Glucosamine Use and Risk of Colorectal Cancer: Results from UK Biobank

Glucosamine and chondroitin are nonvitamin, nonmineral supplements commonly taken for osteoarthritis (1, 2). Research from prospective studies has revealed these supplements, which are commonly taken together in a single daily supplement, may be associated with reduced risk of colorectal adenoma (3) and colorectal cancer (4–7) possibly through an anti-inflammatory mechanism (8–30).

While most studies have observed an inverse association with colorectal cancer risk (4–7, 31, 32), data from several large, U.S. cohorts suggest that this association may vary by time and history of colorectal cancer screening. An exploratory analysis in the VITamins And Lifestyle (VITAL) cohort suggested the association between glucosamine + chondroitin and colorectal cancer was stronger in early years than later years—a difference that was posited to reflect increased measurement error over time, due to use of a single exposure assessment (5). A similar pattern was observed in the Cancer Prevention Study (CPS)-II Nutrition Cohort (7). In exploratory analyses of the Nurses’ Health Study (NHS) and Health Professionals Follow-up Study (HPFS; ref. 6), as well as the CPS-II Nutrition Cohort (7), all of which had repeated exposure measures, the effect estimate for glucosamine/chondroitin use and colorectal cancer strengthened in the never-screened group, raising the possibility that screening is obfuscating the natural history of disease, which may differentially impact associations over time. Further work has suggested potential differences by duration of use (7), nonsteroidal anti-inflammatory drugs (NSAID; refs. 31, 32), and body mass index (BMI; refs. 5, 6), although these associations have not been consistently observed (5–7).

Given the need to identify safe and easily implemented chemopreventive strategies, it is critical to understand if use of these supplements may help reduce colorectal cancer risk within the context of potential modification by screening, time, or other factors. Given that UK Biobank collected data on use of glucosamine, we have used data from UK Biobank to evaluate the association between use of glucosamine and risk of colorectal cancer, and characterize heterogeneity by screening history, age, sex, BMI, and NSAID use. We also study whether the association varies by follow-up time and time of study entry.

This study was conducted within UK Biobank, a large prospective cohort of over 500,000 volunteers, residing in the UK (33–35). Briefly, at the time of baseline assessment (2006–2010), participants provided informed written consent and completed a touchscreen questionnaire and interview that assessed demographics, health history, lifestyle, and diet, among other factors. Of the 502,459 persons, aged 38 to 73 years included in UK Biobank, we excluded those with prior history of invasive colorectal cancer at baseline (n = 27,666). We further excluded persons missing information on the primary exposure of interest, glucosamine (n = 5,884) or covariates missing for less than 5% of study participants (n = 28,913), leaving 439,996 persons for analysis. Although we aimed to be as inclusive as possible, given concern that associations may differ in populations with inflammatory conditions, in sensitivity analyses we examined the associations further excluding those with rheumatoid arthritis (n = 5,095) and inflammatory bowel disease (IBD, n = 3,772); as results remained unchanged, we have not made this exclusion in reported findings.

UK Biobank data were collected in accordance with the UK Biobank Ethics and Governance Council (https://www.ukbiobank.ac.uk/learn-more-about-uk-biobank/about-us/ethics). All participants gave broad consent to use of their anonymized data and samples for any health-related research, and this study was approved by the North West Multicenter Research Ethics Committee in the United Kingdom. It was determined that this analysis did not require oversight of the Memorial Sloan Kettering Cancer Center (New York, NY) Institutional Review Board.

Regular use of glucosamine, as defined in the touchscreen questionnaire, was the primary exposure in this study. Participants were asked, “do you regularly take any of the following?”; a list of supplements was provided from which participants could indicate regular use of glucosamine. Chondroitin use was not assessed on this questionnaire, and glucosamine + chondroitin or chondroitin (yes/no) could not be included as an exposure in this study.

As a secondary exposure, we alternatively defined glucosamine use using a composite variable that additionally included information from an interview on prescription drug use conducted by a trained nurse. This interview was not intended to collect information on supplements, unless for some reason the participant forgot to record it in the touchscreen questionnaire. Therefore, in this secondary analysis, glucosamine use indicates use of glucosamine as reported in either the touchscreen questionnaire or interview. Although this exposure may capture some additional users who forgot to report glucosamine in the initial survey, we have elected not to make this the primary exposure, as it is less clear what this composite exposure represents.

The outcome of this study was invasive colorectal cancer, as determined by registry linkage. Participants were followed for colorectal cancer from the time of baseline questionnaire until date of diagnosis with invasive cancer, death, loss to follow-up or end-of-study (October 31, 2015). Completeness of case ascertainment in English cancer registries is reported to be approximately 98% to 99%, based on a study that linked routine cancer registration with information from the Hospital Episode Statistics database (36).

HRs and corresponding 95% confidence intervals (95% CI) were estimated using Cox proportional hazards regression, with age as the time-axis of analysis.

Minimally-adjusted models adjusted for age (time-axis of analysis), sex, and race (White; non-White). Multivariable models further adjusted for additional variables hypothesized a priori to be associated with exposure and/or outcome, including: household income (<£18,000; £18,000–30,999; £31,000–51,999; £52,000–100,000; >£100,000; missing), BMI (<25 kg/m²; 25–<30 kg/m²; 30–<35 kg/m²; ≥35 kg/m²), history of smoking status (never; former; current), pack-years (0/nonsmokers, ≤12.5, >12.5–26.3, >26.3); alcohol intake (never; special occasions only; 1–3 times/month; 1–4 times/week; daily or almost daily); physical activity (none; any moderate; any vigorous; missing); diabetes (no; yes); history of bowel screening (no; yes), fruit intake (<2.0 servings per day; 2.0–2.9 servings per day; 3.0–3.9 servings per day; ≥4.0 servings per day), vegetable intake (<2.0 servings per day; 2.0–2.9 servings per day; 3.0–3.9 servings per day; ≥4.0 servings per day), red meat intake (<1 time/week; 1.0–1.9 times/week; 2.0–2.9 times/week; 3.0–21.0 times/week), processed meat intake (never; <1.0 time/week; 1.0 time/week; ≥2.0 times/week–1.0 or more times/day), aspirin intake (no; yes), nonaspirin NSAID intake (no; yes), family history of bowel cancer (no; yes; missing), osteoarthritis (no; yes), and menopause status and postmenopausal hormone use (premenopausal/men; postmenopausal without postmenopausal hormone use; postmenopausal with postmenopausal hormone use; missing). A missing indicator was used for covariates missing data on over 5% of participants. As a note, we conducted sensitivity analyses additionally including both supplemental calcium and fish oil; as effect estimates were materially unchanged, these factors were not included in our multivariable models.

For the primary exposure, we evaluated heterogeneity by sex, age (<60 years; ≥60 years), BMI (<30 kg/m²; ≥30 kg/m²), smoking (never; ever), screening history as of baseline (never, ever), and regular NSAID use (no; yes). Stratified results have been presented and the statistical significance of interaction has been assessed using a likelihood ratio test. Given concern regarding potential heterogeneity between normal weight and overweight individuals, we have conducted a sensitivity analysis further disaggregating those with a BMI < 25 kg/m² from those with a BMI of 25 to <30 kg/m²; as results were comparable across these two groups, the results from this sensitivity analysis are not presented. Further, we have conducted a sensitivity analysis in which we have restricted the screening stratification to those 60+ years of age, as this group would have met screening requirements by this age.

We further evaluated heterogeneity by follow-up time by assessing associations separately within the first 5 years of follow-up versus thereafter. In analyses of early follow-up, participants were censored at 5 years. For the latter portion of follow-up, we have entered participants into the model at 5 years and followed participants for the remainder of follow-up (thus, anyone censored in the first 5 years would not contribute to this second set of models).

We have conducted additional exploratory analyses to better understand these associations. In our first exploratory analysis, we have further repeated the stratification by follow-up time (as described above), restricted to those individuals with no history of screening as of baseline questionnaire.

To disentangle the effect of follow-up time from that of calendar time, we have conducted a second exploratory analysis. In this analysis, we examined the stratification by follow-up time separately among those who entered the cohort early (2006–2008) versus those who entered the cohort late (2009–2010). Lastly, in a third exploratory analysis, we have restricted the above-described analysis (stratified on both follow-up time and calendar time) to those with no history of screening as of baseline questionnaire.

All analyses were conducted using R (R Foundation for Statistical Computing).

UK Biobank data are available with approval from UK Biobank (https://www.ukbiobank.ac.uk).

Of the 439,996 study participants, 84,825 (19.3%) reported regular use of glucosamine in the touchscreen questionnaire (Table 1). Of the persons 70 years of age and older, 25.5% reported use of glucosamine, as compared with only 10% of participants under the age of 50 years. Women were more likely to report use (22.1%) than men (16.0%), and White participants were more likely to report use (19.5%) than non-White participants (14.4%). Distributions of additional covariates, by glucosamine use, are provided in Supplementary Table S1. There were some differences between groups. For example, nonsmokers are more likely to report glucosamine use (19.8%) than smokers with a high number of pack-years (15.5%). Those who never drink have a low prevalence of glucosamine use (6.1%) as compared with those who drink one to four times per week (50.3%). Furthermore, those who were premenopausal (11.5%) had a lower prevalence of use than those who were postmenopausal, particularly those who were using postmenopausal hormones (30.3%)

Over a median of 6.7 years of follow-up (maximum: 8.9 years), 2,755 participants developed colorectal cancer. In minimally-adjusted analyses, glucosamine use was associated with 10% lower risk of colorectal cancer (HR = 0.90; 95% CI, 0.92–0.99); however, this association was attenuated in the multivariable model (HR = 0.94; 95% CI, 0.85–1.04; Table 2). In secondary analyses, where regular glucosamine use was additionally ascertained by report in an interview regarding prescription drugs, glucosamine use was not associated with colorectal cancer in the multivariable model (HR = 0.93; 95% CI, 0.84–1.02).

We did not observe heterogeneity in the glucosamine–colorectal cancer association by sex, age, BMI, smoking status, or NSAID use (Table 3). In contrast, there was heterogeneity by screening history (P_interaction = 0.05), with a significant inverse association among the never-screened individuals (HR = 0.86; 95% CI, 0.76–0.98) and no association among those with a screening history (HR = 1.07; 95% CI, 0.92, 1.24). In sensitivity analyses restricted to the smaller subset of participants ages 60+, effect estimates by screening status were comparable with the stratified analyses above (HR never-screened = 0.86; 95% CI, 0.73–1.01 vs. HR ever-screened = 1.04; 95% CI, 0.88–1.22), although the P_interaction was no longer statistically significant (P_interaction = 0.09; Supplementary Table S2).

Given prior observation of heterogeneity by follow-up time, we examined whether the association varied in the first 5 years of follow-up versus thereafter (Table 4). While we did not observe an appreciable difference by follow-up time (HR first 5 years of follow-up = 0.90; 95% CI, 0.81–1.01 vs. HR thereafter = 0.98; 95% CI, 0.82–1.17), the association strengthened in the never-screened group (HR first 5 years of follow-up = 0.83; 95% CI, 0.71–0.97 vs. thereafter HR = 0.88; 95% CI, 0.70–1.11; Supplementary Table S3). In models jointly stratified on follow-up time and calendar time of study entry (2006–2008; 2009–2010), we observed that association was strongest for early follow-up specifically among those who entered the cohort (and for whom exposure was ascertained) in early years (2006–2008; HR = 0.80; 95% CI, 0.67–0.95; Table 5). In comparable models restricted to the never-screened group, again, the only statistically significant association was observed in early follow-up among those who entered the cohort in early years (HR = 0.73; 95% CI, 0.59–0.89; Supplementary Table S4). We did not observe statistically significant associations for the latter portion of follow-up in this early cohort, nor were any significant associations observed for participants surveyed in later years, regardless of follow-up period.

In this large prospective study, there was no overall association between glucosamine and risk of colorectal cancer. However, the association between glucosamine use and risk of colorectal cancer varied by screening status, with a significant inverse association only observed among those with no history of screening at the time of baseline questionnaire. Further, the association was only observed in the first 5 years of follow-up for those entering the cohort in early years (2006–2008). We did not observe differences by sex, age, BMI, smoking status, or NSAID use.

We observed no overall association between glucosamine and risk of colorectal cancer (HR = 0.94). Prior studies have shown inverse associations of stronger effect sizes (4–7). In the prior analyses conducted, exploratory analyses have consistently revealed a stronger association between glucosamine + chondroitin as compared with use of glucosamine alone. In the current study, the touchscreen questionnaire did not ask about use of chondroitin, precluding presentation of this comparison. Thus, it is possible that the association may be weaker here if a larger proportion of persons in the UK (vs. the United States) are using glucosamine alone as opposed to glucosamine + chondroitin. It is also possible that this finding reflects differences across studies in terms of characteristics of exposure (i.e., supplement form), patterns of use (i.e., if more users in this cohort happen to be irregular or short-term users than in other cohorts), differences in residual confounding (i.e., perhaps the association was weaker here due to better control for confounding), or chance.

In this study, we found evidence of heterogeneity by screening status, with the inverse association stronger among the never screened. It is unclear why we see this pattern of association. Although the reasons for this observation are likely multifactorial, one possibility is that by blunting the natural history of disease, screening may obfuscate a glucosamine–colorectal cancer association (6), or that the strong screening–colorectal cancer association makes it challenging to detect a moderate risk factor in the screened group. Understanding this point would require a broader assessment of colorectal cancer risk factors in the context of screening. It is also possible that this pattern of results reflects residual confounding, possibly by other correlated exposures or by other facets of screening, such as recency, frequency, or type of screening. Importantly, this pattern (with results stronger among the never-screened subset) has been consistently observed in prior studies, and is therefore unlikely a chance finding (6, 7).

In this study, we also observed a stronger association in early follow-up than later follow-up, consistent with what we observed in both VITAL (5) and the CPS-II Nutrition Cohort (7). Further analyses in this cohort revealed that the glucosamine–colorectal cancer association was particularly strong in early follow-up among persons who entered the cohort early (2006–2008), before the UK implemented a broad screening strategy. The National Health Service stool-based Bowel Cancer Screening Program started in 2006 in England and was more widely implemented across the UK by 2009/2010 (37, 38); with this, screening coverage would have been relatively low during the early years of our study (2006–2008). These patterns held, and even strengthened, when further restricted to never-screened individuals in exploratory analyses, suggesting that these time-specific results may very well reflect screening. It should also be noted that in the CPS-II Nutrition Cohort, we were also able to examine associations by total duration of use; in these analyses, we observed unexpected stronger associations for persons with a short duration of glucosamine use (7). Interestingly, this difference by duration disappeared when analyses were restricted to never-screened individuals. Thus, it seems likely that differences by time, observed here and elsewhere, may be, in part, interrelated to screening.

In this cohort, we did not observe heterogeneity by age, sex, BMI, smoking status, or NSAID use. This is largely consistent with the literature, although there has been some suggestion of effect modification by factors associated with inflammation-related factors, including BMI and NSAID use (5–7, 31, 32). Specifically, there has been some inconsistent suggestion of effect modification by BMI (5, 6), and other work has suggested that the inverse association may be restricted to NSAID users (31, 32). It should be noted, however, that we have evaluated effect modification by aspirin/non-aspirin NSAID use in other studies, with results of those studies similarly suggesting no effect modification (5–7).

If use of glucosamine does reduce risk of colorectal cancer, it is likely through an anti-inflammatory mechanism, as inflammation has been implicated in the etiology of colorectal cancer, and a combination of laboratory and human data suggest anti-inflammatory effect (39). In vitro models suggest that glucosamine and chondroitin reduce inflammation through inhibition of nuclear factor-kappa B (NFkB), a transcription factor central to the inflammatory cascade, by inhibiting the degradation of its inhibitory subunit, IkB, in a dose-dependent manner (8, 9). Bound by IkB, NFkB is unable to translocate to the nucleus, blocking the inflammatory cascade (40). Supporting animal studies have further shown that glucosamine/chondroitin reduces downstream inflammatory markers as well as NFkB expression in the colon (19–24). Most, but not all, evidence suggests that this association may extend to humans, including evidence from two small human trials (27–29, 41–43).

This study was conducted in a very large prospective cohort, and one that enabled us to further delve into questions regarding the impact of screening and time on study findings. Even so, there are several important limitations to consider. First, there is a high prevalence of glucosamine use in this cohort (19.3% reporting use at baseline) and we do not have information on duration of use; it is therefore possible that this variable is capturing some short-term users, which could explain the weaker association observed here than elsewhere. While the questionnaire did specify regular use, it did not define regular use, and thus it is possible that participants are interpreting regular use more broadly than traditionally defined (4 or 5 days per week or more), which may increase measurement error and thus potentially attenuate associations. Second, the touchscreen questionnaire did not include a question on chondroitin, which is commonly coupled with glucosamine and taken as a single supplement. The prescription interview does have information on chondroitin (as it did not rely on preset list of supplements, unlike in the touchscreen interview) but unfortunately, we cannot present effect estimates from the interview on its own, as it appears to only include information on supplements if forgotten in the original touchscreen questionnaire, which did not ascertain use of chondroitin. It is therefore unclear what the interview data on its own would represent and therefore these data have not been presented. It should also be noted that we only have information on exposure/covariates at baseline, and thus not accounting for changes in use over follow-up could result in attenuation of effect estimates and/or residual confounding due to imperfectly measured covariates. To this end, we explored changes in over time in a small subset of participants for whom we had repeated exposure data; in this subset, nonusers largely remained nonusers; however, nearly half of users discontinued use by the first repeat measurement (and approximately 60% had discontinued by the subsequent assessment), indicating that we may very well have extensive measurement error due to use of a singular baseline measure. We also did not examine subsite-specific associations, but prior studies have not found any interesting or promising subsite-specific results (3, 5–7).

In summary, in this large cohort, we did not observe a statistically significant association between use of glucosamine and risk of colorectal cancer overall. However, our data reinforce prior exploratory findings highlighting an inverse association among individuals with no history of screening. This information is a key first step in informing efforts to evaluate glucosamine as a potential chemopreventive agent for colorectal cancer in unscreened individuals; however, further studies are needed to confirm the complex interplay of glucosamine use and screening, and how this may vary over time.

E.D. Kantor reports grants from NCI during the conduct of the study. P.S. Liang reports grants from Epigenomics, Freenome; and personal fees from Guardant Health outside the submitted work. M. Du reports grants from NCI during the conduct of the study. No disclosures were reported by the other authors.

E.D. Kantor: Conceptualization, data curation, supervision, investigation, methodology, writing–original draft, project administration. K. O'Connell: Data curation, formal analysis, investigation, methodology, writing–review and editing. P.S. Liang: Investigation, methodology, writing–review and editing. S.L. Navarro: Investigation, methodology, writing–review and editing. E.L. Giovannucci: Investigation, methodology, writing–review and editing. M. Du: Data curation, supervision, investigation, methodology, writing–review and editing.

This work is supported by grant P30 CA008748 from the NCI of the NIH (E.D. Kantor, K. O’Connell, and M. Du).

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.