Evaluating colonoscopy screening intervals in patients with Lynch syndrome from a large Canadian registry

Abstract

Background

Lynch syndrome (LS) screening guidelines originally recommended colonoscopy every 1 to 2 years, beginning between the ages of 20 and 25 years. Recent studies have questioned the benefits of these short screening intervals in preventing colorectal cancer (CRC). Our goal is to determine how colonoscopy screening intervals impact CRC in patients with LS.

Methods

We analyzed the demographics, screening practices, and outcomes of patients with LS identified through the clinic based Familial Gastrointestinal Cancer Registry at the Zane Cohen Centre, Sinai Health System, Toronto, Canada.

Results

A total of 429 patients with LS were identified with median follow-up of 9.2 years; 44 developed CRC. We found a positive trend between shorter screening intervals and the number of adenomas detected during colonoscopy. Any new adenoma detected at screening decreased 10-year CRC incidence by 11.3%. For MLH1 carriers, a screening interval of 1-2 years vs 2-3 years led to a 20-year cumulative CRC risk reduction of 28% and 14% in females and males, respectively. For MSH2 carriers, this risk reduction was 29% and 17%, respectively, and for male MSH6 carriers 18%. Individuals without any adenomas detected (53.4% of LS carriers) had an increased 20-year CRC risk of 25.7% and 57.2% for women and men, respectively, compared with those diagnosed with adenomas at screening.

Conclusions

The recommended colonoscopy screening interval of 1-2 years is efficient at detecting adenomas and reducing CRC risk. The observation that 53.4% of LS patients never had an adenoma warrants further investigation about a possible adenoma-free pathway.

In Canada, colorectal cancer (CRC) is the third most commonly diagnosed cancer and the second and third leading cause of cancer death in men and women, respectively (1). Lynch syndrome (LS) is the most common hereditary CRC syndrome, accounting for 2%-5% of all CRC (2,3). LS is an autosomal dominant disorder, with variable penetrance, caused by mutations in DNA mismatch repair (MMR) genes (4). Individuals with LS are at increased risk for multiple cancers, with CRC and endometrial cancer being the most common (5). In North America, cumulative CRC risks by age 70 years among individuals with pathogenic or likely pathogenic variants (P/LPV) in MLH1, MSH2, MSH6, and PMS2 were estimated at 34%, 33%, 11%, and 6% in females and 44%, 37%, 22%, and 12% in males, respectively (6). An earlier study evaluating colonoscopy screening in LS found that screening every 3 years decreased CRC rate by 62% and overall mortality by about 65% (7).

LS screening guidelines originally recommended colonoscopy every 1 to 2 years, beginning between the ages of 20 and 25 years (or 5 years before the youngest age at CRC) (8). In more recent years, the European groups, the European Hereditary Tumour Group, the European Society of Coloproctology, and the European Society of Gastrointestinal Endoscopy, recommend delaying the age to begin colonoscopies to age 35 years, in asymptomatic individuals with P/LPV in the less penetrant MSH6 and PMS2 genes, and similarly, the US National Comprehensive Cancer Network recommends beginning at age 30-35 years for the same genes (9-11). The screening intervals were also extended, with the National Comprehensive Cancer Network recommending every 1-3 years for individuals with P/LPV in MSH6 and PMS2 genes and the European Hereditary Tumour Group and European Society of Coloproctology recommending colonoscopy every 5 years for individuals with PMS2 P/LPV and every 2-3 for the remaining MMR genes. (9,10). The issue of screening interval in patients with LS has been a subject of controversy in recent years where studies have questioned the benefits of shorter colonoscopy screening intervals for all MMR carriers (<2 years vs 2-3 years) in reducing CRC incidence and overall mortality (12-15). The aim of this current study is to robustly evaluate the association between colonoscopy screening intervals and CRC risk by accounting, unlike previous studies, for the full history of colonoscopies and history of adenomas in all patients recruited from a large series of individuals with LS.

Methods

Study population

Patients with LS who had undergone colonoscopy screening at decentralized hospitals were identified through the Familial Gastrointestinal Cancer Registry (FGICR) at the Zane Cohen Centre (ZCC), Sinai Health System, Toronto, Canada. Patients were eligible for the study if they had 1) a proven P/LPV in 1 of the 4 MMR genes (MLH1, MSH2, MSH6, PMS2) or EPCAM; 2) had completed at least 2 colonoscopies; and 3) had no prior colorectal surgery and no history of CRC at study entry.

Colonoscopy data were gathered prospectively for all individuals with age at first colonoscopy used for study entry. As expected with clinical care, screening intervals varied within the same patient and across patients, allowing for in-depth evaluation of the relationship between screening intervals and the detection of adenomas or cancer. Consecutive visits occurring less than 4 months apart were considered as a single visit starting at the age of the first of the 2 visits. Our primary endpoint was time to CRC, and event times were censored at the time of colorectal resection surgery or if they were diagnosed with invasive or in situ CRC.

Demographic and personal data collected for analysis included gender, age at study entry, MMR gene P/LPV status, proband’s CRC status, and age. Screening and clinical data included age at colonoscopy, adenoma, other polyps, CRC detection, cancer stage, and type of colorectal resection surgery. Our definition of adenomas included those classified as adenomatous polyps, villous adenomas, or tubulovillous adenomas. It did not include sessile serrated adenomas, hyperplastic polyps, or inflammatory polyps, although these were tracked separately.

This study has been approved by the institutional human ethics committees.

Statistical analyses

We propose here a new methodology based on an instrumental variable approach (16) where we use screening intervals as an instrument for the number of adenomas detected during the colonoscopy. It assumes that screening intervals affect CRC at a time t only through the number of adenomas detected before t after controlling for confounding variables. Instrumental variable methods were primarily introduced for linear regression models and more recently for survival analysis (17). Here, we extended the 2-step regression approach of Tchetgen Tchetgen et al. (17) to the situation where the exposure (adenoma) is a count data, and the exposure and instrument (screening intervals) are time-dependent variables (see Supplementary Figure 1 and Supplementary Methods, available online, for more details). For the first step, we modeled the association of screening intervals between 2 consecutive colonoscopies (categorized as ≤1 year, 1-2 years, 2-3 years, >3 years) with the number of adenomas detected within the intervals using a hurdle Poisson mixed effects model. For each patient, we considered the full history of colonoscopies and adenomas detected until his or her last colonoscopy, time of colorectal resection surgery, in situ CRC, or invasive CRC, whichever comes first. Each individual contributed multiple screening intervals to the analysis with variable lengths and within each interval, adenoma(s) could be detected or not. From this model, we predicted the total number of adenomas detected in each screening interval for all patients. In the second step of our analysis, we evaluated the association between the predicted number of adenomas obtained from the first step and the time to CRC using a parametric Weibull survival model. All screening intervals of a patient up to a time t were considered in the survival model, and the corresponding predicted total number of adenomas over all these intervals was considered as a time varying covariate, subject to change with each new colonoscopy and affecting CRC at time t (18) (see Supplementary Figure 1, available online). We used the observed number of adenomas detected at the first colonoscopy as a covariate in the model. In addition, both steps of analysis were adjusted for other confounding variables including age at study entry, gender, and age of each LS family’s proband (index case) CRC diagnosis. In additional sensitivity analyses, we also considered several variables that could be related to colonoscopy quality including technical advances in endoscopic equipment measured by year of initiation (> or ≤ 2010); academic or nonacademic hospital, which may suggest clinical awareness of LS (Sinai Health, other academic hospitals in Ontario, other health centers and hospitals in Ontario); and the number of polyps not submitted to pathology at the initial and subsequent colonoscopies because of fulguration, cauterization, or lost sample. The age at study entry and age of proband CRC diagnosis were standardized by subtracting the sample mean and dividing by 10 years. The follow-up period for each patient started at the age of first colonoscopy (assumed time 0) for the eligible individuals up to the age at CRC for individuals with a cancer diagnosis, age at censoring because of colorectal resection surgery or diagnosis of invasive or in situ CRC, or age at last known colonoscopy or last information available for other individuals. Using the model described, we combined the 2 steps of the analysis to estimate and compare CRC risk by hazard ratio (HR) for the various colonoscopy screening intervals for males and females, subcategorized by specific MMR gene status (ie, carrier or noncarrier of a P/LPV) where CRC hazard ratio was assessed based on the cumulative number of adenomas predicted at screening prior to CRC. Descriptive statistics of screening and patient characteristics are presented as median and interquartile range (IQR), and group comparisons were done with Kruskal-Wallis tests or χ² tests where appropriate. The sample median screening interval was derived by computing the median screening interval of each patient and then by taking the median of all these values. Tests of trends across groups are performed with quantile regression at the median value of the variable of interest using a 2-sided t test for the regression coefficient of the group variable, considered as continuous. A P value less than .05 was considered statistically significant for all tests. Statistical analyses were conducted using R software, version 3.6.1 (R Core Team).

Results

Characteristics of patients with LS

A total of 429 patients with LS from 270 families, with a median follow-up of 9.2 (IQR = 5.1-14.1) years, were identified through the Familial Gastrointestinal Cancer Registry (Tables 1-2). This included 130, 178, 87, 26, and 8 individuals with a P/LPV in MLH1, MSH2, MSH6, PMS2, and EPCAM, respectively (Table 2). During colonoscopy screening, 170 (39.6%) individuals had 483 new adenomas diagnosed, and 85 (19.8%) individuals had at least 1 adenoma diagnosed at the first colonoscopy; 229 (53.4%) individuals had no history of adenoma (Table 1). The median number of adenomas diagnosed in patients with a history of adenomas was 2 (IQR = 1-3 adenomas) (Table 1). In those patients, adenomas were diagnosed at an estimated median of 1.9  (IQR = 1.3-2.7) years from the previous colonoscopy. For the entire cohort and for individuals who had at least 1 adenoma detected during their screening, the adenoma detection rate (ADR), which measures the proportion of colonoscopies with at least 1 adenoma detected (16), was 19.8% and 32.9% at first colonoscopy, and 17.5% and 44.1% for subsequent colonoscopies, respectively.

Table 1.

Colonoscopy screening characteristics presented as median and interquartile range (IQR) among the 429 patients with LS

Characteristics	Overal Median (IQR)	Patients with a history of adenomasa Median (IQR)	Patients without history of adenomasb Median (IQR)
Follow-up time, y	n = 429	n = 200	n = 229
Overall	9.2 (5.1-14.1)	10.6 (6.6-16.1)	7.8 (4.0-12.3)
MLH1 carriers	10.4 (5.3-15.0)	11.4 (6.6-15.8)	9.5 (4.5-14.2)
MSH2 carriers	9.8 (6.3-14.8)	11.6 (8.3-17.6)	7.8 (4.5-11.6)
MSH6 or PMS2 carriers	6.8 (3.8-11.8)	9.2 (5.4-12.5)	6.3 (3.7-10.8)
P c	1.5 × 10−3	9.7 × 10−3	.03
No. colonoscopies per patientd	n = 429	n = 200	n = 229
Overall	4 (3-6)	5 (3-7)	3 (2-5)
MLH1 carriers	4 (3-7)	5 (3-7)	4 (2-5)
MSH2 carriers	5 (3-7)	6 (5-8)	3 (2-6)
MSH6 or PMS2 carriers	3 (2-5)	4 (3-5)	3 (2-4)
P	3.5 × 10−7	2.4 × 10−6	7.2 × 10−3
Age at the first colonoscopy, y	n = 429	n = 200	n = 229
Overall	40.9 (30.8-51.2)	44.7 (35.2-53.2)	37.5 (28.0-49.5)
MLH1 carriers	37.1 (27.1-48.7)	39.4 (31.0-49.5)	34.8 (25.1-46.0)
MSH2 carriers	39.3 (30.3-50.1)	44.3 (33.5-52.3)	35.0 (27.0-47.9)
MSH6 or PMS2 carriers	47.4 (39.7-54.4)	51.9 (43.3-58.3)	43.7 (36.1-52.0)
P	6.6 × 10−8	1.4 × 10−5	2.7 × 10−4
Interval between colonoscopies, ye	n = 429	n = 200	n = 229
Overall	2.0 (1.3-3.0)	1.8 (1.2-2.5)	2.0 (1.4-3.3)
MLH1 carriers	2.0 (1.3-3.0)	1.9 (1.2-2.6)	2.1 (1.4-3.0)
MSH2 carriers	1.7 (1.3-2.5)	1.6 (1.2-2.1)	2.0 (1.4-3.2)
MSH6 or PMS2 carriers	2.0 (1.4-3.5)	2.0 (1.3-2.8)	2.0 (1.7-3.9)
P	.03	.10	.30
No. adenomas per patient at the first colonoscopy	—	n = 85	—
Overall	—	1 (1-2)	—
MLH1 carriers	—	1 (1-1)	—
MSH2 carriers	—	1 (1-2)	—
MSH6 or PMS2 carriers	—	1 (1-2)	—
P		.43
No. adenomas per patient after the first colonoscopy, median (IQR)		(n = 170)
Overall	—	2 (1-3)	—
MLH1 carriers	—	1 (1-2)	—
MSH2 carriers	—	2 (1-4)	—
MSH6 or PMS2 carriers	—	1 (0-2)	—
P		3.0 × 10-5
Interval from last colonoscopy to adenoma, ye	—	n = 170	—
Overall	—	1.9 (1.3-2.7)	—
MLH1 carriers	—	2.0 (1.5-3.2)	—
MSH2 carriers	—	1.6 (1.2-2.4)	—
MSH6 or PMS2 carriers	—	2.0 (1.4-2.5)	—
P		.12
Age at CRC onset in individuals with CRC, y	n = 44	n = 31	n = 13
Overall	53.4 (45.6-60.6)	54.2 (45.8-60.7)	52.5 (46.1-58.7)
MLH1 carriers	48.4 (41.9-56.2)	48.4 (36.1-55.8)	60.6 (50.9-62.8)
MSH2 carriers	58.0 (50.2-61.0)	58.5 (50.6-61.0)	52.1 (39.9-58.0)
MSH6 or PMS2 carriers	60.7 (56.2-67.1)	63.9 (59.1-69.2)	51.0 (43.0-60.0)
P	.01	8.4 × 10-3	.84
Interval from last colonoscopy to CRC for individuals with CRC, y	n = 44	n = 31	n = 13
Overall	2.5 (1.9-3.9)	2.5 (1.7-4.0)	2.6 (2.3-3.4)
MLH1 carriers	2.6 (1.4-4.0)	2.3 (1.3-4.0)	2.6 (2.3-3.8)
MSH2 carriers	2.5 (2.1-2.8)	2.5 (2.1-3.5)	2.6 (2.1-2.8)
MSH6 or PMS2 carriers	3.3 (2.5-12.1)	2.9 (2.2-5.5)	—
P	.47	.67	—

^aIncludes individuals with either adenomas detected at the first colonoscopy (n = 85) or after the first colonoscopy (n = 170) unless the variable of interest concerns specifically only the first colonoscopy or subsequent colonoscopies (in that case n is indicated). CRC = colorectal cancer; LS = Lynch syndrome.

^bIncludes individuals with no adenoma detected at the first colonoscopy or after the first colonoscopy.

^c P value from Kruskal-Wallis test comparing the distribution of the variable of interest across 3 groups of carriers: MLH1, MSH2, MSH6 and PMS2 combined.

^dCorresponds to number of colonoscopy visits before CRC for LS patients with CRC.

^eThe median screening interval is derived as median over all patients’ median value.

Table 2.

Description of the series of Lynch syndrome families from Toronto

Patients	No. of LS families	No. of individuals			No. of individuals with at least 1 adenoma detectedb		No. of individuals with a first CRC		No. of deaths after CRC
		Total	Males	Females	Males	Females	Males	Females	Males	Females
All LS	270	429	132	297	67	133	19	25	4	2
MLH1	78	130	48	82	29	34	11	12	2	1
MSH2	102	178	60	118	32	52	5	11	1	1
MSH6	63	87	12	75	17	23	2	1	1	0
PMS2	21	26	8	18	4	6	1	1	0	0
EPCAM a	7	8	4	4	1	2	0	0	0	0

^aTwo individuals were noted to have pathogenic or likely pathogenic variants extending into MSH2. CRC = colorectal cancer; LS = Lynch syndrome.

^bNote this number also includes 31 individuals (17 women and 14 men) who also had CRC.

Characteristics of screening colonoscopies

Colonoscopies were decentralized, which is representative of a large population, however, 29% of the overall screening was performed at the specialized LS program within the ZCC. Of the remaining colonoscopies performed outside of the ZCC, we estimated that 60% were performed by gastroenterologists, and the remaining were performed by colorectal surgeons. A complete description of colonoscopies is given in Table 1 and Supplementary Figures 2 and 3 (available online). The 429 patients with LS underwent a total of 2044 colonoscopies with a median of 4 (IQR = 3-6) colonoscopies per patient and at a median interval between colonoscopies of 2.0  (IQR = 1.3-3.0 years). The distribution of the number of colonoscopies per patient differs statistically across patients with P/LPV in MLH1, MSH2, and MSH6/PMS2 (P = 3.5 × 10^-7) with a median of 4, 5, and 3, respectively. There was an increasing trend in the median screening interval between patients with a history of adenoma and no CRC (1.7 years), those with no history of adenoma (2.0 years), and those diagnosed with CRC (2.4 years) (P = 8.3 × 10^-3) (Table 3). The distribution of colonoscopies by screening interval categories (Table 4) shows that almost half (49.3%) of the colonoscopies in CRC patients exceeded the 1-2 year recommended interval compared with patients with no CRC (35.5%) (χ², P = 4.9 × 10^-4), and 32 of the 44 (73%) CRC patients (20 of 30 with a history of adenoma, and 12 of 14 with no history of adenoma) exceeded the 1- to 2-year interval directly prior to their cancer diagnosis. There were also significant differences in screening intervals according to patient history of adenoma. The vast majority of colonoscopies in patients with a history of adenoma was less than 2 years apart (ie, 67.0% compared with 59.7% in patients with no history of adenoma; P = 9.2 × 10^-3). This difference was more substantial in patients who eventually developed CRC (55.1% vs 35.1%; P = .046).

Table 3.

Screening interval (in years) between colonoscopies per patient according to adenoma detection (at least 1 adenoma detected) and CRC status

Patients	1. No adenoma and no CRC	2. Adenoma detected and no CRC	3. CRC	P a
Overall
No.	245	140	44
Median (IQR)	2.0 (1.4-3.1)	1.7 (1.2-2.2)	2.4 (1.4-3.7)	8.3 × 10-3
MLH1 carriers
No.	70	37	23
Median (IQR)	2.1 (1.4-3.0)	1.8 (1.2-2.2)	2.7 (1.6-4.1)	.15
MSH2 carriers
No.	95	67	16
Median (IQR)	2.0 (1.4-3.0)	1.6 (1.1-2.1)	2.1 (1.4-3.2)	.07
MSH6 or PMS2 carriers
No.	74	34	5
Median (IQR)	2.1 (1.6-3.9)	1.9 (1.4-2.6)	2.9 (2.5-7.6)	.32

^a P value based on quantile regression to test for a trend between the median screening interval and patient’s group coded as 1 = at least 1 adenoma detected and no CRC; 2 = no adenoma or CRC detected; 3 = CRC. The test of trends across groups is performed using a 2-sided t test for the regression coefficient of the group variable, considered as continuous. The median screening interval is derived as median over all patients’ median value. CRC = colorectal cancer; IQR = interquartile range.

Table 4.

Number of colonoscopy screening intervals (and proportion) with a specific length according to patients’ history of adenoma and patients’ CRC status

Screening interval	All patients	Patients without CRC	Patients with CRC
	(n = 429)	(n = 385)	(n = 44)
	All colonoscopies N (%)	All colonoscopies N (%)	All colonoscopies N (%)	Last colonoscopy prior to CRC diagnosis N (%)
<2 y	1025 (63.5)	947 (64.5)	78 (50.7)	12 (27.3)
2-3 y	286 (17.7)	252 (17.2)	34 (21.9)	16 (36.4)
>3 y	304 (18.8)	261 (18.3)	43 (27.7)	16 (36.4)
Patients with a history of adenoma
<2 y	558 (67.0)	493 (69.0)	65 (55.1)	10 (33.3)
2-3 y	136 (16.3)	115 (16.1)	21 (17.8)	9 (30.0)
>3 y	139 (16.7)	107 (15.0)	32 (27.1)	11 (36.7)
Patients with no history of adenoma
<2 y	467 (59.7)	454 (60.9)	13 (35.1)	2 (14.3)
2-3 y	150 (19.2)	137 (18.4)	13 (35.1)	7 (50.0)
>3 y	165 (21.1)	154 (20.7)	11 (29.7)	5 (35.7)
P a	9.2 × 10-3	3.4 × 10-3	.046	.31

^a P value based on χ² test with 2 degrees of freedom comparing patients with adenomas(s) and patients without adenomas with respect to the distribution of screening intervals. CRC = colorectal cancer.

Two-step regression approach

Our results indicate that the incidence of CRC in patients with LS is largely influenced by the history of colonoscopies and adenomas. This motivated our use of the instrumental variable approach described above based on 2-step regression modeling (see Methods).

Association between colonoscopy screening intervals and the number of adenomas detected during screening (step 1)

We observed a consistent association between colonoscopy interval and the number of adenomas detected (Figure 1). After 20 years of follow-up, the annual number of adenomas per patient was estimated to be 1.6, 1.4, 0.6, and 0.3 for females and 1.4, 1.2, 0.5, and 0.3 for males with screening intervals of less than 1 year, 1-2 years, 2-3 years and more than 3 years, respectively. These estimates corresponded to 20-year cumulative adenoma numbers of 20.1, 12.3, 7.8, and 6.3 for females and 17.7, 10.9, 6.9, and 5.6 for males, for the respective intervals. Compared with a screening interval of less than 1 year and after confounder adjustment (gender, age at study entry, proband age of onset, and number of adenomas detected at the first colonoscopy), a screening interval of 1-2 years (as recommended by clinical guidelines) vs less than 1 year was associated with an estimated 3.9% (P = .04) lower annual rate of adenomas detected per individual (ie, the number of adenomas detected over a follow-up time [in years]). Comparatively, this reduction was 9.5% (P < .001) and 13.9% (P < .0001) for a screening interval of 2-3 years and more than 3 years vs less than 1 year, respectively (Supplementary Methods 3.1, available online). For the 245 individuals who did not have any adenomas found at colonoscopy, an ADR of 0 was not associated with any screening interval, however, it was significantly associated with a younger age at colonoscopy (Supplementary Tables 1 and 4, available online).

Figure 1.

Number of adenomas detected during follow-up with respect to colonoscopy screening intervals in male and female patients with LS. Each solid line represents the predicted number of adenomas obtained from step 1 of the analysis for a fictive patient with LS, with an average age at first colonoscopy of 41.3 years, no adenoma detected at the first colonoscopy, and with a fixed screening interval of 1, 2, 3, or 4 years, which remains the same over the study period. Each curve starts at the time of the second colonoscopy (ie, 1, 2, 3, and 4 years for the screening intervals of ≤1, 1-2, 2-3, and >3 years, respectively). Dotted lines represent the 95% confidence intervals. LS = Lynch syndrome.

Association between the number of adenomas detected during screening and CRC risk (step 2)

After adjusting for gender, age at study entry, proband’s age at onset, and the number of adenomas at the first colonoscopy, we found that any new adenoma detected decreased the 10-year CRC risk by 11.3% (HR = 0.88, 95% confidence interval [CI] = 0.78 to 0.99; P = .04; Supplementary Methods 3.2, available online).

Association between colonoscopy screening intervals and CRC risk (combines steps 1 and 2)

A total of 44 individuals were diagnosed with CRC during follow-up at a mean of 4.0 (4.2) years from previous colonoscopy with a mean age of CRC diagnosis of 52.9 (11.3) years (Table 2; Supplementary Table 2, available online). In women with LS, the 20-year cumulative CRC risk was 10.0% (95% CI = 4.4% to 24.9%), 9.2% (95% CI = 3.6% to 24.5%), 13.4% (95% CI = 6.9% to 28.5%), and 16.4% (95% CI = 9.8% to 32.2%) for screening intervals of less than 1 year, 1-2 years, 2-3 years, and more than 3 years, respectively. In men, these 20-year cumulative CRC risk estimates were 36.2% (95% CI = 21.0% to 65.0%), 35.6% (95% CI = 19.9% to 64.6%), 43.4% (95% CI = 28.5 to 74.3%), and 47.7% (95% CI = 32.8% to 79.6%), respectively (Figure 2, A).

Figure 2.

Cumulative incidence of CRC in male and female patients with LS from the first colonoscopy visit to 20 years of follow-up: A) by screening interval and history of adenoma; B) by gene-specific pathogenic variant status and colonoscopy screening intervals. Each solid curve represents CRC risk for a fictive patient with LS, with an average age at first colonoscopy of 41.3 years, no adenoma detected at the first colonoscopy, and with a fixed screening interval of 1, 2, 3, or 4 years, which remains the same over the study period. Dotted lines represent the 95% confidence intervals. CI = confidence interval; CRC = colorectal cancer; LS = Lynch syndrome.

For LS patients without any adenomas detected during screening, the 20-year CRC cumulative risk was 25.7% (95% CI = 17.9% to 49.3%) in women and 57.2% (95% CI = 39.6% to 88.8%) in men.

Association between colonoscopy screening intervals and CRC risk by MMR gene (combines steps 1 and 2)

We identified significant differences across the gene-specific P/LPV status in terms of number of colonoscopies per patient, age at first colonoscopy, number of adenomas, age at first CRC, and follow-up time (Table 2). There was a significantly increased trend in the interval between colonoscopies across patients with LS who 1) had at least 1 adenoma detected; 2) had no adenoma and no CRC detected; and 3) developed a primary CRC (Table 3). This trend was significant overall (P < 10^-5) for MLH1 (P < 10^-3) and MSH2 (P = .01) carriers but not for the combined group of MSH6 and PMS2 carriers (P = .14).

For patients with LS who developed adenomas during follow-up, we found a clear trend between shorter screening intervals (between 1 and 2 years, between 2 and 3 years, more than 3 years) and CRC incidence reduction. For instance, in MLH1 carriers, a screening interval between 1-2 years vs 2-3 years leads to a 28% (15.6% vs 21.7%) and 14% (48.0% vs 55.9%) reduction in CRC cumulative risk in females and males, respectively, after 20 years of follow-up (Figure 2, B; Supplementary Table 3, available online). In MSH2 carriers, this risk reduction is 29% (6.9% vs 9.7%) in females and 17% (24.0% vs 29.0%) in males, respectively. For MSH6 carriers, we were only able to evaluate the risk reduction in males, which is 18% (16.8% vs.20.6%). The larger reduction in CRC risk observed in females vs males is explained by a higher number of adenomas detected in females compared with males. This risk reduction was more substantial when comparing a screening interval of 1-2 years vs more than 3 years (ie, in MLH1 carriers, 40% and 20% in females and males), respectively (Figure 2, B; Supplementary Table 3, available online). For MSH2 carriers, this risk reduction was 41% and 25%, respectively, and for male MSH6 carriers 26%. Finally, we note almost no difference in CRC risk reduction for a screening interval less than 1 year vs 1-2 years.

CRC stage and overall survival

During screening follow-up, 44 (19 females, 25 males) individuals were diagnosed with CRC. The American Joint Committee on Cancer staging was I (n = 25, 56.8%), II (n = 10, 22.7%), III (n = 7, 15.9%), and unknown (n = 2, 4.5%) (Supplementary Table 2, available online). Among the LS patients who developed an incident primary CRC, 6 (2 females, 4 males) died during follow-up (Table 1). Ten-year overall survival was 88.3% (95% CI = 76.6% to 100%).

Sensitivity analysis

The validity of our statistical approach relies on identifying potential confounding variables that could be associated with adenoma detection (the number of adenomas and the chance to have 0 adenoma detected in a screening interval) and the time to CRC. Because adenoma detection is strongly related to the quality of colonoscopy, we adjusted the 2 steps of our analysis for additional variables that could be related to colonoscopy quality including the year a colonoscopy was performed, the type of hospital that performed the colonoscopy, and the number of polyps not submitted to pathology. The results from our analyses remain unchanged when adjusting for these additional variables (section 4 of Supplementary Material, Supplementary Tables 6 and 7, available online).

Discussion

The evaluation of colonoscopy screening intervals is critical for the clinical management of patients with LS carrying P/LPV in MMR genes, yet it remains a subject of controversy. Our new approach based on an instrumental variable methodology acknowledges that the role of colonoscopy screening intervals on CRC is through the detection of adenomas and no other factors. It contrasts with recent studies in several key aspects.

First, recent studies (12-15) did not account for the full history of colonoscopies in LS patients and used indirect evidence to assess the association between colonoscopy screening intervals and CRC incidence. One study compared CRC incidence in a large prospective cohort of LS patients undergoing active colonoscopy screening (Prospective Lynch Syndrome Database cohort) to that estimated from retrospective family studies (19). This is problematic because colonoscopy screening is not well assessed in family studies (6,20), and intensive colonoscopy screening in prospective studies is likely to lead to an increase in incidence of CRC because of early diagnosis, especially directly after screening is initiated (21). Other studies suggest that CRC incidence and CRC stage at diagnosis are not reduced in patients with LS in countries that adopt intense vs less intense screening policies (colonoscopy every 1-2 years vs 2-3 years) (14-15). In these studies, a fixed screening interval is assumed for each patient based on the country guidelines for LS (eg, the mean screening interval), however, it does not account for differences in adherence to screening guidelines. These studies may not account for decentralized screening, which is more typical for individuals with LS living in a large country. Our results indicate that among patients diagnosed with CRC, 32 of 44 (20 of 30 with a history of adenoma and 12 of 14 with no history of adenoma) patients had their last colonoscopy more than 2 years before their CRC diagnosis. Therefore, having a real account of screening intervals per patient instead of mean intervals is critical when assessing the role of colonoscopy screening intervals on CRC incidence, as having a lapse in interval directly before a CRC diagnosis can be identified.

The second key concept of our study and strong motivation for using the instrumental variable approach was to consider a patient history of adenomas as the main mediator of the association between screening intervals and CRC incidence, in contrast with previous studies that did not assess this information (12-15). Engel et al. (15) reported a positive association between shorter colonoscopy screening intervals and the incidence of adenomas but did not assess specifically how adenoma detection itself could modify CRC incidence. Because adenomas are in the causal pathway that leads to progression to CRC, they cannot be ignored from the analysis. Clinical evidence also suggests that a higher ADR is associated with a reduced risk of interval CRC, advanced-stage interval CRC, and fatal interval CRC (22-25). In our study, we reported a positive trend between shorter screening intervals and the number of adenomas detected during colonoscopy screening. It has been mentioned previously that adenomas could regress on their own although this hypothesis needs further evidence (26). A more plausible explanation is that adenoma or CRC risks are quite heterogeneous across patients with LS. For instance, some individuals carrying a pathogenic variant in the same gene might have a much higher risk than others because of unknown familial risk (6). This heterogeneity could also be due to specific family history of cancer and/or social and behavioral exposures (eg, tobacco, alcohol, diet, exercise) or preventive tools such as aspirin use. A more intensive colonoscopy screening with shorter intervals might be able to detect adenomas at a higher rate and prevent CRC efficiently in these high-risk individuals and thus explains the positive trend between shorter screening intervals and the number of adenomas detected.

The last critical aspect of our instrumental variable approach was to account for patients who had no history of adenoma (53.4% of the LS patients). It is noteworthy that these patients had higher CRC incidence compared with those with a history of adenoma. Because these patients were younger on the whole, it is possible they have not reached the median age to develop adenomas. Another possible explanation is that they adhere less to the screening guidelines and may not be as motivated to have frequent colonoscopies. It is also possible that these individuals had missed adenomas on a prior colonoscopy. Missed rates of 26% for adenomas and 9% for advanced adenomas have been reported in other studies (27), and missed lesions is one of the main causes of postcolonoscopy CRC (24). It has been claimed that improving the quality of colonoscopies of polyp resection and the adherence to screening intervals would prevent most CRCs in LS (28). Interestingly, the chance to have no adenoma detected during a colonoscopy interval was not associated with variables used as proxy for colonoscopy quality or clinical expertise in our study (Supplementary Table 5, available online). This supports a possible adenoma-free progression to cancer, which postulates that the progression of cancer could start from MMR deficient crypt foci in the normal colonic mucosa of LS carriers (29-31).

From a clinical standpoint, our major finding highlights the importance of the 1-2 year colonoscopy interval compared with the 2-3 years, which has recently been the subject of debate (12-15). For MLH1 carriers, a screening interval of 1-2 years vs 2-3 years led to a 20-year cumulative CRC risk reduction of 28% and 14% in females and males. For MSH2 carriers, this risk reduction was 29% and 17%, respectively, and for male MSH6 carriers 18%. Adenomas are known to develop rapidly in LS, therefore complying to screening guidelines is key to prevent CRC. We also found that the reduction in CRC incidence was even more substantial when comparing the 1-2 years to more than 3 years screening interval. This is less relevant from a clinical standpoint when patients with LS are under active screening (because this interval is less likely to be observed), however, it stresses the risk that can incur in patients when they drop out from active screening.

Our current study has several strengths as well as some limitations. One major advantage is the long follow-up of patients with LS with median follow-up of 9.2 years and the use of the full information on colonoscopy history for each patient. The length of follow-up is essential to avoid possible CRC overdiagnosis and to provide accurate modeling of screening intervals on adenoma detection and on CRC, which we could leverage through our instrumental variable approach. This is also the first study, to our knowledge, to provide an evaluation of screening intervals by assessing its effect on adenoma detection through our instrumental variable approach. The association between screening intervals and CRC is therefore directly related to the effectiveness of colonoscopy screening at detecting adenomas, the precursors of CRC, and not biased by other factors, which is the interest of our instrumental variable approach. Our analyses were also adjusted for important confounding factors such as the age at first colonoscopy, the age of the proband in the family, and gender, and our results were presented by MMR gene variants whenever possible. Unfortunately, we did not have information on aspirin use and could not adjust for this confounder. Yet, the limited number of CRCs reported in our study prevented us from investigating further the role of screening intervals on CRC among carriers of P/LPVs for each MMR gene separately. Therefore, our results might be more representative of carriers of P/LPVs in MLH1 and MSH2 genes who represent 72% of our sample. It is possible that patients with more frequent screening might be more health conscious and eat better, exercise, and are nonsmokers, and it would be helpful to integrate this information in future studies. We have not modeled the familial relationships in our analyses because most LS families had only 1 or 2 relatives. There might have been some underreporting of adenomas in our LS registry, but if it did happen, it was unlikely related to screening intervals or CRC risk and therefore should not have biased our results. We did not account for sessile serrated lesions in our definition of adenomas because they were reported in only 31 individuals. Moreover, comparable rate of sessile serrated lesions has been found in LS vs controls, which suggests that the serrated neoplasia pathway to CRC in LS may be comparable with the general population (32). Finally, our proxies of colonoscopy quality used in the sensitivity analysis were based on technical advances in endoscopic equipment by year and academic vs nonacademic hospitals. However, better standard measures of colonoscopy quality, which include colonoscopist-level ADRs, cecal intubation rates, percent withdrawal times of more than 6 minutes, and bowel prep scores (33), were not available in our study.

Our results support the 1-2 yearly colonoscopy intervals recommended for patients with LS. It is efficient at preventing CRCs through the removal of adenomas. The high rate of patients without any history of adenoma remains to be understood but should not change the recommended screening guidelines. In the event that a rapid developing adenoma-free pathway to CRC may be possible, adhering to the 1- to 2-year recommended interval is critical to identifying malignancies at an early stage.

Data availability

Data for this study were contributed by the FGICR investigators. Availability of these data will depend on an agreement with the FGICR investigators.

Author contributions

Malyssa Aronson, MS, (C)CGC (Conceptualization; Data curation; Investigation; Project administration; Resources; Writing – original draft; Writing – review & Editing); Robert Gryfe, MD (Conceptualization; Investigation; Project administration; Resources; Writing – original draft); Choi Yun-Hee, PhD (Conceptualization; Formal analysis; Investigation; Methodology; Software; Writing – original draft; Writing – review & Editing); Kara Semotiuk, MS, (C)CGC (Resources; Writing – original draft); Spring Holter, MS, (C)CGC (Resources; Writing – original draft); Thomas Ward, MS, CGC (Resources; Writing – original draft); Steve Gallinger, MD, MSc (Resources; Writing – review & Editing); Zane Cohen, MD (Conceptualization; Investigation; Resources; Writing – review & Editing); and Laurent Briollais, PhD (Conceptualization; Formal analysis; Funding acquisition; Investigation; Methodology; Writing – original draft; Writing – review & Editing).

Funding

Operating grants from the Canadian Institute of Health Research and discovery grants from the Natural Sciences and Engineering Research Council of Canada.

Conflicts of interest

The authors have no conflicts of interest to disclose.