Metabolic Risk Factors Associated with Early-Onset Colorectal Adenocarcinoma: A Case–Control Study at Kaiser Permanente Southern California

This article is featured in Highlights of This Issue, p. 1767

Colorectal cancer is the third most common cancer in the United States (1). Since screening programs were implemented, the incidence of colorectal cancer has been falling for individuals over age 50. However, early-onset colorectal cancer incidence diagnosed among individuals under age 50 has been on the rise, and now accounts for more than 1 in 10 colorectal cancer cases (1–5). An examination of data from the Surveillance, Epidemiology and End Result (SEER) cancer registries revealed significantly increasing colorectal cancer incidence in those as young as 20 to 29 years of age. Early-onset colorectal cancer (eoCRC) tends to be diagnosed at advanced stages, resulting in considerable morbidity and mortality (6). As such, prevention and screening strategies targeting eoCRC are needed. Among older adults, body mass index (BMI), type II diabetes mellitus (T2DM), diet, smoking, and exercise have been associated with colorectal cancer, while other potential risk factors such as hypertension and dyslipidemia have shown mixed results in prior studies (7–15). For eoCRC, however, the data describing its risk factors have just begun evolving (1, 7, 8, 16–25). The established risk factors for eoCRC include family history and inflammatory bowel disease (IBD), while an association between eoCRC risk and obesity, smoking, alcohol use, diet (e.g., processed meat consumption), and sedentary lifestyle have been observed in prior studies (21, 26–34). We hypothesized that some risk factors associated with colorectal cancer in older adults may also be associated with eoCRC. We are particularly interested in metabolic abnormalities as risk factors because of their observed increasing prevalence in the United States, which could potentially explain the rise of eoCRC incidence (35–37). Understanding the risk factors underlying eoCRC is critical to help define public health recommendations. Utilizing a large number of eoCRC cases, we investigated the association between obesity, T2DM, hypertension, and dyslipidemia with risk of eoCRC adenocarcinoma, the most common histology type of eoCRC, in this population-based case–control study.

This study was conducted in Kaiser Permanente Southern California (KPSC), an integrated health care delivery system serving over 4.4 million members who are broadly representative of the racially/ethnically and socioeconomically diverse residents in southern California (38). We used a population-based case–control study design to evaluate risk factors for eoCRC. KPSC members who met the following inclusion criteria were identified: (i) diagnosed with invasive colorectal cancer at age 15 to 49 years; (ii) diagnosed between 2008 and 2018 (BMI and smoking data became consistently available in 2008), and (iii) did not have another cancer diagnosis prior to the colorectal cancer diagnosis. Cases who met the following exclusion criteria were removed from this analysis: (i) with less than 2 years of KPSC membership prior to eoCRC diagnosis; (ii) had neuroendocrine or squamous cell histology; and (iii) had a history of IBD, given the different clinical carcinogenic pathways for those with IBD. Cases were identified using KPSC's SEER-affiliated cancer registry. Quality of the cancer registry data is assured by the SEER standard and is audited by SEER staff on a regular basis. Cases were matched one-to-five with control subjects based on age (yearly), sex, and length of membership (± 6 months), using the principle of density sampling. The eligibility criteria for controls included (i) being a KPSC member at the diagnosis date for the case; and (ii) having no prior cancer diagnosis before index date. Each control was assigned an index date that was the date of eoCRC diagnosis for his/her matched case. We limited our cases and controls to those with at least two years of prior membership to reduce protopathic bias by using exposure and confounder measurements from before one year prior to the diagnosis/index date. This reduced the chance that risk factor status is affected by the progression of cancer, such as the effect of cancer on weight loss. It also reduced the chance that measurement of risk factors is affected by increased encounters with the health care system for the cancer diagnostic workup in cases. This study was approved by the KPSC's Institutional Review Board and was conducted in accordance with the U.S. Common Rule guidelines. The requirement for informed consent was waived by KPSC's IRB.

Characteristics of the colorectal cancer cases, including age at diagnosis, histology, stage at diagnosis, and anatomic site, were collected from KPSC's cancer registry. Other medical information was obtained using KPSC's electronic health records, which accounts for all inpatient and outpatient encounters. The electronic health records include demographics, clinical diagnoses, anthropometrics, family history, and social history information linkable through a unique member identifier. We collected data for the following exposures of interest: BMI, T2DM, hypertension, and dyslipidemia. Data on the following potential confounders were also collected: race/ethnicity, tobacco smoking (current, former, or never), family history (defined as colorectal cancer diagnosed in first degree relatives), census track income and education level, and health care utilization (8–12, 16, 18, 21, 39–41). The status of each of the above variables from one year prior to the diagnosis/index date was assessed except for family history, which was assessed with all available data from the family history table. BMI, blood pressure, and tobacco smoking were routinely assessed at each patient visit. History of T2DM was defined using a combination of diagnosis codes, laboratory test results, and medication use. Those with T2DM were further classified into controlled versus uncontrolled diabetes defined as Hemoglobin A1C (HbA1C) <7% and ≥7%, respectively, using the latest HbA1C measurement from >1 year prior to the diagnosis/index date. Hypertension was defined based on diagnosis code and use of antihypertensive medication. Dyslipidemia was defined based on abnormal test results from any of the following based on the most recent test: low-density lipoprotein (LDL), high-density lipoprotein (HDL), total cholesterol, and triglycerides. We primarily defined dyslipidemia based on any of the four lipid test results outside of the optimal range (i.e., including near optimal, borderline high, high, and very high values. Please see Supplemental Table S1 for cut-offs). Alternatively, we defined dyslipidemia as any of the four lipid test results in the high or very high range (Supplemental Table S1). Those without any test for one or more of the four lipid components during the eligible membership history (i.e., >12 months prior to index date) were considered to have missing value for that (or those) lipid component(s). We also collected diagnosis codes and lipid lowering drugs (statins and others) and define dyslipidemia based on these as a second alternative definition. We obtained race/ethnicity data from KPSC's cancer registry and membership file, while information on census level socioeconomic status was obtained from routinely Geocoded files. Outpatient, emergency room (ER) and inpatient visits (between 12 and 24 months prior to diagnosis/index date) were collected from KPSC's utilization database. Further, a history of IBD including both Crohn disease and ulcerative colitis was captured using International Classification of Diseases (IDC) 9 and ICD-10 diagnosis codes for study population exclusion. Detailed definitions for each of these variables are available in Supplementary Table S1.

The distribution of demographic and clinical characteristics among cases and controls were calculated. Among cases, the distribution of anatomic location (colon vs. rectum) and stage at diagnosis were further examined by age at diagnosis (15–29, 30–39, and 40–49 years). Bivariate conditional logistic regression was used to evaluate the crude association between each variable and eoCRC adenocarcinoma. BMI was initially considered as <25, 25–29.9, 30–39.9, and ≥40. Because no significant difference was found between the last two groups in terms of colorectal cancer risk, BMI was modeled as normal weight (BMI < 25), overweight (BMI 25–29.9), and obesity (BMI ≥ 30). T2DM was stratified as no diabetes versus controlled (HbA1c < 7%) versus uncontrolled (HbA1c ≥ 7%).

A multivariable logistic regression was constructed, mutually adjusting for all exposures of interest as well as race/ethnicity, family history, smoking, census track education, and income levels, and heath care utilization. The Bayesian Improved Surname Geocoding imputation method was used to impute for the missing race/ethnicity information (5% missing), while Rubin's multiple imputations were used to address missing information in BMI (11% missing), smoking (10% missing), census track education and income level (0.5% missing), and dyslipidemia (18% missing) (42). The analysis was then repeated, stratified by the location of the cancer (colon vs. rectum). Two-way interaction terms between obesity and age (≥40 vs. <40 years), and between obesity and sex were created and tested using the likelihood ratio test. Alternative definitions for dyslipidemia were used in separate models.

Genetic testing information was not available for patients in this study. To further assess risk factors in patients without a predisposition for eoCRC adenocarcinoma, another set of analysis was conducted that excluded those with family history. In a separate sensitivity analyses, we redefined family history to be colorectal cancer diagnosed in first-degree relatives under age 60. All analyses were performed using Statistical Analysis Software (SAS).

A total of 1,828 invasive colorectal cancer cases were diagnosed between ages 15 and 49 years from 2008 to 2018 at KPSC. Of these, 1,023 cases of adenocarcinoma met the additional eligibility criteria (Fig. 1), with 5,128 matched controls selected. There was an equal distribution of sex among the cases (50% male, Table 1). The majority of the cases were diagnosed between ages 40 and 49 years (76%). Twenty-three percent were metastatic at diagnosis overall, and this varied with age. Among patients 15 to 29, 30% were metastatic at diagnosis, versus 33% for patients 30 to 39, and 20% for patients 40 to 49 (Table 2). In terms of tumor location, about 60% of colorectal cancers were found in the colon while about 40% were in the rectum or rectosigmoid colon (Table 2).

Univariate analysis showed that the following metabolic risk factors were associated with eoCRC adenocarcinoma: obesity and hypertension (Table 3). On multivariate analysis, obesity [OR = 1.41; 95% confidence interval (CI), 1.15–1.74] was statistically significantly associated with risk of eoCRC adenocarcinoma (Table 3). T2DM [controlled (OR = 0.80; 95% CI, 0.56–1.14) or uncontrolled (OR = 0.97; 95% CI, 0.69–1.36)], hypertension (OR = 1.14; 0.94–1.40), and dyslipidemia OR = 1.05 (95% confidence interval 0.85–1.31) were not statistically significantly associated with risk of colorectal adenocarcinoma. Results for dyslipidemia were similar when using alternative definitions [OR = 1.05 (0.88–1.26) when defined based on test results in the high or very high range (30% of the study subjects met this definition); OR = 0.95 (0.78–1.16) when defined based on diagnosis code and/or lipid lowering medication use (22% of the study subjects met this definition), data not shown]. In addition, as expected, family history (OR = 3.13; 2.34–4.18) were significantly associated with risk of eoCRC adenocarcinoma. Cases were also more likely to have any ER and/or inpatient utilization in the prior 12 to 24 months (OR = 3.25 (2.70–3.90). No clear associations were found for outpatient visits [in the prior 12–24 months, OR = 1.01 (1.00–1.02) per additional visit, P = 0.07], smoking [OR = 0.82 (0.62–1.07) and 0.78 (0.63–0.97) for current and former smokers, respectively, in reference to nonsmokers], census track education [OR = 1.05 (0.85–1.29) and 1.18 (0.90–1.54) for those residing in census blocks with 51% to 75% and >75% adults with college degree, respectively, in reference to those residing in census blocks with ≤50% adults with college degree], and census track income levels [OR = 0.89 (0.69–1.14) and 0.95 (0.70–1.27) for those residing in census blocks with $45,001 to $80,000 and >$80,000 median household income, in reference to those residing in census blocks with ≤$45,000 median household income). Interaction terms between obesity and age as well as between obesity and sex were not statistically significant (P = 0.43 and 0.37, respectively).

In the stratified analyses by cancer location, obesity was significantly associated with colon adenocarcinoma OR = 1.56 (1.17–2.07), but not for rectal adenocarcinoma OR = 1.19 (0.85–1.68); (Table 3). No other metabolic risk factors were found to be significant in both locations. Family history and prior ER/inpatient utilization were both significantly associated with risk of colon and rectum adenocarcinoma, although the magnitude of association for family history appeared to be stronger for rectum cancer OR = 3.84 (2.42–6.09) than for colon cancer (OR = 2.61; 1.74–3.92).

In a separate analysis, cases were restricted to patients without family history (n = 924, 90% of all cases). Similar significant associations for obesity were found for colorectal adenocarcinoma in this restricted cohort compared with all cases. When analysis was restricted based on cancer location, there was a statistically significant association between obesity and colon adenocarcinoma, but not for rectal adenocarcinoma (Table 4).

When we used a more restrictive definition of family history (first-degree relatives diagnosed with colorectal cancer under age 60), results for the association with obesity, diabetes, hypertension, and dyslipidemia were similar. Compared with a family history of colorectal cancer at any age, this narrower family history variable showed a stronger association with colorectal adenocarcinoma (OR = 5.00; 95% CI, 2.79–8.97), colon adenocarcinoma OR = 2.96 (1.38–6.37), and rectal adenocarcinoma OR = 12.91 (4.01–41.54).

In this study based on 1,032 early-onset colorectal adenocarcinoma, we found that obesity was significantly associated with increased risk of early-onset colorectal adenocarcinoma. On the other hand, contrary to CRC in older adults, no clear association was found for T2DM, hypertension or dyslipidemia. These results help shed lights on the etiology and prevention of eoCRC, and suggest a role of obesity in helping to identify patients who may benefit from earlier colorectal cancer screening, especially those who do not meet the current early screening indications (i.e., family history and IBD).

The prevalence of obesity in U.S. adults has been rising since the 1970s (43). Therefore, it is possible that obesity may partially explain the increasing trend of eoCRC since mid-1990 (44). Our finding on the positive association between obesity and risk of eoCRC is consistent with several prior studies (29, 33, 34). These included the study of Syed and colleagues, which identified 5,710 eoCRC cases using EMR and reported an odds ratio of 2.88 for obesity. In addition, our findings are in line with published studies of older adults, and suggest at least some similarity among risk factors for colorectal cancer in young adults versus older adults (9, 15, 18, 21). However, lack of association between obesity and eoCRC has also been reported by other case–control studies (21, 26, 30), as well as an ecological study which examined state level trend of obesity and found no correlation to eoCRC incidence (22). A well-recognized limitation in most prior studies is the relatively small sample size, which may explain the inconsistency in their observations related to obesity. Therefore, additional large studies based on high quality, prospectively collected data remain needed to confirm the role of obesity. Future research should also clarify the role of obesity in eoCRC development independent of diet and sedentary lifestyle, which have both been linked to risk of eoCRC (21, 31). Furthermore, none of the studies conducted to date evaluated the effect of obesity from different life periods (e.g., childhood, adolescents, young adults) on risk of eoCRC. Thus, the relationship between obesity and eoCRC remains to be further elucidated.

When we separately examined colon and rectal adenocarcinoma, obesity remained a significant risk factor for colon adenocarcinoma (OR = 1.56), but not for rectal adenocarcinoma (OR = 1.19). Several studies have suggested that risk factors may have differential associations with colon versus rectal cancer (31, 45, 46). That said, given that an elevated odds ratio was still suggested for rectal adenocarcinoma, the lack of statistical significance may be due to insufficient power in this subset analysis. The effect of obesity on early-onset rectal cancer remains to be confirmed (26, 27, 29, 33).

Our null association for T2DM was consistent with a small number of prior studies that examined this potential risk factor (21, 26). It should be noted that the power for estimating the association between T2DM and eoCRC may be limited in this study, as only 10% of the study population had T2DM. That said, the OR estimates for T2DM, both controlled and uncontrolled, are consistent with a lack of a positive association. Although we did not observe any clear association between hypertension and eoCRC risk, Syed and colleagues reported an elevated odds ratio (2.86, P < 0.001) for hypertension and risk of eoCRC (34). In older adults, a systematic review and meta-analysis concluded that hypertension was associated with risk of overall colorectal cancer, although there was less clear evidence when colon and rectal cancers were examined separately (13). Few studies have examined the relationship between dyslipidemia and risk of eoCRC. In the study by Syed and colleagues, hyperlipidemia was strongly associated with eoCRC (OR = 2.39, P < 0.001; ref. 34). However, it does not appear that prior health care utilization was adjusted for in their analysis, and it is not clear whether and what potential confounders were adjusted in their statistical models. Accounting for the potential differential prior utilization between those with and without eoCRC may be particularly important for hypertension and dyslipidemia, given these conditions are both asymptomatic and are thus prone to detection bias. Furthermore, it was unclear how the temporal relationship was defined in the study by Syed and colleagues for hyperlipidemia, hypertension, and obesity, and whether the observed association might be due to increased clinical workup for cases before cancer diagnosis. In older adults, findings for dyslipidemia have been mixed, although a meta-analysis based on prospective studies concluded that dyslipidemia is a risk factor for colorectal cancer (14). A meta-analysis or pooled analysis aggregating a large number of cases may be similarly needed to further determine the effect of dyslipidemia on eoCRC risk.

There are several limitations to our study that should be considered when interpreting our results. First, we did not have high quality information on alcohol use, diet, or sedentary behaviors. Prior studies have shown an association between these lifestyle factors and risk of eoCRC (21, 31, 34). Therefore, it is possible that the observed association with obesity could be partially or fully explained by these potential confounders. Second, the assessment of family history may not be complete for all individuals and may be more complete for cases than controls. Indeed, the percentage of patients with colorectal cancer with family history in our study (10.3%) was lower than the reported rate of family history/genetic syndromes in prior studies (16%–25%; refs. 4, 47, 48). Third, it is possible that we had under-ascertainment of asymptomatic conditions such as dyslipidemia, especially when it is not routinely screened across this age group (blood pressure, on the other hand, is routinely measured at every medical encounter). In addition, many young adults may not have had routine encounters with the health care system in general. This could lead to differential misclassification if cases and controls have differential utilization. We attempted to lower this likelihood by using exposures status measured from at least 1 year prior to diagnosis/index date. This approach also mitigated the concern of the exposure status being affected by the cancer itself, although this remains a possibility given the presumed long induction time of colorectal cancer. Furthermore, utilization could also be related to the exposures of interest and severity of the medical conditions, such as T2DM and obesity, and introduce potential bias via differential misclassification. However, we found similar degree of completeness (90%) for BMI and smoking (both routinely assessed in medical encounters) among cases and controls, suggesting that this potential bias is unlikely to fully explain the results. Furthermore, we have adjusted for heath utilization from 1 year prior to diagnosis/index date in the multivariable models, which further mitigates the potential impact of differential health care utilization on our findings. Fourth, we did not define the length of exposure or age of onset for any of the metabolic abnormalities. Therefore, we cannot draw conclusions regarding long-term exposure to these factors. Fifth, with regards to the effects of BMI, those who are underweight may have a different risk of eoCRC compared to those with normal weight. However, given the small number of subjects who were underweight, we were unable to separately examine this BMI category. Finally, the demographics of our patient population may differ from other parts of the United States. Specifically, 40% of our colorectal cancer cases are Hispanic, which represents the population in Southern California, but is higher than the U.S. population (19% Hispanic; ref. 49). Given that Hispanics have higher prevalence of T2DM, including among the youth population (50), the relationship between diabetes and risk of eoCRC should be further studies by race/ethnicity, as well as by time since diabetes onset.

Despite these limitations, our study has several important strengths, including a relatively large sample size of eoCRC adenocarcinoma cases, the race/ethnicity diversity, and the prospectively collected high-quality EMR data.

Because of the rising incidence of colorectal cancer among patients <50 years old, the age cutoff for screening recommendations has been reexamined. The ACS recently revised its screening recommendations to include average risk patients starting at age 45 (51). However, there is diminishing return to lowering the screening cutoff further, and age alone may not be the best selection criteria for screening in younger patients. Early screening for persons with IBD and family history are already incorporated into national screening guidelines, although the details of these guidelines may continue to be refined (51–54). The major risk factors for colorectal cancer in our study were family history and obesity, suggesting that strategies aimed both at early detection as well as prevention will continue to play important roles in improving outcomes for eoCRC. These factors could be used to help guide which patients may be the best candidates for earlier screening. It remains critically important to further our understanding of the etiology of eoCRC, including the impact of early life exposures and environmental exposures in order to better inform prevention and screening strategies for eoCRC.

No disclosures were reported.

A.J. Schumacher: Conceptualization, resources, supervision, investigation, methodology, writing–original draft. Q. Chen: Data curation, formal analysis, writing–review and editing. V. Attaluri: Writing–review and editing. E.C. McLemore: Writing–review and editing. C.R. Chao: Conceptualization, resources, supervision, investigation, methodology, writing–original draft.

The authors thank the patients of Kaiser Permanente for helping us improve care through the use of information collected through our electronic health record systems.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.