Evaluation of Early-Life Factors and Early-Onset Colorectal Cancer Among Men and Women in the UK Biobank

In contrast to individuals older than 50 years, the incidence of early-onset colorectal cancer (EOCRC) in the United States has been increasing since at least the mid-1990s, especially among persons aged 40–49 years.1Murphy C.C. et al.Gastroenterology. 2017; 152: 1809-1812.e3Abstract Full Text Full Text PDF PubMed Scopus (47) Google Scholar Beyond well-established risk factors for EOCRC, including inflammatory bowel disease (IBD) and family history of CRC, recent data suggest nonmodifiable (eg, male sex) and modifiable (eg, hyperlipidemia, obesity, and alcohol consumption) factors may also play a role.2Gausman V. et al.Clin Gastroenterol Hepatol. 2020; 18: 2752-2759.e2Abstract Full Text Full Text PDF PubMed Scopus (91) Google Scholar,3O’Sullivan D.E. et al.Clin Gastroenterol Hepatol. 2021; PubMed Google Scholar As colorectal carcinogenesis is typically a decades-long process, the etiologically relevant time period for EOCRC may extend into early childhood. However, there is a paucity of data for early-life factors and EOCRC, especially in infancy and childhood. The only epidemiologic study assessing early-life factors in the context of EOCRC reported a compelling association between obesity at age 18 years and EOCRC in women.4Liu P.H. et al.JAMA Oncol. 2019; 5: 37-44Crossref PubMed Scopus (211) Google Scholar Moreover, it has been hypothesized that maternal factors may play a role in EOCRC development,5Zhang Q. et al.Med Hypotheses. 2018; 121: 152-159Crossref PubMed Scopus (15) Google Scholar but no epidemiologic studies have investigated this question. We used data from the UK Biobank cohort to explore the relationship between multiple early-life factors and EOCRC risk. UK Biobank is a large prospective cohort of more than 500,000 residents in the United Kingdom.6UK Biobank. https://www.ukbiobank.ac.uk/Google Scholar At baseline (2006–2010), participants completed a detailed questionnaire assessing lifestyle and health history. Participant information was linked with health records and registries. We aimed to compare the distribution of early-life factors in individuals who developed EOCRC (defined as diagnosis before age 50 years) compared with those who did not. Diagnosis occurred either before the baseline assessment (prevalent cases) or during follow-up (incident cases). Controls included participants without prior history of invasive cancer at baseline, except for nonmelanoma skin cancer. We excluded individuals missing all early-life exposures, leaving 451,615 persons (455 prevalent EOCRC cases, 85 incident EOCRC cases, and 451,075 controls) aged 38–73 years for analysis. Exposures included the following 6 early-life factors assessed in the baseline questionnaire: breastfeeding in infancy, maternal smoking at birth, comparative body size and height at age 10 years, age at menarche for women, and relative age of first facial hair for men. We estimated the association between early-life factors and EOCRC using minimally and fully adjusted multivariable logistic regression models (Supplementary Methods) and tested for trend across ordinal variables. We included prevalent cases to maximize statistical power to study this rare outcome. Because the exposures we studied occurred well before CRC diagnosis, we do not anticipate appreciable bias from this design. We performed sensitivity analyses using multivariable Cox proportional hazards regression restricting the outcome to incident cases. We also conducted sensitivity analyses excluding individuals with IBD and explored results by cancer subsite, year of diagnosis, and age at diagnosis. Compared with controls, patients with EOCRC were younger and more likely to be male (Supplementary Table 1). Family history of bowel cancer was 2-fold higher in individuals with EOCRC (n = 540) than controls (21.7% vs 10.8%; P < .001). None of the 6 early-life factors showed statistically significant associations with EOCRC in either minimally or fully adjusted multivariable models (Table 1). Sensitivity analyses examining the time to incident EOCRC diagnosis (n = 85) similarly showed no statistically significant associations (Supplementary Table 2), although the interpretability of these estimates was hampered by limited statistical power. Excluding individuals with IBD and stratifying by year and age of diagnosis did not affect the interpretation of results. In subsite-stratified exploratory analyses, height at age 10 years was positively associated with early-onset colon cancer (vs shorter: odds ratio [about average], 1.39; 95% confidence interval, 1.02–1.94; odds ratio [taller], 1.54; 95% confidence interval, 1.09–2.20; P trend = .02), but not rectal cancer. Although height has been linked with overall CRC, this observation requires further study because it may have been a chance finding due to the large number of exploratory tests.Table 1Associations Between Early-Life Factors and Early-Onset Colorectal CancerPrevalent + incident EOCRCControl, n (%)(n = 451,075)Case, n (%)(n = 540)Minimally adjustedaAdjusted for age (years), race (White; non-White), and sex. Minimally adjusted models for age at menarche (women only) and age of first facial hair (men only) were not adjusted for sex. OR (95% CI)Fully adjustedbAdjusted for age (years), race (White; non-White), sex, household income (<£18,000; £18,000–30,999; £31,000–51,999; £52,000–100,000; and >£100,000; do not know/prefer not to answer/missing), family history of bowel cancer (no; yes; do not know/prefer not to answer/missing), and the other early-life factors. OR (95% CI)Breastfed as baby No97,710 (21.7)126 (23.3)ReferenceReference Yes251,714 (55.8)298 (55.2)1.04 (0.84–1.29)1.04 (0.84–1.29) Missing101,651 (22.5)116 (21.5)1.05 (0.81–1.36)1.04 (0.80–1.36)Comparative body size to peers at age 10 y Thinner149,831 (33.2)171 (31.7)ReferenceReference About average229,443 (50.9)275 (50.9)1.06 (0.88–1.29)1.04 (0.85–1.26) Plumper71,801 (15.9)94 (17.4)1.15 (0.89–1.47)1.13 (0.87–1.45) P trend——.37.32Comparative height to peers at age 10 y Shorter92,398 (20.5)95 (17.6)ReferenceReference About average245,157 (54.4)309 (57.2)1.23 (0.98–1.55)1.21 (0.96–1.53) Taller113,520 (25.2)136 (25.2)1.15 (0.89–1.50)1.13 (0.87–1.48) P trend——.36.45Maternal smoking at birth No275,959 (61.2)335 (62.0)ReferenceReference Yes115,586 (25.6)137 (25.4)0.95 (0.78–1.16)0.94 (0.77–1.15) Missing59,530 (13.2)68 (12.6)0.97 (0.74–1.25)0.97 (0.74–1.26)Age at menarche 11 y or younger54,907 (22.2)52 (19.8)ReferenceReference 12–13 y104,613 (42.2)113 (43.0)1.12 (0.81–1.57)1.13 (0.82–1.59) 14 y or older88,311 (35.6)98 (37.3)1.16 (0.83–1.64)1.19 (0.85–1.68) P trend——.40.32Relative age of first facial hair Younger than average13,910 (6.8)26 (9.4)ReferenceReference About average162,871 (80.1)221 (79.8)0.80 (0.54–1.22)0.80 (0.54–1.23) Older than average26,463 (13.0)30 (10.8)0.61 (0.36–1.04)0.64 (0.38–1.10) P trend——.06.10CI, confidence interval; OR, odds ratio.a Adjusted for age (years), race (White; non-White), and sex. Minimally adjusted models for age at menarche (women only) and age of first facial hair (men only) were not adjusted for sex.b Adjusted for age (years), race (White; non-White), sex, household income (<£18,000; £18,000–30,999; £31,000–51,999; £52,000–100,000; and >£100,000; do not know/prefer not to answer/missing), family history of bowel cancer (no; yes; do not know/prefer not to answer/missing), and the other early-life factors. Open table in a new tab CI, confidence interval; OR, odds ratio. This is one of the first population-based studies examining multiple early-life factors and EOCRC. We found no association for 6 early-life factors, suggesting that these factors are unlikely to play a prominent role in the carcinogenesis of EOCRC. In response to the rising incidence of EOCRC, the US Preventive Services Task Force recently lowered the recommended CRC screening age for average-risk individuals to 45 years.7US Preventive Services Task Force et al.JAMA. 2021; 325: 1965-1977Crossref PubMed Scopus (328) Google Scholar Accordingly, understanding the drivers of this troubling trend is paramount for risk stratification in those ineligible for screening and will allow mitigation of identified risk factors on a population level. There are especially limited data on early-life risk factors, which provide a long and biologically plausible lag time between timing of exposure and cancer development. The major strength of this study was the availability of early-life exposures in a cohort with a large number of EOCRC cases. However, recall bias may have influenced the findings, particularly given 84% of cases were diagnosed before baseline. UK Biobank is not specific to CRC and our panel of early-life factors are not established cancer risk factors, which helps reduce this bias; moreover, our interpretations remained unchanged after restricting to incident cases. The number of EOCRC cases in our study is nearly 5-fold larger than the only other epidemiologic study to assess early-life factors and EOCRC4Liu P.H. et al.JAMA Oncol. 2019; 5: 37-44Crossref PubMed Scopus (211) Google Scholar; we had sufficient power to detect clinically meaningful effect sizes smaller than those reported in that study (Supplementary Methods). However, we are unable to rule out weak associations, which will require combined/meta-analyses of this rare outcome. There is interest in early-life factors, such as mode of delivery (cesarean or vaginal) and childhood antibiotic use because these may influence carcinogenesis through the microbiome8Akimoto N. et al.Nat Rev Clin Oncol. 2021; 18: 230-243Crossref PubMed Scopus (137) Google Scholar; however, missing data for these variables (>65%) exceeded our a priori threshold for inclusion. Our study is therefore only a first step in filling the knowledge gap for the etiology of EOCRC and warrants follow-up in other large studies evaluating early-life exposures. In summary, our data suggest a panel of 6 early-life factors does not play a clinically meaningful role in the development of EOCRC. A variety of lifestyle, clinical, and biologic factors may contribute to the increasing incidence and mortality of EOCRC that has been observed in the United States and several other countries. Additional large studies are needed to confirm and extend our findings on measurable early-life factors, which may lead to improved risk stratification for EOCRC. This research was conducted using the UK Biobank Resource under application number 19115. Data are available to other researchers by applying to the UK Biobank. Analytic methods will be made available to other researchers on request. Valerie Gausman, MD (Writing – original draft: Lead; Writing – review & editing: Equal). Peter S. Liang, MD, MPH (Conceptualization: Equal; Investigation: Equal; Supervision: Equal; Writing – original draft: Equal; Writing – review & editing: Equal). Kelli O’Connell, MSPH (Conceptualization: Supporting; Data curation: Equal; Formal analysis: Lead; Methodology: Equal; Writing – review & editing: Equal). Elizabeth D. Kantor, PhD, MPH (Data curation: Equal; Investigation: Equal; Writing – review & editing: Equal). Mengmeng Du, ScD (Conceptualization: Equal; Data curation: Equal; Investigation: Equal; Methodology: Equal; Supervision: Equal; Writing – review & editing: Equal). Minimally adjusted multivariable models included age, sex, and race unless otherwise specified. Fully adjusted models also included household income, family history of bowel cancer, and the other early-life factors. For sex-specific variables (age at menarche and age at first facial hair), individuals of the opposite sex were included as part of the reference category in the multivariable model. We assessed covariates at the baseline survey between 2006 and 2010. We have largely used a complete case approach, in which analyses include only participants with complete data (ie, those missing covariate data were excluded). However, we have alternatively used a missing indicator for variables missing >5% data (eg, income and family history of bowel cancer). This approach avoids the issue of model nonconvergence due to small cell counts (resulting from small numbers missing for most covariates), but also avoids excluding a large number of participants because of missing data for a particular covariate. We decided a priori to exclude early-life factors with substantial missing data from the analysis (>30% missing or cell counts <10) to ensure the internal validity of the data. These included factors such as mode of delivery, birth weight, and childhood antibiotic use. In post-hoc power calculation, we estimated that for a dichotomous variable such as maternal smoking, we had 80% power to detect odds ratio >1.33 or odds ratio <0.73 assuming 2-sided α = .05.Supplementary Table 1Population Characteristics by Case-Control StatusCharacteristicControls (n = 451,075)Cases (n = 540)Age,aAge represents the age at which subjects completed their baseline assessment. n (%) Younger than 50 y109,380 (24.3)181 (33.5) 50–60 y152,362 (33.8)214 (39.6) 60–70 y187,312 (41.5)244 (26.7) 70 y or older2021 (0.5)1 (0.2)Sex, n (%) Female247,831 (54.9)263 (48.7) Male203,244 (45.1)277 (51.3)Race, n (%) White427,351 (94.7)511 (94.6) Non-White23,724 (5.3)29 (5.4)Household income, n (%) <£18,00085,121 (18.9)110 (20.4) £18,000–£30,99997,727 (21.7)110 (20.4) £31,000–£51,999102,505 (22.7)136 (25.2) £52,000–£100,00081,152 (18.0)102 (18.9) >£100,00021,697 (4.8)26 (4.8) Missing62,873 (13.9)56 (10.4)Body mass index, n (%) <25 kg/m2149,126 (33.1)177 (32.8) 25 to <30 kg/m2190,654 (42.3)219 (40.6) 30 to <35 kg/m278,129 (17.3)101 (7.8) ≥35 kg/m230,945 (6.9)42 (7.8) Missing2221 (0.5)1 (0.2)Smoking status, n (%) Never246,715 (54.7)290 (53.7) Former155,490 (34.5)180 (33.3) Current47,403 (10.5)66 (12.2) Missing1467 (0.3)4 (0.7)Height, cm, mean (SD)168.5 (9.3)169.5 (9.1)Family history of bowel cancer, n (%) No344,706 (76.4)366 (67.8) Yes48,910 (10.8)117 (21.7) Missing57,459 (12.7)57 (10.6)Breastfed as baby, n (%) No97,710 (21.7)126 (23.3) Yes251,714 (55.8)298 (55.2) Missing101,651 (22.5)116 (21.5)Comparative body size to peers at age 10 y, n (%) Thinner149,831 (33.2)171 (31.7) About average229,443 (50.9)275 (50.9) Plumper71,801 (15.9)94 (17.4)Comparative height to peers at age 10 y, n (%) Shorter92,398 (20.5)95 (17.6) About average245,157 (54.4)309 (57.2) Taller113,520 (25.2)136 (25.2)Maternal smoking at birth, n (%) No275,959 (61.2)335 (62.0) Yes115,586 (25.6)137 (25.4) Missing59,530 (13.2)68 (12.6)Age at menarche, n (%) 11 y or younger54,907 (22.2)52 (19.8) 12–13 y104,613 (42.2)113 (43.0) 14 y or older88,311 (35.6)98 (37.3)Relative age of first facial hair, n (%) Younger than average13,910 (6.8)26 (9.4) About average162,871 (80.1)221 (79.8) Older than average46,463 (13.0)30 (10.8)SD, standard deviation.a Age represents the age at which subjects completed their baseline assessment. Open table in a new tab Supplementary Table 2Associations Between Early-Life Factors and Incident Early-Onset Colorectal CancerIncident EOCRCControls, n (%)Case, n (%)Minimally adjustedaAdjusted for age (y; time axis of analysis) and sex. Minimally adjusted models for age at menarche (women only) and age of first facial hair (men only) were not adjusted for sex.,bSome categories have been collapsed to ensure sufficient cell counts (n ≥ 10) when possible. HR (95% CI)Breastfed as baby No37,936 (34.7)29 (34.1)Reference Yes55,588 (50.9)41 (48.2)1.04 (0.64–1.69) Missing15,728 (14.4)15 (17.7)1.39 (0.74–2.61)Comparative body size at age 10 y Thinner36,723 (33.6)33 (38.8)Reference About average/plumper72,529 (66.4)52 (61.2)0.80 (0.52–1.25)Comparative height size at age 10 y Shorter22,652 (20.7)20 (23.5)Reference About average57,801 (52.9)46 (54.1)0.91 (0.54–1.53) Taller28,799 (26.4)19 (22.4)0.74 (0.39–1.38)Maternal smoking at birth No69,452 (71.1)55 (70.5)Reference Yes28,161 (28.9)23 (29.5)1.07 (0.65–1.75)Age at menarche 13 y or younger39,068 (64.2)31 (64.6)Reference 14 y or older21,783 (35.8)17 (35.4)0.99 (0.55–1.79)Relative age of first facial hair Younger than average4350 (9.0)6 (16.2)Reference About average/older than average44,051 (91.0)31 (83.8)0.52 (0.22–1.24)CI, confidence interval; HR, hazard ratio.a Adjusted for age (y; time axis of analysis) and sex. Minimally adjusted models for age at menarche (women only) and age of first facial hair (men only) were not adjusted for sex.b Some categories have been collapsed to ensure sufficient cell counts (n ≥ 10) when possible. Open table in a new tab SD, standard deviation. CI, confidence interval; HR, hazard ratio. Early Life: An Important Window of Susceptibility for Colorectal CancerGastroenterologyVol. 163Issue 2PreviewIncidence rates of colorectal cancer have increased in young adults (age <50 years) in the US since the early 1990s, and more recently, incidence rates have increased in adults in their early 50s.1 The shifting epidemiology of colorectal cancer has forced researchers to reconsider what we know about the causes of this disease. Importantly, incidence rates of colorectal cancer have increased across generations or birth cohorts, starting with those born in the 1960s.2 This so-called “birth cohort effect” implicates exposures in early life as risk factors, consistently with a large literature demonstrating the importance of gestation, infancy, and childhood for several adult cancers. Full-Text PDF