Colorectal Cancer Incidence After Colonoscopy at Ages 45–49 or 50–54 Years

Abstract

BACKGROUND & AIMS:

Colorectal cancer (CRC) incidence at ages younger than 50 years is increasing, leading to proposals to lower the CRC screening initiation age to 45 years. Data on the effectiveness of CRC screening at ages 45–49 years are lacking.

METHODS:

We studied the association between undergoing colonoscopy at ages 45–49 or 50–54 years and CRC incidence in a retrospective population-based cohort study using Florida’s linked Healthcare Cost and Utilization Project databases with mandated reporting from 2005 to 2017 and Cox models extended for time-varying exposure.

RESULTS:

Among 195,600 persons with and 2.6 million without exposure to colonoscopy at ages 45–49 years, 276 and 4844 developed CRC, resulting in CRC incidence rates of 20.8 (95% CI, 18.5–23.4) and 30.6 (95% CI, 29.8–31.5) per 100,000 person-years, respectively. Among 660,248 persons with and 2.4 million without exposure to colonoscopy at ages 50–54 years, 798 and 6757 developed CRC, resulting in CRC incidence rates of 19.0 (95% CI, 17.7–20.4) and 51.9 (95% CI, 50.7–53.1) per 100,000 person-years, respectively. The adjusted hazard ratios for incident CRC after undergoing compared with not undergoing colonoscopy were 0.50 (95% CI, 0.44–0.56) at ages 45–49 years and 0.32 (95% CI, 0.29–0.34) at ages 50–54 years. The results were similar for women and men (hazard ratio, 0.48; 95% CI, 0.40–0.57 and hazard ratio, 0.52; 95% CI, 0.43–0.62 at ages 45–49 years, and hazard ratio, 0.35; 95% CI, 0.31–0.39 and hazard ratio, 0.29; 95% CI, 0.26–0.32 at ages 50–54 years, respectively).

CONCLUSIONS:

Colonoscopy at ages 45–49 or 50–54 years was associated with substantial decreases in subsequent CRC incidence. These findings can inform screening guidelines.

Colorectal cancer (CRC) is the third leading cause of cancer death worldwide.¹ CRC incidence and mortality rates are associated with the human development index² and, in the United States, CRC is the second leading cause of cancer death.³ Substantial declines in CRC incidence in older persons in the United States are attributed in large part to widespread screening. In contrast, CRC incidence is increasing in persons younger than age 50 years in the United States and many other countries, possibly due to early-life exposures.^4,5

Most CRCs arise from adenomatous⁶ or sessile serrated lesions,⁷ and colonoscopy with polypectomy can prevent progression to CRC.⁸ Randomized controlled trials of sigmoidoscopy and guaiac fecal occult blood testing (gFOBT) demonstrate that screening decreases CRC incidence and mortality.⁹ In the United States, colonoscopy is the most common CRC screening test.¹⁰ Colonoscopy is accepted for screening¹¹ based on observational data^8,12-15 and by extension from the results of the randomized trials of sigmoidoscopy and gFOBT.⁹

In 2018, the American Cancer Society (ACS) recommended lowering the average-risk CRC screening initiation age from 50 to 45 years.¹⁶ The US Preventive Services Task Force is currently reviewing its guidance and may endorse screening at age 45 years.¹⁷ The US Multi-Society Task Force, and the National Comprehensive Cancer Network still recommend an initiation age of 50 years.^11,18,19 The ACS acknowledged the paucity of data on CRC screening at ages 45–49 years, and based its recommendation on the expectation that screening will perform similarly before and after age 50 years.

Our aims were to study the association between colonoscopy and subsequent development of CRC in people aged 45–49 years, and to compare the results with those aged 50–54 years. We performed a population-based cohort study in Florida, a large and diverse state, using statewide databases with state-mandated reporting.

Methods

General Study Design

We performed a retrospective, population-based cohort study comparing CRC incidence among Florida residents who underwent or did not undergo colonoscopy at ages 45–49 or 50–54 years in 2005–2016. We selected Florida because it is a large and diverse state with little outmigration,²⁰ statewide databases covering all insurance statuses, state-mandated reporting of all colonoscopies, and unique anonymized patient-level identifiers that allow longitudinal follow-up of individuals.

Our study’s design was consistent with multiple prior studies of colonoscopy and CRC, many of which used administrative databases,^13,21-28 including the established precedent of using claims data to study the efficacy of tests seeking to prevent cancer.²⁹

Our research question, study setting, and study design mandated careful consideration of methodological issues.^30,31 Colonoscopy can screen for CRC or its precursors in asymptomatic persons, it can be diagnostic of CRC in symptomatic persons, and it can prevent some CRCs through removal of CRC precursors in asymptomatic or symptomatic persons.^30,31 This unique relationship, in which a test can be both preventive and diagnostic of a disease, presents challenges when studying the association between the test as an exposure and future disease as an outcome.^30,31

In randomized controlled trials of endoscopic screening for CRC, prevalent asymptomatic CRCs are initially uncovered with screening, and these must be included when comparing overall CRC incidence in the screening arm vs the nonscreening control arm. Initially, discovery of prevalent cases leads to a higher CRC rate in the screening arm.³² Subsequent CRC incidence in the screening arm is decreased by removal of CRC precursors at screening, and eventually cumulative CRC incidence in the screening arm becomes lower than in the nonscreening control arm.³² In contrast, because our observational study spanned a study period in which most colonoscopies at ages 45–49 years were anticipated to have nonscreening indications, and because most CRCs are diagnosed at colonoscopy, it was necessary to distinguish colonoscopies diagnostic of CRC, which by definition cannot prevent CRC,^30,31 from exposure colonoscopies that can decrease subsequent CRC risk. We defined colonoscopy as diagnostic of CRC if it occurred within the 6 months preceding the first occurrence of a CRC diagnosis.³³ Furthermore, we had to exclude persons whose initial encounter in the database included a CRC diagnosis with or without colonoscopy because the absence of prior encounters precluded us from determining whether these persons had a previous exposure colonoscopy, and therefore from being able to assign them to a study group. Given these aspects of our study design, our results should be understood to reflect the association between an exposure colonoscopy that, by definition, did not diagnose CRC but may have included removal of CRC precursors, and the development of future CRC.^30,31

Weiss³¹ has suggested that, in cohort studies of screening and cancer incidence, accrual of cases and person-time in those exposed to a screening test should be delayed until prevalent cases in unscreened persons would become clinically evident in order to minimize a healthy screenee bias due to removal of cancers at screening. Weiss acknowledges that the duration of the occult invasive phase of a malignancy cannot be known in any specific instance, and recognizes that when the occult invasive phase may be short relative to the pre-invasive detectable phase, as is likely the case for CRC and polyps, errors in estimating the duration of the occult invasive phase would not detract to any great extent from the validity of the estimate.³¹ Weiss refers to a landmark study in the field that defined exposure to colonoscopy based on questionnaires performed at least 2 years before cancer diagnosis.^12,31 We applied these concepts to the interpretation of our findings and in a post-hoc analysis, as explained in the Statistical Analysis section.

Setting and Data Sources

Florida’s State Ambulatory Surgery Database, State Emergency Department database, and State Inpatient Database are a part of the Healthcare Cost and Utilization Project (HCUP) family of databases, the largest US collection of longitudinal ambulatory, emergency department, and hospital data beginning in 1988.³⁴ HCUP includes databases derived from administrative data developed through federal–state–industry partnerships under the Agency for Healthcare Research and Quality.³⁴ This national information resource undergoes extensive quality checks and supports research on multiple topics.³⁴ For instance, postsurgical infections were studied using HCUP databases including Florida’s,³⁵ and we studied colonoscopy complications using California’s databases.³⁶

The Florida State Ambulatory Surgery Database collects data on outpatient procedures and services, including all colonoscopies. Florida’s Agency for Health Care Administration requires all licensed ambulatory surgery centers with ≥200 records/quarter to submit 1 Ambulatory Patient Data Report per visit.³⁷ The Florida State Emergency Department database collects data from all licensed hospital-owned emergency departments, including standalone departments. The Florida State Inpatient Database includes discharge records for all patients, regardless of payer, from all Florida community hospitals. These datasets contain >100 variables, including patient demographics, residence ZIP codes, all diagnoses and procedure codes, hospital identifiers, and variables that allow longitudinal patient follow-up in compliance with Health Insurance Portability and Accountability Act guidelines.

Each person represented in Florida’s State Ambulatory Surgery Database, State Emergency Department database, and/or State Inpatient Database is assigned a unique identifier based on encrypted personal information, linking all records for a given patient across facilities and time. A variable representing the difference in days between some unique arbitrary date and the date of a clinical encounter in a given calendar year enables longitudinal follow-up of each individual.

Study Population, Exposure and Outcome

We studied Florida residents who were or were not exposed to colonoscopy at ages 45–49 or 50–54 years. All colonoscopies other than those diagnostic of CRC as defined above, regardless of indication, were defined as exposure colonoscopies.²⁹

Our primary analysis focused on CRC incidence in persons exposed to colonoscopy between ages 45 and 49 years in 2005–2016 and persons not exposed to colonoscopy at those ages or younger. We also studied persons exposed to colonoscopy between ages 50 and 54 years in 2005–2016 and persons not exposed to colonoscopy at those ages or younger. All persons were followed until December 31, 2017. Subjects were excluded at study entry if they had a prior diagnosis of CRC, inflammatory bowel disease, or a cancer genetic syndrome during any encounter, or (in order to exclude visitors and minimize failures to ascertain colonoscopy or CRC) a non-Florida residence ZIP code (Supplementary Table 1).

First, to construct the study groups, we began with all eligible persons who had at least 1 medical encounter in any of the databases during years 2005–2016. Second, we excluded those whose initial encounter had a colonoscopy diagnostic of CRC or a CRC diagnosis because the absence of prior encounters precluded us from determining whether these persons previously lived in Florida, with or without exposure colonoscopy, as noted above. All other CRC diagnoses were included in the primary outcome. Third, all persons in the groups at ages 45–49 or 50–54 years were required to be at most 45 or 50 years of age, respectively, at study start. The follow-up period started once these individuals turned 45 or 50 years, respectively. Finally, subjects contributed observation time to these groups until they were diagnosed with CRC, died, had an exposure colonoscopy within the specified ages of interest, or until December 31, 2017, whichever came first. Persons with exposure colonoscopy within the specified ages of interest then switched groups and contributed the rest of their follow-up period and any incident CRC to the group with exposure colonoscopy. Colonoscopies occurring after the specified exposure periods were not considered as primary predictors in order to reflect as closely as possible the clinical decision concerning exposure to colonoscopy at the ages of interest.

Determination of Colonoscopy Exposure, Colonoscopy Indication, Intervention at Colonoscopy, and Colorectal Cancer Diagnosis

We identified colonoscopies and classified indications into 6 groups, as described previously,³⁸ based on Current Procedural Terminology,³⁹ International Classification of Diseases, revisions 9 and 10,^40,41 and Healthcare Common Procedural Coding Systems⁴² codes (Supplementary Tables 2 and 3). In brief, diagnoses were clustered into HCUP Clinical Classifications Software categories, and the Clinical Classifications Software categories were mapped into groups of indications (Table 1). Intervention at colonoscopy (defined as polypectomy, lesion removal, or biopsy) was identified as described previously³⁸ (Supplementary Table 2). CRC was identified by ICD-9 and ICD-10 codes (Supplementary Table 4). Previous research has shown that a diagnosis of CRC in administrative data is very reliable. Hwang et al⁴³ reported sensitivity and specificity for CRC coded in 1 or more encounters in administrative data of 100% and 98.86%, respectively, with positive and negative predictive values of 99.68% and 100%, respectively. Other studies on CRC identification have relied on similar definitions.⁴⁴

Table 1.

Demographic and Clinical Characteristics of the Study Population

Characteristic	Ages 45–49 y		Ages 50–54 y
	With colonoscopy	Without colonoscopy	With colonoscopy	Without colonoscopy
N	195,600	2,570,184	660,248	2,431,744
Sex, n (%)
Women	110,909 (56.7)	1,309,005 (50.9)	353,650 (53.6)	1,222,238 (50.3)
Men	84,691 (43.3)	1,261,179 (49.1)	306,598 (46.4)	1,209,506 (49.7)
Age at entry, n (%)
45 y	31,729 (16.2)	1,990,796 (77.5)	—	—
46 y	43,056 (22.0)	191,695 (7.5)	—	—
47 y	40,598 (20.8)	153,689 (6.0)	—	—
48 y	37,291 (19.0)	127,935 (5.0)	—	—
49 y	42,926 (22.0)	106,069 (4.1)	—	—
50 y	—	—	159,877 (24.2)	1,929,806 (79.4)
51 y	—	—	183,590 (27.8)	170,886 (7.0)
52 y	—	—	131,058 (19.9)	134,009 (5.5)
53 y	—	—	98,092 (14.9)	108,615 (4.5)
54 y	—	—	87,631 (13.3)	88,428 (3.6)
Age, y, mean (SD)	47.1 (1.5)	45.5 (1.1)	51.7 (1.4)	50.5 (1.0)
Follow-up, y, median (IQR)a	7.1 (4.4–9.1)	5.9 (3.3–8.8)	6.4 (3.7–8.9)	4.8 (2.6–7.9)
Race/ethnicity, n (%)
White	121,971 (62.4)	1,536,775 (59.8)	440,189 (66.7)	1,519.924 (62.5)
Black	28,719 (14.7)	426,567 (16.6)	88,211 (13.4)	390,315 (16.1)
Hispanic	33,370 (17.1)	447,430 (17.4)	92,333 (14.0)	382,634 (15.7)
Asian/Pacific Islander	2438 (1.3)	31,055 (1.2)	9707 (1.5)	25,559 (1.1)
Native American	336 (0.17)	4523 (0.18)	1037 (0.16)	4066 (0.17)
Other	5886 (3.0)	88,797 (3.5)	18,739 (2.8)	76,744 (3.2)
Missing	2880 (1.5)	35,037 (1.4)	10,032 (1.5)	32,502 (1.3)
Indication for exposure colonoscopy,b n (%)
Group 1: Gastrointestinal bleeding, occult blood in stools, or iron-deficiency anemia	61,827 (31.6)	—	67,124 (10.2)	—
Group 2: Abnormal finding in radiologic examination of gastrointestinal tract	5862 (3.0)	—	7136 (1.1)	—
Group 3: Weight loss, abdominal pain, or intestinal obstruction	31,104 (15.9)	—	41,124 (6.2)	—
Group 4: Change in bowel habits, diverticulitis/diverticulosis, constipation, diarrhea, irritable colon, or noninfectious gastroenteritis	47,314 (24.2)	—	162,096 (24.6)	—
Group 5: Screening, benign neoplasm, or polyps of large intestine	38,586 (19.7)	—	365,152 (55.3)	—
Group 6: Other indications	10,907 (5.6)	—	17,616 (2.7)	—
Type of exposure colonoscopyc
Without intervention	95,210 (48.7)	—	329,158 (49.9)	—
With intervention	100,390 (51.3)	—	331,090 (50.2)	—

IQR, interquartile range.

^aIn the 45- to 49-year-old group without colonoscopy, 137,324 persons (5.3%), and in the 50- to 54-year-old group without colonoscopy, 403,077 persons (16.6%) switched to the exposed (with colonoscopy) group upon undergoing colonoscopy; 23,778 (0.93%) and 31,544 (1.3%) persons, respectively, died during the study period, and 2,409,802 (93.7%) and 1,997,123 (82.1%) persons, respectively, were alive and in the study group at the end of the study period. In the 45- to 49-year-old and 50- to 54-year-old groups with colonoscopy, 2922 (1.5%) and 7792 (1.2%) persons died during the study period, and 192,678 (98.5%) and 652,456 (98.8%) persons, respectively, were alive at the end of the study period.

^bExposure colonoscopy was defined as colonoscopy with no subsequent CRC diagnosis within 6 months.

^cColonoscopy with or without intervention was determined by International Classification of Diseases, revisions 9 and 10^40,41 and Current Procedural Terminology³⁹ procedure codes (Supplementary Table 2).

Statistical Analysis

CRC incidence rates were calculated per 100,000 person-years of follow-up with 95% confidence intervals (CIs) determined using Byar’s approximation of the exact limits.⁴⁵ Cox models extended to accommodate a time-varying exposure were used to compare the hazard of CRC rates between cohorts with and without exposure to colonoscopy, with adjustment for age, sex, and race/ethnicity. The models incorporated a time-variant variable to capture the occurrence of colonoscopy during the specified age periods. To address the possibility of a healthy screenee bias, we interpreted our figures with consideration of a range of delays in accrual of cases and person-time in those exposed to colonoscopy, as suggested by Weiss,^30,31 and we performed a sensitivity analysis, detailed below, delaying accrual of cases in the colonoscopy group, as done by Nishihara et al.¹² Secondary analyses were performed stratifying by sex, race/ethnicity, and colonoscopy indication.

We performed 4 post-hoc sensitivity analyses. First, to address the possibility of a healthy screenee bias, we performed an analysis in which accrual of CRC cases in the cohorts exposed to colonoscopy was delayed by 2 years.¹² Second, although our analyses were designed a priori to compare cohorts based on colonoscopy during the exposure period without consideration of subsequent colonoscopies, we performed a sensitivity analysis in which persons with subsequent colonoscopy beyond the exposure period were excluded. Third, to address the possibility that persons with only 1 visit with a CRC diagnostic code might not represent true cases of CRC, we performed a second sensitivity analysis in which these persons were excluded. For this analysis, persons with ICD-9 and ICD-10 codes for a history of CRC that occurred after the initial visit with a CRC diagnostic code were retained as CRC cases. Fourth, to address the concern that individual follow-up of ≤1 year might not be clinically meaningful, we performed a sensitivity analysis in which persons with ≤1 year of follow-up were excluded.

Analyses were conducted using SAS/STAT, version 9.4 (SAS Institute, Cary, NC).

Results

Study Population

The final 4 study groups included 195,600 persons with and 2.6 million without exposure to colonoscopy at ages 45–49 years, and 660,248 persons with and 2.4 million without exposure to colonoscopy at ages 50–54 years. These groups contributed, respectively, 1.3, 15.8, 4.2, and 13.0 million person-years of follow-up. Table 1 shows demographic data for the 4 groups, including colonoscopy indications and interventions.

A plurality of exposure colonoscopies performed at ages 45–49 years were for gastrointestinal bleeding or anemia (31.6%), in contrast to only 10.2% at ages 50–54 years (Table 1). The majority (55.3%) of exposure colonoscopies performed at ages 50–54 years were performed for screening, in contrast to only 19.7% at ages 40–49 years (Table 1). Approximately one-half of exposure colonoscopies in each cohort included an intervention (Table 1).

Colorectal Cancer Incidence Rates

In the group with exposure to colonoscopy at ages 45–49 years, 276 persons developed CRC, compared with 4844 persons who developed CRC in the group without exposure to colonoscopy at ages 45–49 years (Table 2). CRC incidence in the group with exposure to colonoscopy was 20.8 (95% CI, 18.5–23.4) per 100,000 person-years, and in the group without exposure to colonoscopy, CRC incidence was nearly 50% higher at 30.6 (95% CI, 29.8–31.5) per 100,000 person-years (Table 2).

Table 2.

Colorectal Cancer Incidence in Persons With and Without Exposure Colonoscopy^a

	Ages 45–49 y		Ages 50–54 y
Variable	With colonoscopy	Without colonoscopy	With colonoscopy	Without colonoscopy
Overall
Total no. of people	195,600	2,570,184	660,248	2,431,744
Person-years of follow-up	1,329,676	15,806,958	4,197,869	13,027,088
No. of persons with CRC	276	4844	798	6757
Incidence rate of CRC per 100,000 person-years	20.8	30.6	19.0	51.9
95% CI	18.5–23.4	29.8–31.5	17.7–20.4	50.7–53.1
Sex
Women
Total no. of people	110,909	1,309,005	353,650	1,222,238
Person-years of follow-up	753,992	8,022,547	2,248,030	6,482,274
No. of persons with CRC	140	2312	428	3064
Incidence rate of CRC per 100,000 person-years	18.6	28.8	19.0	47.3
95% CI	15.7–1.9	27.7–30.0	17.3–20.9	45.6–49.0
Men
Total no. of people	84,691	1,261,179	306,598	1,209,506
Person-years of follow-up	575,684	7,784,411	1,949,839	6,544,814
No. of persons with CRC	136	2532	370	3693
Incidence rate of CRC per 100,000 person-years	23.6	32.5	19.0	56.4
95% CI	20.0–28.0	31.3–33.8	17.1–21.0	54.6–58.3
Race/ethnicity
White
Total no. of people	121,971	1,536,775	440,189	1,519,924
Person-years of follow-up	850,230	9,637,230	2,865,981	8,223,154
No. of persons with CRC	171	3057	539	4329
Incidence of CRC rate per 100,000 person-years	20.1	31.7	18.8	52.6
95% CI	17.3–23.4	30.6–32.9	17.3–20.5	51.1–54.2
African American
Total no. of people	28,719	426,567	88211	390,315
Person-years of follow-up	187,045	2,639,285	554,093	2,120,042
No. of persons with CRC	51	894	143	1276
Incidence rate of CRC per 100,000 person-years	27.3	33.9	25.8	60.2
95% CI	20.7–35.9	31.7–36.1	21.8–30.3	56.9–63.5
Hispanic
Total no. of people	33,370	447,430	92,333	382,634
Person-years of follow-up	214,758	2,583,534	529,294	1,951,672
No. of persons with CRC	39	645	85	860
Incidence rate of CRC per 100,000 person-years	18.2	25.0	16.1	44.1
95% CI	13.1–24.5	23.1–26.9	12.9–19.7	41.2–47.1
Other
Total no. of people	8660	124,375	29,483	106,369
Person-years of follow-up	57,991	735,302	183,303	559,242
No. of persons with CRC	12	196	22	229
Incidence rate of CRC per 100,000 person-years	20.7	26.7	12.0	40.9
95% CI	11.3–35.0	23.1–30.6	7.7–17.8	35.9–46.5
Missing race
Total no. of persons	2880	35,037	10,032	32,502

^aExposure colonoscopy was defined as colonoscopy with no subsequent CRC diagnosis within 6 months.

In the group with exposure to colonoscopy at ages 50–54 years, 798 persons developed CRC, with CRC incidence of 19.0 (95% CI, 17.7–20.4) per 100,000 person-years (Table 2). In the group without exposure to colonoscopy at ages 50–54 years, 6757 developed CRC, with CRC incidence of 51.9 (95% CI, 50.7–53.1) per 100,000 person-years, a rate that was more than double that observed in the group with exposure to colonoscopy (Table 2).

Colorectal Cancer Hazard Ratios

Figure 1 shows the proportions of persons free of CRC over time in the groups exposed vs not exposed to colonoscopy. The relatively linear declines in the proportions of persons free of CRC that were observed through approximately day 2000 after exposure to colonoscopy at ages 45–49 years and even longer at ages 50–54 years, and the progressively greater divergence over time in the curves of the unexposed vs exposed groups (Figure 1), suggest that delaying accrual of CRC cases and person-time for up to 3–4 years in these exposed groups in order to account for a possible healthy screenee bias would not substantially affect the results.

Figure 1.

Proportions of persons free of CRC over time in cohorts exposed vs not exposed to colonoscopy at ages 45–49 years (A) or 50–54 years (B). Adjusted Cox models extended to accommodate time-varying exposures showed hazard ratios of 0.50 (95% CI, 0.44–0.56; P < .0001) and 0.32 (95% CI, 0.29–0.34; P < .0001) for developing CRC in the cohorts exposed vs not exposed to colonoscopy at ages 45–49 years and 50–54 years, respectively.

Unadjusted Cox models showed hazard ratios of 0.62 (95% CI, 0.55-0.70; P < .0001) and 0.36 (95% CI, 0.33-0.38; P < .0001) for developing CRC in the groups with vs those without exposure to colonoscopy at ages 45–49 years and 50–54 years, respectively. After adjustment for age, the hazard ratios were 0.49 (95% CI, 0.43–0.56; P < .0001) and 0.32 (95% CI, 0.29–0.34; P < .0001), respectively. These hazard ratios were not affected by further adjustment for sex and race/ethnicity in addition to age: hazard ratio, 0.50 (95% CI, 0.44–0.56; P < .0001) and hazard ratio, 0.32 (95% CI, 0.29-0.34; P < .0001), respectively.

Analyses by Sex and Race/Ethnicity

CRC incidence rates in the groups without exposure to colonoscopy were higher in men than women, higher in African American persons than in White persons, and lower in Hispanic persons than in White persons (Table 2). The results in all strata showed differences in CRC incidence rates in the sex and race/ethnicity subgroups with vs without exposure to colonoscopy at ages 45–49 or 50–54 years that were similar to those observed for the groups overall (Table 2).

In women, the unadjusted hazard ratios for developing CRC in the groups with vs those without exposure to colonoscopy at ages 45–49 and 50–54 years were 0.59 (95% CI, 0.50–0.70) and 0.40 (95% CI, 0.36–0.44), respectively. In men, the respective hazard ratios were 0.67 (95% CI, 0.56–0.80) and 0.33 (95% CI, 0.29–0.36). After adjustment for age, the hazard ratios were 0.48 (95% CI, 0.40–0.58; P < .0001) and 0.35 (95% CI, 0.31–0.39; P < .0001), respectively, in women, and 0.52 (95% CI, 0.43–0.62; P < .0001) and 0.29 (95% CI, 0.26–0.32; P < .0001), respectively, in men. After further adjustment for race/ethnicity in addition to age, the hazard ratios were essentially unchanged: 0.48 (95% CI, 0.40–0.57; P < .0001) and 0.35 (95% CI, 0.31–0.39; P < .0001), respectively, in women, and 0.52 (95% CI, 0.43–0.62; P < .0001) and 0.29 (95% CI, 0.26–0.32; P < .0001), respectively, in men.

Analyses by Colonoscopy Indication

Table 3 shows CRC incidence rates by colonoscopy indication, all of which were lower than without exposure to colonoscopy (Table 2). CRC rates were higher in the group exposed to colonoscopy at ages 50–54 years than in the group exposed to colonoscopy at ages 40–49 years for most indications, with the exception of a screening indication (15.7; 95% CI, 14.1–17.5 vs 23.7; 95% CI, 18.4–30.2 per 100,000 person-years). In both groups, CRC rates were higher after colonoscopies performed for bleeding, anemia, or abnormal radiographic findings than for other symptoms (Table 3). CRC rates were also higher in both groups after colonoscopies with vs without intervention (Table 3).

Table 3.

Colorectal Cancer Incidence by Exposure Colonoscopy Indication and Type

Variable	Incidence rate of CRC per 100,000 person-years
	Ages 45–49 y		Ages 50–54 y
	Rate	95% CI	Rate	95% CI
Indication for exposure colonoscopya
Group 1: Gastrointestinal bleeding, occult blood in stools, or iron-deficiency anemia	22.9	18.6–27.8	30.5	25.9–35.6
Group 2: Abnormal finding in radiologic examination of gastrointestinal tract	26.7	14.2–46.2	37.8	23.8–57.2
Group 3: Weight loss, abdominal pain, or intestinal obstruction	14.6	10.1–20.6	15.7	11.5–20.8
Group 4: Change in bowel habits, diverticulitis or diverticulosis, constipation, diarrhea, irritable colon, or noninfectious gastroenteritis	17.2	13.2–22.2	18.3	15.9–21.0
Group 5: Screening, polyp or benign neoplasm of large intestine, or history of polyp	23.7	18.4–30.2	15.7	14.1–17.5
Group 6: Other	27.2	17.3–40.8	34.0	25.5–44.6
Type of index colonoscopyb
Colonoscopy without intervention	16.2	13.4–19.6	14.0	12.5–15.7
Colonoscopy with intervention	25.4	21.8–29.5	24.4	22.4–26.7
Comparator group without exposure to colonoscopy (see Table 2)
Without colonoscopy	30.6	29.8–31.5	51.9	50.7–53.1

NOTE. See Supplementary Table 4 for detailed definitions.

^aExposure colonoscopy was defined as colonoscopy with no subsequent CRC diagnosis within 6 months.

^bColonoscopy with and without intervention was determined by International Classification of Diseases, revisions 9 and 10^40,41 and Current Procedural Terminology³⁹ procedure codes (Supplementary Table 2).

Sensitivity Analyses

The 4 post-hoc sensitivity analyses addressing an exclusion period for accrual of cases in the cohorts exposed to colonoscopy, colonoscopies after the exposure period, persons with only a single visit with a diagnostic code for CRC, and persons who had follow-up of ≤1 year all resulted in hazard ratio estimates that did not differ substantially from those of the original results (Supplementary Material).

Discussion

Our population-based statewide study, with a diverse cohort, large sample size, and relatively long follow-up, suggests a strong inverse association between exposure to colonoscopy at ages 45–49 or 50–54 years and subsequent CRC incidence. These results address the pressing need for data on CRC outcomes after colonoscopy at ages 45–49 years in order to inform guidelines and clinical practice on the optimal age to begin CRC screening.^4,5

The 2018 ACS guideline is currently the only one to recommend lowering the CRC screening initiation age.^11,16,18,19 Review of the US Preventive Services Task Force guideline is underway.¹⁷ The ACS guideline acknowledged the limited evidence on the benefit of screening at ages 45–49 years.^46,47 The gFOBT trials that enrolled persons as young as age 45 years^48,49 were not designed to study the 45- to 49-year-old age group with adequate power. The US Multi-Society Task Force on Colorectal Cancer and National Comprehensive Cancer Network have not changed their guidance.^11,18,19

Several observations in our study are consistent with prior knowledge and thus provide confidence in our approach. The observed CRC rates and the patterns of relative CRC incidence between sexes and races/ethnicities (Table 2) are consistent with published data.^3,4 Blood in stool is associated with colorectal neoplasia, and advanced polyps requiring intervention at colonoscopy are associated with higher rates of subsequent CRC,^50,51 as seen in our groups exposed to colonoscopy with intervention (Table 3). The hazard ratios for incident CRC in persons exposed vs not exposed to colonoscopy at ages 50–54 years are consistent with the results of previous studies.^8,12-15 CRC incidence rates were lower after colonoscopy with any indication compared with no colonoscopy (Tables 2 and 3).²⁹ Our study design accounts for the unique challenges faced in observational studies of a test that can be both preventive of future disease and diagnostic of current disease.^30,31 Delaying accrual of CRC cases and person-time for up to 3–4 years in the cohorts exposed to colonoscopy to account for a possible healthy screenee bias does not appear to substantially affect the results (Figure 1, post-hoc analysis in the Supplementary Material).

The apparently greater protective association between colonoscopy and incident CRC observed in persons exposed to colonoscopy at ages 50–54 vs 45–49 years must be interpreted with great caution. We do not believe that this can be taken as evidence that colonoscopy at ages 45–49 years is not as effective at preventing CRC as colonoscopy at ages 50–54 years. It is more likely that the younger cohort exposed to colonoscopy was enriched in persons at higher than average risk for developing CRC. Indications for colonoscopy differed between the 2 groups, with more colonoscopies being performed for bleeding or anemia, and fewer for screening, in those exposed to colonoscopy at ages 45–49 years (Table 1). This is not surprising, given that our data are from years when screening was recommended starting at age 50 years. Furthermore, persons exposed to colonoscopy at ages 45–49 years may represent a group enriched in family history of CRC and cancer genetic syndromes, despite our attempt to exclude these syndromes. These persons have elevated CRC risk despite colonoscopy.^52-54 These selection biases would attenuate any observed protective effect from colonoscopy at ages 45–49 years, yet despite this, a protective association was still observed.

A high-quality colonoscopy that detects no neoplasia identifies persons who are at low CRC risk, but does not change natural history.^55-57 In contrast, colonoscopy with polypectomy can decrease CRC incidence.^8,12-15 The CRC rates observed in our study overall (Table 2) and after colonoscopy with or without intervention (Table 3), and the rates of intervention (Table 1), allow for an exploratory calculation of the benefit of intervention during colonoscopy, with the caveat above (see formula in the Supplementary Material). If those who underwent colonoscopy with intervention at ages 45–49 years had not undergone intervention, the expected CRC incidence can be estimated as 44.3 per 100,000 person-years. A comparison with the observed rate with intervention of 25.4 per 100,000 person-years (Table 3) suggests that intervention reduced CRC incidence by approximately 43%. For those who underwent colonoscopy with intervention at ages 50–54 years, the estimated expected vs observed rates were 89.5 per 100,000 person-years vs 24.4 per 100,000 person-years (Table 3), respectively, suggesting that intervention reduced CRC incidence by approximately 73%.

Our study has limitations. First, the groups exposed vs not exposed to colonoscopy may differ in ways that were not measured, including factors associated with subsequent colonoscopy beyond the initial exposure period. We had no information on CRC lifestyle risk factors, such as obesity or smoking. We would expect that selection for colonoscopy of younger persons at higher risk for CRC would attenuate the estimated protective effect from colonoscopy, which is what we observed. It is unlikely that in the study period, before the ACS recommendation, health-conscious 45- to 49-year-old persons at low risk of developing CRC self-selected for colonoscopy. Furthermore, our results for the 50- to 54-year age group are consistent with published studies, providing confidence in our methodological approach.^8,12-15 Second, although we limited our analysis to persons with residence in Florida in order to minimize loss to follow-up, we were unable to identify persons who moved out of Florida after inclusion. US Census data suggest that only about 2% of Florida residents move out of state every year.²⁰ We cannot exclude that, among some state residents, colonoscopy or CRC might have occurred outside of Florida. Third, no information was available on CRC stage and we could not ascertain CRC mortality, which future studies should address. One major strength is that we included all Florida residents regardless of insurance status or type, unlike studies employing insurance-based administrative databases that fail to capture the uninsured or lose patients on insurance change. Additional strengths are the inclusion of all colonoscopies regardless of indication,²⁹ and a sample size that is many times larger than those of previous studies of colonoscopy and CRC.^13,22,25,27 Multiple sensitivity analyses did not substantially affect the results or alter the conclusions.

We believe that our results can inform guidelines on the age to start CRC screening. Although randomized controlled trials provide the strongest evidence on which to base policy, current CRC screening guidelines, especially relating to colonoscopy, also consider high-quality observational data.^11,16,18,58 Flexible sigmoidoscopy and gFOBT decreased CRC incidence and mortality in randomized controlled trials,⁹ but these tests have been replaced by colonoscopy and the fecal immunochemical test in practice.¹⁰ The endorsements of screening colonoscopy and fecal immunochemical test^11,16,18 stand on the shoulders of the randomized controlled trials of sigmoidoscopy and gFOBT,⁹ and recognition that it is not feasible to perform large randomized trials of all CRC screening options.⁵⁹ It is questionable whether a randomized controlled trial of CRC screening at ages 45–49 years can be conducted, given the requirement for even larger sample sizes than any CRC screening trial to date, due to the lower event rate in younger people.^9,58,60 Our population-based observational study provides valuable information on the potential effect of screening that seeks to prevent cancer²⁹ in persons younger than 50 years.

We expect that data like ours will be supplemented by observational studies on lesion detection rates during screening at ages 45–49 years, followed by extrapolation to infer long-term outcomes. In addition, observational studies of screening colonoscopy at these ages should allow ascertainment and inclusion of CRCs diagnosed at first colonoscopy^30,32 and, with adequate follow-up, any impact on future CRC risk in the intervention arm. In such studies, an adequate nonscreening control group will need to be identified. As observational data accumulate, the societal decision about changing the screening initiation age will have to take into account the potential trade-offs and economic consequences, including the lower absolute rates of CRC at ages 45–49 years vs at ages 50–54 years (in this study, the absolute CRC rates were 30.6; 95% CI, 29.8–31.5 vs 51.9; 95% CI, 50.7–53.1 per 100,000 person-years in the groups not exposed to colonoscopy at ages 45–49 years vs 50–54 years). Our recent decision analysis suggests that starting CRC screening at age 45 years in the United States is likely to be highly cost-effective, but that, if resources are limited, greater benefits at lower costs could be achieved by improving screening rates in unscreened persons older than 50 years, or improving follow-up colonoscopy rates after abnormal fecal immunochemical test.⁶¹

In summary, we found a strong association between exposure to colonoscopy at ages 45–49 years and a substantial reduction in subsequent CRC incidence. Given that randomized controlled trials of CRC screening at ages 45–49 years may never be conducted, data like ours may provide the best available basis for assessing the potential benefits of beginning CRC screening at age 45 years and to inform future guidelines and clinical practice.

Estimation of the Effect of Colonoscopy With Intervention

Assuming that the cohorts exposed and not exposed to colonoscopy would have experienced the same CRC natural history if the exposed cohort had instead been unexposed, the expected CRC rate in the subgroup that underwent colonoscopy with intervention, had that subgroup not undergone such intervention, can be estimated based on the following relationship:

$$\begin{gathered} P_{colonoint}\ast CR{C}_{colonoprecursor}+ \\ {P}_{coloint}\ast CR{C}_{coloprecursor}=CR{C}_{nocolo} \end{gathered}$$

where ${\mathrm{P}}_{\text{colo}\text{no}\text{int}}$ = proportion of cohort with exposure to colonoscopy that underwent colonoscopy with no intervention; ${\mathrm{P}}_{\text{colo}\text{int}}$ = proportion of cohort with exposure to colonoscopy that underwent colonoscopy with intervention; ${\text{CRC}}_{\text{colo}\text{no}\text{precursor}}$ = observed CRC rate in subcohort with colonoscopy with no intervention, where it is assumed that no intervention means that no colorectal cancer precursor was found; ${\text{CRC}}_{\text{colo}\text{precursor}}$ = calculated CRC rate for the subcohort with colonoscopy with intervention, if the CRC precursor that was presumably removed had instead been left in place; ${\text{CRC}}_{\text{no}\text{colo}}$ = observed CRC rate in entire cohort without exposure to colonoscopy.

Results of Sensitivity Analyses

The 4 post-hoc sensitivity analyses resulted in estimates that did not differ substantially from those of the original results (unadjusted hazard ratios, 0.62; 95% CI, 0.55–0.70; P < .0001 and 0.36; 95% CI, 0.33–0.38; P < .0001) for developing CRC in the groups with vs those without exposure to colonoscopy at ages 45–49 years and 50–54 years, respectively).

First, instituting an exclusion period of 2 years before accrual of cases in the cohorts exposed to colonoscopy yielded unadjusted hazard ratios of 0.54 (95% CI, 0.47–0.61; P < .0001) and 0.31 (95% CI, 0.29–0.34; P < .0001) for developing CRC in the groups with vs those without exposure to colonoscopy at ages 45–49 years and 50–54 years, respectively.

Second, the numbers (percent) of persons in each cohort who underwent colonoscopy after the exposure period were 44,898 (23.0%) among those exposed and 255,052 (10.5%) among those unexposed to colonoscopy at ages 45–49 years, and 126,067 (19.1%) among those exposed and 141,255 (7.0%) among those unexposed to colonoscopy at ages 50–54 years. Exclusion of these persons resulted in unadjusted hazard ratios of 0.67 (95% CI, 0.59–0.76; P < .0001) and 0.40 (95% CI, 0.37–0.44; P < .0001) for developing CRC in the groups with vs those without exposure to colonoscopy at ages 45–49 years and 50–54 years, respectively.

Third, among those with a diagnostic code for CRC, the numbers (percent) of persons in each cohort who had only a single initial visit with such a code, and no subsequent visits with a diagnostic code for CRC or history of CRC, were 74 of 276 (27%) among those exposed and 932 of 4844 (19.2%) among those unexposed to colonoscopy at ages 45–49 years, and 200 of 798 (25%) among those exposed and 1347 of 6757 (19.9%) among those unexposed to colonoscopy at ages 50–54 years. Exclusion of these persons resulted in unadjusted hazard ratios of 0.58 (95% CI, 0.50–0.68; P < .0001) and 0.34 (95% CI, 0.31–0.37; P < .0001) for developing CRC in the groups with vs those without exposure to colonoscopy at ages 45–49 and 50–54 years, respectively.

Fourth, the numbers (percent) of persons in each cohort who had follow-up ≤1 year were 2122 (1.1%) among those exposed and 49,523 (1.9%) among those unexposed to colonoscopy at ages 45–49 years, and 7224 (1.1%) among those exposed and 166,817 (6.9%) among those unexposed to colonoscopy at ages 50–54 years. Exclusion of these persons resulted in unadjusted hazard ratios of 0.60 (95% CI, 0.52–0.70; P < .0001) and 0.34 (95% CI, 0.31–0.38; P < .0001) for developing CRC in the groups with vs those without exposure to colonoscopy at ages 45–49 years and 50–54 years, respectively.

Supplementary Material

Supplementary Table 1

Supplementary Table 1. Identification of Exclusions: Diagnosis Codes and Description

Supplementary Table 2

Supplementary Table 2. International Classification of Diseases 9th and 10th Revisions,^40,41 Current Procedural Terminology,³⁹ and Healthcare Common Procedure Coding System⁴² Codes for Identification and Characterization of Colonoscopy

Supplementary Table 3

Supplementary Table 3. International Classification of Diseases 9th and 10th Revisions^40,41 Diagnostic Codes Associated With Colonoscopy

Supplementary Table 4

Supplementary Table 4. International Classification of Diseases 9th and 10th Revisions^40,41 Diagnostic Codes to Identify Colorectal Cancer

Note: To access the supplementary material accompanying this article, visit the online version of Gastroenterology at www.gastrojournal.org, and at http://doi.org/10.1053/j.gastro.2021.02.015.

WHAT YOU NEED TO KNOW:

BACKGROUND AND CONTEXT

Colorectal cancer (CRC) rates are increasing at ages <50 years. It has been recommended to lower the CRC screening age to 45 years, but evidence is lacking about the effectiveness of CRC screening at ages 45-49 years.

NEW FINDINGS

Colonoscopy at ages 45-49 years was associated with a 50% decrease in the risk of subsequent CRC. The finding that colonoscopy at ages 50-54 years was associated with a 68% decrease in the risk of subsequent CRC is consistent with published literature, providing confidence in the study’s methods.

LIMITATIONS

This was not a randomized controlled trial of screening at ages 45-49 years, but such a trial is unlikely to ever be undertaken.

IMPACT

These results can be used by groups that draft CRC screening guidelines and policy as they grapple with decisions regarding the optimal age to start average-risk CRC screening.

Abbreviations used in this paper:

ACS

American Cancer Society

CI

confidence interval

gFOBT

guaiac fecal occult blood testing

HCUP

Healthcare Cost and Utilization Project