Risk of Colorectal Cancer and Colorectal Cancer Mortality Beginning Ten Years after a Negative Colonoscopy, among Screen-Eligible Adults 76 to 85 Years Old

This article is featured in Highlights of This Issue, p. 1

Age is a strong predictor of colorectal cancer risk and mortality (1), but other characteristics could bear on the decision of older adults to undergo screening for colorectal cancer. Recent recommendations from the United States Preventive Services Task Force (USPSTF), American Cancer Society, and the Multi-Society Task Force advise that, for people ages 76 to 85 years, the decision should be shared between clinicians and patients (2–4). The guidelines further suggest consideration of prior screening history and overall health (3, 4) because the degree to which a person may benefit from screening depends on both their risk of colorectal cancer and life expectancy over which the benefit may accrue (5–8). Understanding for whom continued colorectal cancer screening could be of great or only marginal benefit (because they have a high or low risk of colorectal cancer, respectively) may inform patient and provider decision-making and more targeted test offerings to older adults, which are key components of the USPSTF's new recommendations for colorectal cancer screening in adults 76 to 85 years old (4).

However, there is great uncertainty about the value of continued screening beyond age 75 (9, 10) and few data are available to inform the decision. Prior studies have shown a low incidence of colorectal cancer in the years immediately following a negative colonoscopy (11) or sigmoidoscopy (12), particularly for persons ages 50 to 75 years. Existing recommendations for older adults are based primarily on microsimulation models of colorectal cancer incidence, colonoscopy harms, and life expectancy (13). This evidence gap leaves age as a default (and sometimes preferred) determinant (14), which ignores tremendous variability in biological age and health status (7, 15) among adults ages 76 to 85 years and may obscure differences in risk of colorectal cancer and colorectal cancer-related mortality, especially based on screening history. It is unknown how these differences may translate to differences in the cumulative risk of colorectal cancer incidence and mortality.

Using a large multi-healthcare system dataset of patients in Washington and California, we provide absolute estimates of colorectal cancer incidence and mortality among people ages 76 to 85 years 10 years after a negative colonoscopy, that is, at the point in time when they might consider screening again.

We conducted a retrospective cohort study among older adult members of three Kaiser Permanente integrated healthcare systems: Northern California (KPNC), Southern California (KPSC), and Washington (KPWA). These healthcare systems contribute to the Optimizing Colorectal Cancer Screening PREcision and Outcomes in CommunIty-baSEd Populations (PRECISE) Research Center, part of the NCI-funded Population-based Research to Optimize the Screening Process (PROSPR II) consortium (16). Details of these healthcare systems, particularly with respect to the colorectal cancer screening process, have been previously published (16). Cumulatively, these populations include more than 1 of every 70 persons of the total United States population and are generally representative of the census demographics of their regional populations (17–20).

All three integrated healthcare systems include older persons. Compared with the broader United States population ages 65 and older, the study population had similar levels of Medicare coverage (∼90%) and greater racial and ethnic diversity. Further details about the PRECISE cohort have been published elsewhere (16, 21). The study was conducted in accordance with the U.S. Common Rule. A waiver of consent was granted and the study was approved by Institutional Review Boards (IRB) at the study sites and the University of Washington.

The study population included patients from the PRECISE cohort who had previously undergone a colonoscopy (for any indication) and were eligible for colorectal cancer screening (based on the eligibility criteria enumerated below) sometime between ages 76 to 85 years (Supplementary Fig. S1). The population of interest comprised patients older than 75 years, the upper recommended screening age for universal screening, but younger than age 86 years, when guidelines strongly suggest cessation of screening. Between these age cutoffs, patients are recommended to engage in shared decision-making with their clinician about whether to continue screening (4). Patients were included in the study population if they had a negative colonoscopy (i.e., no adenoma or cancer detected) between January 1, 2000, and December 31, 2009, when they were 66 to 75 years old, and if they were continuously enrolled and alive 10 years following that colonoscopy. The time point 10 years after colonoscopy was considered to be the study's index date (i.e., the time point at which the person might consider screening again). To ensure a screening-eligible population, patients with a history of inflammatory bowel disease (IBD), colorectal cancer, or gastrointestinal surgery prior to their index date were excluded. Patients with colorectal cancer testing between their negative colonoscopy and index date (e.g., lower endoscopy or any positive fecal test during the ten years prior to their index date, or any fecal test during the one year prior to their index date) were also excluded, because these individuals would have been up-to-date or otherwise ineligible for screening at index date. Patients were considered ineligible for screening if they had a CT colonography, barium enema, or abdominal CT during the 180 days prior to their index date, because these procedures could be used to detect colorectal cancer, and thus anyone with a recent procedure would not be eligible for screening in the near term. Patients were excluded if they had no recorded healthcare encounters (acute inpatient stay, institutional stay, or primary care visit) with the healthcare system during the 365 days prior to the index date, which served as a measure of an interaction with a healthcare provider at which symptoms, if present, would have been reported. This exclusion also improves the likelihood of up-to-date coding of other relevant conditions for analysis (e.g., for calculation of comorbidity index described in the “Measures” section below).

Finally, patients with any relevant symptoms (i.e., abdominal pain, iron deficiency or unspecified anemias, gastrointestinal bleeding or blood in stools, diarrhea, weight loss or underweight, diverticulitis, constipation, abdominal mass, or change in bowel habits) during the 180 days prior to their index date were also excluded because the presence of these signs or symptoms makes patients ineligible for screening, which, by definition, occurs in patients without possible signs or symptoms of cancer.

The PRECISE Research Center collected information on patient demographics and clinical characteristics, colorectal cancer screening process data (i.e., risk factors, screening tests, diagnostic evaluations, treatment procedures, and outcomes), colorectal cancer screening history prior to cohort entry, cancer diagnoses, and cancer deaths. Data on colorectal cancer tests, patient demographics, and comorbidities diagnosed during cohort eligibility were obtained from administrative and clinical databases including electronic health records. Patient data regarding prior testing history and conditions were also collected from these sources, with look-back period depending on site constraints: KPWA extended back to January 1, 1993, whereas KPNC and KPSC looked back to January 1, 2000, for enrollment, visits, and colorectal cancer tests, and extended further back (to the full extent of information available in electronic records databases) for colorectal cancer diagnoses and gastrointestinal surgeries. Colorectal cancer diagnoses were obtained from local and central cancer registries: Seattle – Puget Sound Surveillance, Epidemiology and End Results (SEER) registry for KPWA, and the facility-based cancer registries at KPNC and KPSC, which report to the State of California Cancer Registry. Information on patient deaths was sourced from state vital records data, as well as a variety of internal sources (e.g., insurance membership, discharge status on claims, etc.). Colorectal cancer-related deaths were ascertained using each site's state death records as the primary source for cause of death. In the rare case where a patient was recorded with a colorectal cancer-related death but without a colorectal cancer diagnosis date, the date of diagnosis was imputed as the date of colorectal cancer-related death.

For procedures occurring after the index date, colonoscopy indication was assigned based on manual chart review and natural language processing or using a modified version of a colonoscopy indication algorithm that incorporates administrative and clinical data (22) and considers data elements related to recent procedures, IBD, signs and symptoms, past findings of colorectal cancer screening procedures, and personal history of colorectal cancer.

The Charlson Comorbidity Index (CCI) was used to measure comorbidity burden and calculated by applying modified Charlson/Deyo comorbidity algorithm weights (23) to an updated set of diagnosis and procedure codes associated with both inpatient and outpatient visits. CCI was assessed as of the last day of the calendar quarter prior to the index date, using a look-back period of 365 days for relevant codes, and categorized (0, 1, 2, 3, 4, ≥5) for analysis. Sensitivity analyses stratified patients based on the individual comorbid conditions that contributed to the CCI. Other covariates included age at index date, sex, ethnicity and race, and recent encounters with the healthcare system, which were classified as primary care or inpatient stays (both acute and institutional) and used as a supplementary proxy for health status, along with CCI.

Descriptive statistics were calculated to characterize the cohort demographics. Cumulative incidence functions (CIF) were used to estimate cumulative incidence for two outcomes: (i) colorectal cancer incidence and (2) colorectal cancer mortality in the years following the index date, with estimates at 2, 5, and 8 years post-index date. Follow-up began at index date (i.e., 10 years after a prior negative colonoscopy where no adenoma or colorectal cancer was detected; Fig. 1) and continued until diagnosis of colorectal cancer or death from colorectal cancer, aging out of cohort eligibility (96th birthday), disenrollment from health plan, moving out of SEER coverage area, end of cohort follow-up on December 31, 2019, or 180 days after a subsequent screening colonoscopy (mortality and incidence analyses) or fecal test (mortality analysis only). Follow-up was censored for these reasons so the estimates describe colorectal cancer incidence and mortality beginning 10 years after a negative colonoscopy in the absence of screening interventions that may have affected the risk of the outcome. The exact date of censoring was delayed to 180 days after a screening colonoscopy to avoid the exclusion of cancers detected by the procedure. Follow-up was not censored after a colonoscopy with a diagnostic indication since diagnostic colonoscopy is routine clinical care when indicated, rather than a preventative health measure; thus, a decision about whether to screen would have no impact on the occurrence of a diagnostic procedure. Follow-up for colorectal cancer mortality was additionally censored at 180 days after a fecal test (fecal immunochemical test or fecal occult blood test) because the test reduces risk of colorectal cancer mortality (24–26); a 180-day censoring delay avoids excluding colorectal cancer deaths.

A Fine-Gray subdistribution hazard model was used to simultaneously estimate the absolute risk of colorectal cancer outcome (incidence or mortality) and non-colorectal cancer mortality (27). Deaths from causes other than colorectal cancer were counted as competing events (28, 29). The more common method (30, 31) for estimating cumulative incidence of events over time—one minus the Kaplan–Meier (KM) estimate of the survival function—censors follow-up at death from another cause, which means deceased participants are assumed to have the same risk of the outcome (colorectal cancer) as the alive individuals remaining in the population. This approach thus misrepresents the incidence of colorectal cancer among older adults remaining alive. To demonstrate the effect of this difference in calculation, cumulative mortality and incidence for the total population were also computed using a Kaplan–Meier approach as a secondary analysis. CIF incidence and mortality were stratified by patient demographic and health status variables: age at index date (in years); sex (male vs. female); CCI (0, 1, 2, 3, 4, ≥5); number of primary care encounters or inpatient stays (both acute and institutional) during the 180 days prior to index date.

A sensitivity analysis was also used to estimate cumulative colorectal cancer incidence and mortality without censoring after screening colonoscopy, because retrospective classification of colonoscopy indication based on electronic data is known to be imperfect (22) and censoring at a screening colonoscopy could lead to an inaccurate estimate of colorectal cancer risk if persons who have a screening colonoscopy after age 75 years are not representative of colorectal cancer risks and lifespan.

The data generated in this manuscript are not publicly available due to IRB restrictions to protect patient privacy and consent. Processes for accessing PROSPR data are described at: https://healthcaredelivery.cancer.gov/prospr/datashare/.

Characteristics of the 25,974 screen-eligible patients ages 76 to 85 years at their index date (i.e., with a negative colonoscopy ten years prior) are shown in Table 1. The study population skewed slightly towards younger ages in that range (54.7% ages 76–80 years vs. 45.3% ages 81–85 years). Cohort members were primarily identified as non-Hispanic White (66.2%), Hispanic (13.4%), non-Hispanic Asian (11.3%), and non-Hispanic Black (9.1%). Other racial groups comprised less than 2% among those with a recorded racial identity. There was a higher proportion of females (58.0%) than males (42.0%). In general, the cohort was highly heterogenous in terms of comorbidity burden: nearly one quarter had no major comorbidities (CCI = 0) and more than half (54.1%) had a CCI score of two or more. The three most prevalent comorbid conditions at the index date were peripheral vascular disease (37.8%), renal disease (29.3%), and diabetes (24.3%). Characteristics of the study population stratified by healthcare system are shown in Supplementary Table S1. We observed slight differences in the distribution of characteristics by healthcare system, with greater racial diversity at KPNC and KPSC and lower comorbidity scores at KPWA (Supplementary Table S1).

Overall, cumulative colorectal cancer incidence and mortality were low in this population. Cumulative incidence of colorectal cancer was estimated as 0.39% (95% CI, 0.31%–0.48%) 2 years after index date and 1.29% (95% CI, 1.02%–1.61%) 8 years after index date (Table 2). Cumulative colorectal cancer mortality was an order of magnitude lower: 0.04% (95% CI, 0.02%–0.08%) 2 years after index date and 0.46% (95% CI, 0.30%–0.70%) 8 years after index date (Table 3). Sensitivity analyses found that these estimates did not change if subsequent screening colonoscopy was not treated as a censoring event (Supplementary Tables S2 and S3). Estimates for cumulative colorectal cancer incidence and mortality are provided in Tables 2 and 3, respectively; corresponding estimates of cumulative mortality from other causes are provided in Supplementary Table S4. Cumulative incidence curves that generated overall estimates are illustrated in Fig. 2, with additional stratified curves provided in Supplementary Fig. S2.

Estimates of cumulative risk of death from non-colorectal cancer causes were 100-fold higher than their corresponding colorectal cancer mortality estimates. Cumulative mortality from non-colorectal cancer causes was estimated as 8.24% (95% CI, 7.83%–8.66%) 2 years after index date and 41.45% (95% CI, 38.74%–43.16%) 8 years after index date (Supplementary Table S4).

Point estimates for cumulative colorectal cancer incidence and mortality were higher among males compared with females and colorectal cancer mortality was higher among 81- to 85-year-olds compared with 76- to 80-year-olds (Fig. 2). Strata defined by both characteristics showed even more variation across time points (Supplementary Figs. S2A and S2E). At 2 years following index date, the highest incidence was observed among males ages 81 to 85 years and lowest among females ages 76 to 80 years.

CCI scores did not have a clear association with cumulative risk of colorectal cancer or colorectal cancer mortality (Tables 2 and 3). Estimates for specific comorbid conditions are shown in Supplementary Tables S2 and S3. Individuals with diabetes appeared to have an elevated risk of colorectal cancer mortality (0.12% 2-year risk; 95% CI, 0.05%–0.27%) compared with overall study population risks (0.04% 2-year risk).

At 8 years following index date, patients with no encounters during 180 days prior to index date had nearly equivalent colorectal cancer risk to patients with acute inpatient or institutional stays prior to index date (Table 2). For colorectal cancer mortality, patients with no encounters had higher risk at 8 years (0.81%; 95% CI, 0.31%–1.79%) than patients with any recent encounters (Table 3). Meanwhile, the corresponding risk of death from other causes at the same time point was much higher for patients with acute inpatient or institutional stays compared with no encounters (62.83% vs. 40.24%; Supplementary Table S4).

Colorectal cancer risk was not appreciably different across race and ethnicity groups. Eight-year cumulative risk estimates for colorectal cancer incidence were 0.91% (95% CI, 0.55%–1.43%) for Hispanic patients; 1.34% (1.00%–1.75%) for non-Hispanic White patients; 1.60% (0.62%–3.43%) for non-Hispanic Black patients; and 1.09% (0.55%–1.96%) for non-Hispanic Asian patients (Table 2). Eight-year cumulative risk estimates for colorectal cancer mortality were 0.33% (95% CI, 0.10%–0.89%) for Hispanic patients; 0.44% (0.25%–0.73%) for non-Hispanic White patients; 0.77% (0.21%–2.13%) for non-Hispanic Black patients; and 0.56% (0.11%–1.98%) for non-Hispanic Asian patients (Table 3). Descriptive estimates were not robust for patients who identified as Native American/Alaska Native or Native Hawaiian/Other Pacific Islander, due to small numbers (≤0.5% study population).

In this study, across three integrated healthcare systems in the western United States, adults ages 76 to 85 years who had a negative colonoscopy 10 years earlier and were eligible for screening were at low risk of being diagnosed with colorectal cancer and of dying from this disease. These low 8-year colorectal cancer incidence and mortality risk estimates occurred in the context of much higher cumulative risks of death from other causes.

These data substantially extend existing information on colorectal cancer incidence and mortality by incorporating screening history into estimates, as well as stratifying by health status. SEER estimates for nearly the same age group (75–84 years) are higher: [cumulative incidence of colorectal cancer] 0.58% (2-year) and 2.28% (8-year); [cumulative mortality from colorectal cancer] 0.20% (2-year) and 0.78% (8-year; ref. 32). However, SEER estimates reflect the risks of a population that is heterogeneous and, on average, higher risk than this study population, because SEER includes individuals who may have less-than-adequate screening histories (e.g., none, or longer than ten years ago) and/or prior positive findings on a colorectal cancer test.

Importantly, these current estimates account for a critical selection factor that affects most disease risk estimates in older adults: death from other causes. The current analysis acknowledges that death from causes other than colorectal cancer is a significant consideration for the assessment of disease risk among older adults. Without accounting for risk of death from another cause [i.e., by using one minus the Kaplan–Meier estimator (1−KM) and censoring patients at time of death], risk estimates of colorectal cancer incidence and mortality were higher: 1.58% vs. 1.29% and 0.60% vs. 0.46%, respectively, 8 years after index date. Accounting for deaths from other causes in the generation of cumulative risk estimates thus implicitly acknowledges that lower life expectancy ultimately affects cumulative risk of colorectal cancer. Because discussions of individual life expectancy are difficult during shared decision-making conversations between older adults and their care providers (33), these cumulative risk figures may be a useful population-level alternative.

Estimates of cumulative incidence and mortality differed little when stratified by a limited set of ethnicity and race identities due to wide confidence intervals, and caution in their interpretation is warranted. Although some participants reported belonging to more than one race category, the categories were limited and may not capture the full diversity of the population. Furthermore, although this study cohort of 76 to 85 years old has a higher representation of Black (9.5% vs. 8.8%) and Asian (11.8% vs. 4.0%%) patients than the general United States population (34), representation was still insufficient to provide stratified estimates for other listed race categories. We also caution that stratifying a disease-specific risk estimate by race may exacerbate inequality if a group with higher cumulative risk of other causes of death has lower risk of colorectal cancer and thus is denied screening opportunities without confronting the underlying disparities driving the competing risk of death.

There are several strengths and limitations of this study. Strengths include the high-quality long-term data (including prior testing history) and use of a large and diverse screen-eligible cohort. However, the estimates presented in this analysis are not individual-level risk prediction (as the ability to gauge person-level life expectancy was quite limited) but rather provide empirical evidence of population-level colorectal cancer risk among previously tested older adults. These estimates are stratified by univariate characteristics, one or two at a time; individuals are, of course, much more complex. Furthermore, the estimates are specific to a population of individuals with negative colonoscopy 10 years ago and do not include patients undergoing surveillance for a polyp or other findings of concern. The risks estimated in this study should be considered in the context of the described population, and any application to a new population setting should evaluate both differences in background rates of colorectal cancer as well as population characteristics. Another limitation is that despite efforts made to include a diverse cohort, absolute number of events in ethnicity and race strata were insufficient to produce robust estimates stratified by ethnicity and race.

In summary, this study documents that adults ages 76 to 85 years who had a negative colonoscopy 10 years earlier were at low risk for colorectal cancer and colorectal cancer-related mortality during the ensuing 8 years. Knowledge of these results could well have a bearing on older adults’ decision to undergo or not undergo further colorectal cancer screening, including choice of modality, should they decide to continue screening.

R.A. Ziebell reports grants from NCI during the conduct of the study. A. Kamineni reports grants from NCI during the conduct of the study. D.A. Corley reports grants from NCI during the conduct of the study. B.B. Green reports grants from NCI during the conduct of the study; and was on the screening committee for the National Colorectal Cancer Round Table, which is sponsored by the American Cancer Society and the Centers for Disease and Infection Control; The Round Table paid for travel to Washington, DC, on August 11, 2022, for an in person meeting, B.B. Green reports receiving no other compensation for this position. T. Levin reports grants from NCI during the conduct of the study; grants from Freenome, Inc. outside the submitted work. J.E. Schottinger reports grants from NIH during the conduct of the study; grants from NIH outside the submitted work. J. Chubak reports grants from NCI during the conduct of the study; grants from Amgen, Inc. outside the submitted work. No disclosures were reported by the other authors.

R.R. Dalmat: Conceptualization, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. R.A. Ziebell: Data curation, validation, investigation, writing–review and editing. A. Kamineni: Conceptualization, resources, funding acquisition, methodology, project administration, writing–review and editing. A.I. Phipps: Conceptualization, supervision, methodology, writing–review and editing. N.S. Weiss: Conceptualization, supervision, methodology, writing–review and editing. E.S. Breslau: Resources, writing–review and editing. D.A. Corley: Resources, funding acquisition, project administration, writing–review and editing. B.B. Green: Writing–review and editing. E.A. Halm: Resources, funding acquisition, project administration, writing–review and editing. T. Levin: Writing–review and editing. J.E. Schottinger: Resources, funding acquisition, project administration, writing–review and editing. J. Chubak: Conceptualization, resources, supervision, funding acquisition, methodology, project administration, writing–review and editing.

J. Chubak, D.A. Corley, E.A. Halm, A. Kamineni, and J.E. Schottinger received funding from the NCI at the NIH (grant no.: UM1CA222035). This manuscript was written as part of the Population-based Research to Optimize the Screening Process (PROSPR II) Consortium. The overall aim of PROSPR II is to conduct multisite, coordinated, transdisciplinary research to evaluate and improve cervical, colorectal, and lung cancer screening processes. This work was also supported by grant no. T32CA09168 from the NCI. The views expressed here are those of the authors only and do not necessarily represent the views of the NCI or NIH.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.