1.78)
|
0.85
|
|
Metastasis-free survival (MFS)
|
98.7 (98.0–99.4)
|
95.8 (94.2–97.4)
|
98.8 (97.9–99.7)
|
96.7 (94.9–98.5)
|
1.41 (0.75–2.65)
|
0.28
|
|
Cancer-specific survival (CSS)
|
99.3 (98.7–99.8)
|
98.0 (96.9–99.1)
|
99.0 (98.2–99.8)
|
97.6 (96.1–99.0)
|
1.05 (0.47–2.34)
|
0.91
|
|
Overall survival
|
97.0 (95.9–98.1)
|
91.9 (89.9–94.0)
|
87.4 (84.8–90.0)
|
75.3 (71.4–79.3)
|
0.51 (0.38–0.68)
|
<0.001
|
Table 3 Multivariable Cox proportional hazard regression analyses for comparison of TTR, MFS,
CSS, and overall survival of screen-detected vs. non-screen-detected T1 colorectal
cancers.
|
Covariate, per analysis
|
Hazard ratio (95%CI)
|
|
ASA, American Society of Anesthesiology.
|
|
Time to recurrence (TTR)
|
|
Detection method (reference: non-screen detected)
|
|
|
1.05 (0.62–1.79)
|
|
Location (reference: right-sided colon)
|
|
|
0.80 (0.35–1.82)
|
|
|
2.09 (0.90–4.79)
|
|
Treatment (reference: endoscopic only)
|
|
|
1.03 (0.52–2.04)
|
|
|
0.37 (0.17–0.80)
|
|
Differentiation (reference: well/moderate)
|
|
|
1.20 (0.42–3.45)
|
|
Lymphovascular invasion (reference: absent)
|
|
|
2.50 (1.35–4.63)
|
|
Resection margin (reference: R0)
|
|
|
2.12 (1.13–3.94)
|
|
Metastasis-free survival (MFS)
|
|
Detection method (reference: non-screen-detected)
|
|
|
1.41 (0.75–2.65)
|
|
Location (reference: right-sided colon)
|
|
|
1.00 (0.37–2.68)
|
|
|
2.70 (0.99–7.38)
|
|
Treatment (reference: endoscopic only)
|
|
|
1.29 (0.58–2.86)
|
|
|
0.53 (0.23–1.24)
|
|
Differentiation (reference: well/moderate)
|
|
|
1.50 (0.51–4.43)
|
|
Lymphovascular invasion (reference: absent)
|
|
|
2.67 (1.35–5.25)
|
|
Resection margin (reference: R0)
|
|
|
2.19 (1.07–4.49)
|
|
Cancer-specific survival (CSS)
|
|
Detection method (reference: non-screen-detected)
|
|
|
1.05 (0.47–2.34)
|
|
Location (reference: right-sided colon)
|
|
|
1.03 (0.35–3.02)
|
|
|
1.55 (0.47–5.14)
|
|
Treatment (reference: endoscopic only)
|
|
|
2.87 (0.99–8.29)
|
|
|
0.49 (0.14–1.67)
|
|
Differentiation (reference: well/moderate)
|
|
|
2.05 (0.64–6.62)
|
|
Lymphovascular invasion (reference: absent)
|
|
|
3.60 (1.51–8.55)
|
|
Resection margin (reference: R0)
|
|
|
2.62 (0.85–8.06)
|
|
Overall survival
|
|
Detection method (reference: non-screen-detected)
|
|
|
0.51 (0.38–0.68)
|
|
Age at diagnosis
|
1.07 (1.05–1.09)
|
|
Sex (reference: male)
|
|
|
0.83 (0.63–1.10)
|
|
ASA classification (reference: I)
|
|
|
2.86 (1.57– 5.24)
|
|
|
5.87 (3.10– 11.10)
|
Distant metastases were found in 32/38 patients with cancer recurrence in the screen-detected
group (84.2%), compared with 15/24 patients in the non-screen-detected group (62.5%).
MFS at 3 and 5 years did not differ significantly between the two groups (adjusted
HR 1.41, 95%CI 0.75–2.65) ([Fig. 3]
b). The estimated 3- and 5-year CSSs were high in both groups and were not significantly
different between screen-detected and non-screen-detected patients (adjusted HR 1.05,
95%CI 0.47–2.34) ([Fig. 3]
c). Further analyses of oncologic outcomes for the three treatment strategies separately
did not show a difference in cancer recurrence or CRC-related death between screen-detected
and non-screen-detected T1 CRCs for any of the strategies either (Table 3s).
As opposed to the cancer-specific outcomes, overall survival was significantly higher
in the screen-detected T1 CRC population, with overall survival rates of 97.0% and
91.9% at 3 and 5 years, compared with 87.4% and 75.3% among non-screen-detected T1
CRC patients, even when adjusted for age at diagnosis, sex, and ASA classification
(adjusted HR 0.51, 95%CI 0.38–0.68) ([Fig. 3]d).
In the sensitivity analyses, which we restricted to patients aged 55–80 years, still
no differences were found in TTR, MFS, and CSS between screen-detected and non-screen-detected
T1 CRCs (TTR: adjusted HR 1.08, 95%CI 0.60–1.94; MFS: adjusted HR 1.36, 95%CI 0.69–2.71;
CSS: adjusted HR 1.35, 95%CI 0.52–3.49). Overall survival remained significantly higher
in the screen-detected T1 CRC patients (3-year overall survival: screen detected 97.0%
vs. non-screen detected 88.7%; 5-year overall survival: screen detected 91.9% vs.
non-screen detected 79.4%; adjusted HR 0.53, 95%CI 0.39–0.72) (Fig. 3s).
Within the subgroup of 1268 patients who were treated with endoscopic resection first,
oncologic outcomes also did not differ between the screen-detected population and
non-screen-detected patients (TTR: adjusted HR 1.02, 95%CI 0.54–1.91; MFS: adjusted
HR 1.35, 95%CI 0.63–2.93; CSS: adjusted HR 0.79, 95%CI 0.23–2.69) ([Fig. 2]). In addition, in this subgroup, overall survival was significantly higher in the
screen-detected patients (adjusted HR 0.52, 95%CI 0.37–0.74).
Discussion
Within this study, we showed that T1 CRCs detected through population-based screening
have comparable rates of LNM, and 3- and 5-year TTR, MFS, and CSS compared with non-screen-detected
T1 CRCs. In addition, no significant difference in the performance of a prediction
model for LNM based on conventional clinical and histologic risk factors was observed
between the two groups (AUROC 0.66 vs. 0.60; P = 0.47). These findings suggest that the current risk stratification models, based
on non-screen-detected T1 CRCs, can be applied to screen-detected T1 CRCs as well.
Consequently, there is no need for differential treatment approaches between screen-detected
T1 CRCs and non-screen-detected T1 CRCs. However, screen-detected T1 CRC patients
had a significantly higher overall survival owing to a lower risk of mortality unrelated
to CRC (adjusted HR 0.51, 95%CI 0.38–0.68).
The findings of the present study did not provide support for our hypothesis that
there may be disparities in stage distribution and oncologic outcomes between T1 CRCs
detected through population-based screening and outside of the screening program.
The proportion of LNM did not significantly differ between the two groups, and the
observed rates of LNM were consistent with those reported in the literature prior
to the implementation of screening programs [11]
[12]. Distant metastasis at time of diagnosis (stage IV disease) was rare in this cohort,
both for screen-detected (0.7%) and non-screen-detected T1 CRCs (0.1%). Therefore,
although population-based screening has led to an increased rate of detection of T1
CRCs, the stage distribution of this specific group of patients has not changed. The
same holds true for oncologic outcomes during the follow-up period. Within the total
cohort, 2.6% of T1 CRCs showed distant metastasis during follow-up, which is in line
with previous reports on non-screen-detected T1 CRCs [13]
[14]
[15]. Moreover, CSS rates were high for both screen-detected and non-screen-detected
T1 CRC patients.
In contrast to the similar oncologic outcomes across the two groups, a significant
difference was observed in overall survival, even after adjusting for age at diagnosis,
sex, and ASA classification. This difference was mainly explained by a difference
in mortality unrelated to CRC. A similar observation was made by de Neree tot Babberich
et al. [16], who observed a significantly lower risk of surgical and non-surgical complications
in Dutch screen-detected CRC patients, even after extensive case-mix correction for
clinical, surgical, and histopathologic factors. Both the lower non-CRC-related mortality
and the lower risk of (non-)surgical complications may be explained by a healthy user
bias in the screen-detected population [17]. Factors such as socioeconomic status, ethnicity, educational level, body mass index,
and smoking habit were not registered, while there is evidence that these factors
vary between screening participants and non-participants [18]
[19]
[20]. Therefore, unmeasured confounding in the overall survival analysis cannot be ruled
out. Furthermore, patients with a non-screen-detected T1 CRC had an indication to
undergo a colonoscopy because of either symptoms, underlying illness, or risk profile,
which may have caused a selection bias toward an unhealthier population.
Some limitations need to be taken into account. First, we studied the screen-detected
population and the non-screen-detected population in a period during which the screening
program was implemented in a stepwise manner by age cohort. This influenced the group
composition of screen-detected and non-screen-detected T1 CRC patients in our study;
however, we did not find a significant difference in the proportions of LNM between
the age cohorts that had not yet received an invitation and those who had been invited
to take part in the screening program (data not shown). In addition, preliminary results
from the Dutch CRC screening program show that, after two subsequent screening rounds,
the stage distribution of screen-detected CRCs remained unchanged [21]. Therefore, we do not believe that the proportion of stage III T1 CRCs at diagnosis
in screen-detected and non-screen-detected T1 CRC patients will differ significantly
in the current situation where the screening program is fully implemented.
A second limitation is the difference in treatment strategies between screen-detected
and non-screen-detected T1 CRCs. As previously observed in a population-based study
[1], screen-detected T1 CRCs were more often treated with local excision first. This
may have caused a selection bias toward more high risk cases in the screen-detected
primary surgery group. Eventually, however, the same proportion of screen-detected
and non-screen-detected patients underwent surgical resection with removal of the
draining lymph nodes, as the proportion of completion surgery was higher in the screen-detected
group, leading to a similar overall surgical intervention rate. Moreover, screen-detected
patients who only underwent local excision showed a similar 5-year TTR and MFS compared
with non-screen-detected patients, indicating that both groups seem to have a similar
oncologic risk profile.
A post hoc multivariable logistic regression analysis, in which we studied potential
factors associated with primary surgery, showed that the difference in treatment strategy
may be partly explained by the higher proportion of right-sided tumors found in non-screen-detected
patients (data not shown). Local resection options in the proximal colon are limited
and sometimes only performed by experienced endoscopists in expert centers. Other
factors independently associated with a primary surgical resection were nonpedunculated
morphology and larger polyp size. However, this could not completely explain the higher
local resection rate in screen-detected patients, because detection through screening
showed an independent association with a lower primary surgical resection rate, also
after adjustment for ASA classification, age at diagnosis, year of diagnosis, and
the aforementioned factors.
It is therefore not completely clear why primary surgery was more often chosen in
non-screen-detected patients. Perhaps unmeasured parameters, such as the experience
of the endoscopists, differences in optical features, perception of the impact of
surgery, and patient preference, may have played an important role. Because a screening
endoscopist needs specific accreditation to perform colonoscopies in the Dutch CRC
screening program, it might also be that these endoscopists are better equipped to
recognize and treat malignant polyps, which could have influenced the treatment strategy.
Finally, although this is a large-sized study, we cannot rule out that an even larger
study population would have revealed a significant difference in the proportion of
LNM between screen-detected and non-screen-detected patients. The sensitivity analysis,
which was restricted to patients diagnosed between the ages of 55 and 80, showed that
screen-detected patients had a 1.52 times higher chance of LNM compared with patients
outside the screening program, with a confidence interval of 0.91–2.53. It therefore
cannot be concluded with complete certainty that screen-detected patients do not have
a slightly increased risk of LNM. This potential unrecognized difference in the risk
of LNM is however not reflected in a difference in risk of local recurrence and distant
metastasis, which are the most important oncologic outcomes. It is therefore more
likely that the observed difference in LNM is based on chance or selection, rather
than that there is a true clinically relevant difference.
In conclusion, both the rise in T1 CRCs detected through screening and the ongoing
shift toward more local resections of early-stage CRC make it important to urgently
unravel the oncologic risk profile of screen-detected T1 CRC. Our multicenter study
shows that the risks of LNM and cancer recurrence are not significantly different
in screen-detected and non-screen-detected T1 CRCs. The comparable performance of
conventional risk factors for LNM and the similar rates of TTR, MFS, and CSS observed
in the two groups further confirm that T1 CRCs detected through screening can be treated
similarly to T1 CRCs detected outside of the screening program.
