Type 2 Diabetes and Risk of Early-Onset Colorectal Cancer

Background and Aims Early-onset colorectal cancer (CRC) is increasing in many developed countries. Type 2 diabetes mellitus has increased substantially in younger adults; however, its role in early-onset CRC remains unidentified. Methods We conducted a claims-based nested case-control study using IBM MarketScan Commercial Database (2006–2015). Incident early-onset CRC diagnosed at ages 18–49 was identified by the International Classification of Diseases, ninth Revision, Clinical Modification diagnosis code, and the first coded diagnostic pathology date was assigned as the index date. Controls were frequency matched with cases. Type 2 diabetes, stratified by severity, was identified through International Classification of Diseases, ninth Revision, Clinical Modification using the Klabunde algorithm. Multivariable logistic regressions were used to estimate odds ratios (ORs) and 95% confidence intervals (CIs). Results A total of 6001 early-onset CRC and 52,104 controls were included. Type 2 diabetes was associated with an increased risk of early-onset CRC (5.0% in cases vs 3.7% in controls; OR = 1.24, 95% CI: 1.09–1.41). The positive association was more pronounced for uncontrolled (OR = 1.37; 95% CI: 1.12–1.67) or complicated (OR = 1.59, 95% CI: 1.08–2.35) type 2 diabetes compared with controlled diabetes (OR = 1.13, 95% CI: 0.94–1.36). Conclusion Individuals with type 2 diabetes have a higher risk of early-onset CRC, with stronger associations for uncontrolled diabetes and complicated diabetes. The rising prevalence of type 2 diabetes among younger adults may partially contribute to the increasing incidence of early-onset CRC.


Introduction
C olorectal cancer (CRC) is the second most common cause of cancer-related death in the United States. 1 Because of increased CRC screening, the incidence and mortality rates of CRC have declined for several decades among adults aged 50 years and older.In contrast, the incidence and mortality of early-onset CRC (diagnosis before age 50 years) have been increasing since the mid-1990s. 2During 2012-2016, the incidence of proximal colon, distal colon, and rectal cancer all rose at 1.8% annually among adults younger than 50 years. 2Such an alarming increase in early-onset CRC contributed to a 6-year drop in the median age of CRC diagnosis, from 72 years during 1988-1989 to 66 years during 2015-2016. 2Further elucidation of risk factors that contributed to this increase is pivotal.
Thus far, obesity 3 and sedentary lifestyle 4 are among the potential contributors to the rise in early-onset CRC, pointing to a possible role of insulin dysregulation.6][7] Although the association between type 2 diabetes and average to late-onset CRC is established, 8 the emerging molecular characteristics of earlyonset CRC 9 support the necessity to revisit such association in a younger population.The link between early-onset CRC with obesity and prolonged sitting, both of which are risk factors for type 2 diabetes, 10,11 further lends support to this need.In addition, type 2 diabetes, if etiologically relevant to early-onset CRC, likely contributes to the rising incidence of early-onset CRC because of the paralleled increase of type 2 diabetes [12][13][14] and the rise in early-onset CRC. 15 Specifically in the United States, between 1988 and 2012, type 2 diabetes has increased from 2.7% to 4.5% for ages 20-44 years and from 13.3% to 16.2% among ages 45-64. 13In addition, when compared with older adults with type 2 diabetes, type 2 diabetes before age 45 appears to be a more aggressive disease with an increased risk of requiring insulin. 16Therefore, investigating the role of type 2 diabetes in early-onset CRC in a population-based study will likely generate significant insights into the etiology, prevention, and early detection of early-onset CRC.
To address these critical knowledge gaps, we used the IBM MarketScan Commercial Database (2006-2015), a longitudinal database that contains individual-level commercial health insurance claims data from over 113 million individuals from all geographic areas of the United States, to comprehensively examine the association between type 2 diabetes and risk of early-onset CRC.

Study Population
We conducted a nested case-control study of early-onset CRC using the MarketScan Database (2006-2015), a longitudinal, deidentified, individual-level health care claims database which comprised more than 113 million commercially insured U.S. adults younger than age of 65 years. 17The database captures information on outpatient and inpatient insurancereimbursable services, prescription data, type of health plan, and demographic information.Compared with nonelderly people with employer-sponsored insurance, MarketScan enrollees have a similar age and sex distribution. 18This study was exempt from institutional review board approval for its deidentified limited data set analysis.All authors had access to the study data and approved the final manuscript.

Ascertainment of Cases and Controls
All adults with an incident diagnosis of CRC between ages 18 and 49 years were considered as early-onset CRC and were identified by an International Classification of Diseases, ninth Revision, Clinical Modification (ICD-9-CM) diagnosis code (153.0-153.9,154.0, 154.1, and 154.8).To reduce false positives, we included only pathology-coded CRC cases and assigned the first diagnostic pathology date as the index date.As confirmed pathology diagnoses are automated into pathology ICD-9-CM diagnosis codes and are considered accurate in reporting pathology findings, 19 additional codes to confirm pathology were not required.We restricted our analyses to adults with at least 2 years of enrollment before the index dates, as well as 90 days of enrollment after the index dates to derive metastatic status.By requiring 2 years of data before index CRC and up to 90 days of data after, the index dates included were between  20 We excluded patients with CRC and with any prior/concurrent cancer history (V10.x) or genetic susceptibility to malignant neoplasm (V84.0x), as well as patients with CRC and with any cancer except nonmelanoma skin cancer identified through Healthcare Cost and Utilization Project's Clinical Classification Software (HCUP CCS) 21 within 2 years before the index dates.The identification of CRC as cases or the other cancers for exclusion for both cases and controls was based on the study by Klabunde et al, 22 which requires at least one inpatient facility claim and/or 2 outpatient or provider claims 31-365 days apart.
Controls without CRC were frequency matched with cases by up to an 8:1 ratio based on age (18-24 and every 5 years thereafter), sex (female, male), geographical region (Northeast, North Central, South, West, unknown) due to geographic variations in the incidence of early-onset CRC, 23 duration of insurance enrollment before index diagnosis (years), and prescription drug coverage (yes, no).Controls were selected to ensure that the distribution of the control index dates matched the distribution of index dates among the cases to account for changes over time.Controls also had at least 2 years of enrollment before the assigned random index dates and were selected to match the year of the corresponding case's index date.Controls with genetic susceptibility and personal cancer history were also excluded.

Ascertainment of Type 2 Diabetes Mellitus
To reduce bias due to increased detection of diabetes and other comorbidities for CRC cases during the workup period but not among controls, 24 we restricted our exposures and covariates from 91 days to 2 years before the index dates.Two years before the index dates were chosen to maximize statistical power.2 which has been used to identify type 2 diabetes in claims data. 26Type 2 diabetes was further classified as controlled/not stated as uncontrolled without complications (with 250.00, 250.10, 250.20, 250.30 for all the encounters), uncontrolled without complications (with 250.02, 250.12, 250.22,250.32 in any of the encounters), 27

Assessment of Covariates and Other Clinical Information
We extracted demographic information, including employment status, urban/rural residence, geographical region, and health plan, and derived the Charlson comorbidity index. 28In addition, the Charlson comorbidity index (without diabetes and cancer except nonmelanoma skin cancer) was calculated for both cases and controls.We extracted information on potential confounders, including inflammatory bowel diseases (IBDs), obesity, and family history of gastrointestinal cancer between 91 days and 2 years before the index dates.Information on fecal occult blood testing (FOBT) and screening/other colonoscopies during the same time period was retrieved.We also obtained information on a list of prespecified early signs/ symptoms for CRC, including gastrointestinal bleeding, abdominal pain, anemia, change of bowel habits, diarrhea, constipation, and weight loss between 91 days and 2 years before the index dates. 29

Statistical Analysis
To evaluate the association between type 2 diabetes and the risk of early-onset CRC, multivariable logistic regression models were used to estimate odds ratios (ORs) and 95% confidence intervals (CIs).We first adjusted for matching factors 30 including age (years), sex, duration of insurance enrollment (number of completed years of enrollment), geographical region (Northeast, North Central, South, West, unknown), and prescription drug coverage before the index dates.We then additionally adjusted for full-time employment status, residence (urban, rural, unknown), health plan (preferred provider organization, health maintenance organization, others), Charlson comorbidity index (continuous), IBD, obesity, family history of gastrointestinal cancer, screening colonoscopy, other colonoscopy, and FOBT.Individuals without geographic region or residence (<2%) were included in the adjustment using missing indicators.To examine the robustness and generalizability of these findings to an asymptomatic population, we conducted sensitivity analyses by restricting to participants without the following: IBD, prior colonoscopy/FOBT, or prespecified early signs/symptoms of CRC.We conducted sensitivity analyses by restricting to diabetes from 1 year up to 2 years before the index dates.We also examined whether the association differed according to the severity of type 2 diabetes (controlled, uncontrolled, and complicated).
We further examined if the association between type 2 diabetes and early-onset CRC differed as per the tumor anatomical site (colon [proximal colon, distal colon, unspecified colon], rectum).We also conducted stratified analyses to evaluate the association among subgroups, including sex, age at the index date (18-45 vs 46-49 years), birth year ( 1965 vs >1965), and geographical region (South vs others).The P value for interaction was estimated using a Wald test on the crossproduct term of type 2 diabetes and each stratification factor.All the analyses were performed using SAS, version 9.4 (SAS Institute, Cary, NC).All the statistical tests were 2-sided, and P values <.05 were considered statistically significant.

Results
A total of 6001 early-onset CRC cases and 52,104 controls were included in the analyses (Table 1).The mean age of patients with early-onset CRC was 43.0 years.Compared with controls, early-onset CRC cases were more likely to have IBD and be coded for obesity.They also had higher rates of colonoscopies other than for screening and FOBT tests before the index dates.Our analysis included 2248 individuals with type 2 diabetes; the majority (n ¼ 1934) had claims for diabetes starting between 1 year and 2 years before early-onset CRC diagnosis.
Type 2 diabetes was present in 5.0% of patients with earlyonset CRC, compared with 3.7% among the controls (Table 2).In comparison with those without type 2 diabetes, individuals with type 2 diabetes had a 24% increased risk of early-onset CRC (OR ¼ 1.24, 95% CI: 1.09-1.41),after adjusting for the matching factors and a range of potential confounders, including full-time employment status, residence, type of commercial health insurance, Charlson comorbidity index, IBD, obesity, family history of gastrointestinal cancer, personal history of screening colonoscopy, other colonoscopy, and FOBT.The positive association remained similar when we restricted the analysis to individuals without IBD, without family history of gastrointestinal cancer, without previous colonoscopy/FOBT, or without a list of early signs/symptoms of CRC.When we restricted to diabetes from 1 year to up to 2 years before the index, the association with early-onset CRC remained similar with an OR of 1.22 (95% CI 1.07-1.40).Secondary analyses of early-onset CRC as per the severity of type 2 diabetes revealed a significant positive association for uncontrolled (OR ¼ 1.37, 95% CI: 1.12-1.67)and complicated type 2 diabetes (OR ¼ 1.59; 95% CI: 1.08-2.35).However, there was no association between controlled type 2 diabetes and risk of early-onset CRC (OR ¼ 1.13; 95% CI: 0.94-1.36).

Discussion
In this nested case-control study leveraging real-world claims data with 6001 early-onset CRC cases, we found that type 2 diabetes was associated with a 24% increased risk of developing early-onset CRC compared with individuals without type 2 diabetes.The positive association, largely observed for proximal and distal colon cancer, remained after restricting to individuals without IBD, family history of gastrointestinal cancer, previous colonoscopy/ FOBT, or early signs/symptoms of CRC.We found that this association was more pronounced for uncontrolled or complicated type 2 diabetes than for controlled type 2 diabetes.Our findings suggest that type 2 diabetes contributes, in part, to the rising incidence of early-onset CRC.
A recent systematic review 31 estimated that type 2 diabetes was associated with a 27% increased risk of CRC among older individuals.Our study supports a similar association between type 2 diabetes and the risk of earlyonset CRC.Findings from recent studies attempting to examine the link between type 2 diabetes and early-onset CRC were mixed.In determining the association between metabolic risk and early-onset CRC, Schumacher et al and Low et al did not observe an association between type 2 diabetes and early-onset CRC, but these studies were limited by the number of patients with type 2 diabetes. 6,7However, a Swedish nationwide cohort reported that type 2 diabetes was associated with increased risk of early-onset CRC. 5 Similarly, a recent larger-scale pooled analysis also reported a suggestive association between type 2 diabetes and early-onset CRC with an OR of 1.25 (95% CI 0.93-1.68),including 168 incident early-onset CRC cases among individuals who had type 2 diabetes. 32With a total of 302 patients with type 2 diabetes preceding 6001 early-onset CRC cases, our study is among the first with adequate power to provide a reliable effect estimate for the association between type 2 diabetes and early-onset CRC.Notably, our results are likely generalizable to the U.S. population as the prevalence of type 2 diabetes among our controls  A total of 41 cases with more than one anatomical site were excluded.e Metastatic colon cancer was determined using treatment encounters of liver or lung metastasis within 3 months after the index dates.corresponds to the U.S. national data. 13The similar associations observed across strata of sex, birth year, and geographical regions (South, historically described as the "diabetes belt", 33 vs others) further ensure the generalizability of our findings.
The underlying mechanisms of the association between type 2 diabetes and CRC are not fully understood.Insulin resistance, hyperglycemia, and hyperinsulinemia may play important roles.Our observed positive association for uncontrolled but not for controlled type 2 diabetes further supports that hyperglycemia/hyperinsulinemia are likely critical to colorectal carcinogenesis.Impaired insulin receptor activation and subsequent defective PI3K signaling pathway could lead to insulin resistance and hyperinsulinemia, 34 as well as high levels of insulin-like growth factor (IGF) 1. 35 The insulin-PI3K pathway has been shown to have profound effects on cancer initiation 34 by stimulating colonic mucosal cell growth and sustaining tumor growth.IGFs are key regulators in signal transduction networks that have important roles in neoplasia. 35Uncontrolled hyperglycemia in type 2 diabetes might contribute to DNA damage and aberrant RNA expression that could promote carcinogenesis and cancer progression. 36Moreover, a dysregulated immune system in patients with type 2 diabetes and chronic inflammation accompanied with elevated cytokine levels such as interleukin-6, tumor necrosis factora, and C-reactive protein could promote CRC tumorigenesis. 37Finally, altered host-microbiota cross talk 38 and increased colonic transit time in type 2 diabetes have been linked to dysregulation of bile acid metabolism, 39 which could also contribute to colorectal carcinogenesis.
The more apparent associations between type 2 diabetes and early-onset colon compared with rectal cancer require validation.Emerging evidence indicates that the molecular Additionally adjusted for employment status (full time/others), residence (urban, rural, unknown), health plan (PPO, HMO, others), Charlson comorbidity index without diabetes and cancer (continuous), and any of the following conditions between 91 days and 2 years before the index dates: IBD (yes/no), obesity (yes/no), family history of gastrointestinal cancer (yes/no), screening colonoscopy (yes/no), other colonoscopy (yes/no), and fecal occult blood test (yes/no).c Early signs/symptoms included any of the following conditions between 91 days and 2 years before the index dates: gastrointestinal bleeding, abdominal pain, anemia, change of bowel habits, diarrhea, constipation, and weight loss.features of CRC vary by anatomic subsites.Microsatellite instability, CpG island methylator phenotype, and BRAF mutation gradually increase from the rectum to the ascending colon. 40Rectal cancers also exhibit more TP53 mutations and fewer PIK3CA mutations or CTNNB1 mutations. 41Given the heterogeneous nature of CRC, it is expected that risk factors also differ by anatomic location.In line with our findings in younger adults, a case-control analysis involving 21,744 CRCs from veterans (median age 68 years) reported that diabetes was associated with 29%, 15%, and 12% increased risk of proximal, distal, and rectal cancer, respectively. 42It is hypothesized that IGF1 activates the PI3K/AKT pathway through the PIK3CA mutation. 34The lower frequency of PIK3CA mutations in rectal cancer may, in part, explain the lack of association between type 2 diabetes and rectal cancer.Further elucidations of mechanisms underlying these differential associations are critical for the development of precision prevention strategies.
Our study has several strengths.This large, nested casecontrol study leveraged longitudinal claims data from close to half of the U.S. adult population.Such an unprecedented sample size provided a unique opportunity to examine the association of interest.This type of examination between type 2 diabetes and early-onset CRC is not otherwise feasible in existing prospective cohort studies or other realworld electronic health record-based databases because of the relatively low prevalence of type 2 diabetes and low incidence of early-onset CRC in younger adults.Our rigorous study design restricted CRC cases to those only with confirmed pathology claims, and patients with type 2 diabetes were identified through ICD-9-CM coding followed by an established algorithm to maximize reliability.To minimize potential detection bias from patients presenting with signs/symptoms that directly led to a diagnosis of CRC, we leveraged claims data from 91 days to 2 years before CRC diagnosis and adjusted for a list of variables associated with detection.We also conducted sensitivity analyses to minimize the influence of comorbidities/symptoms that may have led to differential detection of type 2 diabetes among cases and controls.
The study also has a few limitations.First, we were not able to reliably identify the first date of type 2 diabetes diagnosis and thus could not assess the impact of duration of diabetes.However, among the prospective studies that allowed such assessment, there were a limited number of CRC cases with diabetes, 43 and such investigation is largely infeasible for early-onset CRC.Second, residual confounding could not be ruled out because of limited information on putative confounders such as tobacco use, alcohol intake, diet, and physical activity.However, in analyses among older populations, these factors minimally confound the association between type 2 diabetes and CRC of lateronset. 44In addition, the associations between these factors with early-onset CRC remain to be elucidated.For instance, a recent analysis assessing nongenetic risk factors for earlyonset CRC did not find an association with tobacco use. 32hird, we acknowledge the percent of obese patients in our sample was low because of under coding in claims data.However, in claims data, diagnostic codes for obesity primarily captured morbid obesity with high positive predictive values 45 and have been widely used to define obesity in studies assessing surgical outcomes. 46In addition, type 2 diabetes is in the causal pathway between obesity and CRC, or in other words, a major link between obesity and CRC is primarily through insulin resistance and subsequent type 2 diabetes. 47Obesity thus serves as an instrumental variable, not a confounder, for the association between type 2 diabetes and early-onset CRC.Finally, the MarketScan database does not provide information on race/ethnicity and is restricted to individuals with commercial insurance.Thus, further validation in diverse groups is warranted.
In conclusion, in this large U.S. claims-based nested casecontrol study, type 2 diabetes was associated with increased risk of early-onset CRC, suggesting that the rising incidence Adjusted for the same covariates as the model b in Table 2 without the stratification factor.b P for interaction was calculated by the Wald test using the cross-product term of type 2 diabetes and each stratification factor.
of early-onset CRC may be partially attributed to the surging prevalence of type 2 diabetes.The more pronounced association for uncontrolled or complicated diabetes further highlights the importance of early detection and intervention of type 2 diabetes at younger ages.Our findings lend support to the promise of type 2 diabetes control as an emerging CRC prevention strategy among younger adults. d

dA
total of 18 patients with type 2 diabetes (out of 302) among cases and 137 among controls (out of 1946) could not be classified as controlled, uncontrolled, or complicated.
Metastatic status was imputed using coded diagnosis and/or treatment records for liver/lung metastasis within 90 days of diagnosis.

Table 1 .
Characteristics of Participants as Per Case and Control Status, MarketScan Commercial Database (2006-2015) The Charlson comorbidity index was calculated without diabetes and cancer.
HMO, health maintenance organization; IQR, interquartile range; PPO, preferred provider organization; SD, standard deviation.aControls were matched based on age (18-24 years and every 5 years thereafter), sex (female, male), geographical region (Northeast, North Central, South, West, unknown), duration of insurance enrollment before index diagnosis (years), and prescription drug coverage (yes, no).b Between 91 days and 2 years before the index date.c

Table 3 .
Type 2 Diabetes and Risk of Early-onset Colorectal Cancer as Per Anatomical Site a Adjusted for the same set of covariates as model a in Table2.bAdjustedfor the same set of covariates as model b in Table2.c A total of 41 cases with more than one anatomical site were excluded.

Table 2 .
Type 2 Diabetes and Risk of Early-onset Colorectal CancerAdjusted for matching factors including age (year), sex (female, male), duration of insurance enrollment (year), region (Northeast, North Central, South, West, unknown), and prescription drug coverage (yes/no). a

Table 4 .
Stratified Analyses of Type 2 Diabetes and Risk of Early-onset Colorectal Cancer a