Risk and survival in colorectal cancer with increasing body mass index: A nationwide population- based cohort study

Aim: The aim was to explore potential associations between the body mass index (BMI) and the risk of colorectal cancer (CRC), including subsites of the colon, and cancer- specific death. Methods: A registry- based cohort study was conducted with baseline data gathered from the Norwegian Tuberculosis Screening Programme, collected between 1963 and 1975, and linked to follow- up data from the Cancer Registry of Norway and the Norwegian Cause of Death Registry. Cox regression models were used to explore associations between BMI and CRC risk and cancer- specific death. Results: Of 1 723 692 included individuals, 76 616 developed CRC during 55 370 707 person- years of follow- up. In men, a 5 kg/m 2 increase in BMI was associated with an in - creased risk of colon cancer, including both right and left subsites, and rectal cancer. Allowing for nonlinearities, we found a U- shaped association for the right colon and an in - verse U- shape for the left colon and rectum cancer. In women, a 5 kg/m 2 increase in BMI in early adulthood was associated with increased risk of colon cancer, including both subsites. In women, an increased risk of CRC death with increasing BMI was found for colon cancer. Conclusions: Men of all ages have an increased risk of CRC with increasing BMI, with the highest risk for right- sided colon cancer. An increased risk for colon cancer was also found in women with high BMI in early adulthood. Furthermore, women of all age groups appeared to have an increased

Recently, attention has been given to the 'obesity paradox in cancer', a phenomenon characterized by reduced cancer death combined with an increased risk of developing cancer in subjects with obesity [15,16]. This paradox has also been observed in the context of CRC, but most such research has limited sample sizes and short-term follow-up, which makes interpretation challenging, particularly for CRC subsite analyses.
There is thus a need for larger studies with longer follow-up [12].
The aim of this study was to investigate the association between BMI and risk of CRC, including colon subsites and stratified by gender. Furthermore, CRC tumour stage and cancer-specific death with regard to BMI were explored. The study is based on the largest population cohort available in Norway, consisting of almost 2 million out of 4 million inhabitants at the time of study inclusion and a maximum follow-up of up to 50 years [17].

Study design
A registry-based cohort study with baseline data from the Norwegian Tuberculosis Screening Programme (NTSP) and followup data from the Cancer Registry of Norway (CRN), the Norwegian Cause of Death Registry and the National Population Register was performed. The study adheres to the STROBE guidelines for reporting of observational studies [18].

The Norwegian Tuberculosis Screening Programme
The NTSP commenced during 1943. A nationwide mass survey was performed between 1963 and 1975 where height and weight were registered electronically in all participants using the unique 11-digit personal identification number assigned to Norwegian residents [17,19,20]. The NTSP data are owned and administered by the Norwegian Institute of Public Health [20].

Data from Norwegian population-based registries
Registration of cancer diagnoses in the CRN is mandated by law in Norway and the registry has a documented high completeness [21]. The different data sources in this project were linked using the unique 11-digit personal identification number allocated to all Norwegian residents.

Study cohort and outcomes
All individuals aged 16-75 years at the time of the NTSP screening were included ( Figure 1). BMI was calculated based on measured body height and weight (kg/m 2 ). Persons with missing data on height or weight were excluded, as were persons with short stature, defined as women <150 cm and men <161 cm [22]. Individuals diagnosed with cancer within 1 year of NTSP screening were also excluded. Few patients had a BMI under 15 or over 50 (0.02% of the total cohort). These individuals were excluded to reduce statistical uncertainty and avoid model extrapolation.
Colon and rectal cancers were classified using the ICD-10 lexicon and subclassified as right-or left-sided colon using the ICD-0-3 lexicon. Only adenocarcinomas were included (Appendix S1). Based on clinical and pathological data from CRN records, disease stage at diagnosis was categorized as localized, regional metastatic (regional lymph nodes) or distant organ metastatic using the Union for International Cancer Control (UICC) lexicon [23].
The study cohort was categorized into three periods based on screening year to account for potential changes in BMI over time: period 1, 1963-1967; period 2, 1968-1971; and period 3, 1972-1975. To investigate whether BMI at a given age affected the outcomes of interest, the cohort was also grouped by age at screening: age group 1, 16-29 years; age group 2, 30-49 years; and age group 3, 50-75 years.
For analyses focused on CRC incidence, individuals were followed from the date of screening until whichever occurred first: CRC diagnosis, death, turning 75 years old, or administrative censoring (31 December 2018). For analyses focused on disease stage at diagnosis, we followed individuals until a CRC diagnosis with a given stage.
Individuals were censored at CRC diagnosis if their disease stage was not the stage of interest. Time since birth was used as the time scale for follow-up. Thus, regarding incidence, there was no need for explicit adjustment for age because this ensured that we always compared

What does this paper add to the literature?
The long follow-up time with high number of cancer cases together with exceptional quality of data has enabled a unique and novel in-depth analysis of the association between body mass index and colorectal cancer with regard to gender, age, cancer stage and subsite.
groups of people with the same attained age. For survival analyses, patients were followed from the date of their CRC diagnosis to CRC death or date of administrative censoring, whichever occurred first. Individuals who died due to any other cause were censored at death.
In these analyses, time since diagnosis was used as the time scale.
The effect of BMI changing over time was evaluated in a subgroup of patients with two BMI measurements. For these patients, the first measurement occurred at the NTSP screening and the second at the time of surgery registered in the CRN (after 2016).

Statistical analysis
Standard descriptive statistics were presented using mean and standard deviations for continuous variables and absolute and relative frequencies for categorical variables. The association between BMI and CRC incidence was modelled using multivariable Cox proportional hazards regression models with adjusted age groups as defined at screening as a categorical variable. In the main analyses BMI was handled as a continuous variable, assuming log-linearity, allowing us to present estimated hazard ratios (HRs) with 95% CI per 5 kg/m 2 increase in BMI. To explore potential nonlinear effects, BMI was modelled using restricted cubic splines with 5 df adjusted for categorical age at screening [5]. When investigating the effect of BMI across age groups we fitted models with interaction terms between BMI and age groups.
Similarly, the effect of BMI on CRC-specific death was estimated using multivariable Cox regression models under the assumption of log-linearity of BMI, adjusted for categorical age at screening, continuous age at diagnosis, time period of diagnosis and UICC cancer stage. Effects of BMI on cancer risk and cancer-specific death were analysed separately in men and women.
Population attributable risk analyses were conducted using the formula provided in the supplementary appendix in an article published by Bhaskaran et al. [5]. The estimation of the risk of CRC cancer related to BMI was categorized into three groups based on WHO classifications-underweight (<18.5), normal weight (18.5-24.9) and overweight (>25)-and adjusted for age group at screening.
Estimates of the prevalence of underweight and overweight were obtained from Statistics Norway [24].
All statistical analyses were performed using Stata version 16.1 [25]. The significance level was set to 5% throughout.

Ethics and approvals
This study was approved by the Regional Committee for Medical and

RE SULTS
A total of 1 911 598 individuals were enrolled in the NTSP between 1963 and 1975. After excluding non-eligible participants, the study cohort consisted of 1 723 692 individuals of whom 894 611 were women (51.9%) ( Figure 1).
Characteristics of the study cohort are given in Table 1. Mean age and BMI did not change over time. The mean ages at initial BMI measurement and diagnosis across CRC subsites and age groups are given in Table S2.  (Table S1).

Colorectal cancer risk
In men, a 5 kg/m 2 increase in BMI was associated with an increased  Figure S1). For right-sided colon cancer all age groups presented a U-shaped association of BMI and CRC cancer risk, as did age group 3 (50-75 years) in left-sided cancer.
In women, there were no clear indications of nonlinear effects of BMI except within age group 1 (16-29 years) who exhibited a close to linear increase in risk of colon cancer. Figure 4 shows the relative risk of presenting with specific UICC stages of CRC per 5 kg/m 2 increase in BMI. For men, the risk of presenting with any stages of colon cancer increased with higher BMI.

Cancer stage
The effect was consistent across all stages, most BMI measurement age groups at time of NTSP screening and subsites. For men, the risk of rectal cancer was slightly increased for all stages, but this was

Cancer-specific death
There was no association between BMI and CRC-specific death in men, regardless of subtype and age group at time of BMI measurement. However, for women there was an increased risk of

DISCUSS ION
This study including 76 616 CRC cases extends the available evidence about the associations between increasing BMI and risk of CRC. Our findings also indicate an increased CRC-specific death for women with higher BMI.

Colorectal cancer risk
In men, the increased CRC risk was most prominent for colon cancer and was consistent in both linear and nonlinear models. Although previous studies indicate comparable findings regarding CRC subtypes and age groups, to the best of our knowledge these associations have F I G U R E 2 Forest plots showing the hazard ratio (HR) with 95% CI for colorectal cancer by subsite and age group for men (A) and women (B). A 5 kg/ m 2 increase in body mass index (BMI) is shown for each sex and BMI is fitted as a linear effect. Number of cases is given in parentheses.
not been observed within the same cohort. Previous reports suggest a weaker association between right-sided colon cancer and BMI compared to left-sided cancer [9]. Our results indicate that the strongest association is between increasing BMI and right-sided colon cancer in men, especially when allowing for nonlinear effects of BMI ( Figure 3).
Under this assumption, the results for left-sided colon cancer correspond to the curve for rectal cancer, whereas for right-sided cancer the curve shape is opposite and U-shaped.
For women, a novel positive association for colon cancer in age group 1 (16-29 years) was observed in both linear and nonlinear models. This finding resembles the risk found for men and was present regardless of subsite (Figures 3 and S1).
In a study of 109 771 health professionals, women with high body fatness in early life had an increased risk of CRC [26]. This was a selected cohort with limited sample size using recall registration of early body fatness. Another study did not find any association between BMI in early adulthood (age 25) and CRC cancer incidence and death [27]. However, colon and rectal cancer was analysed combined and recall weights from 20 to 40 years earlier were applied. A recent study reported an association between early adulthood obesity and early onset of CRC [28]. However, only patients with cancer diagnoses before the age of 50 were included. The Nurses' Health Study II observed an increased risk of early onset CRC (diagnosis before age 50) in women with obesity (BMI > 30) at age 25-42 [28].
In our study the mean age at initial BMI screening for women who developed colon cancer in age group 1 (16-29 years) was 23.4 (SD 4.0) years, while the mean age at time of CRC diagnosis for the same group was 62.5 (SD 9.3) years (Table S2). We found no elevated risk for early onset colon cancer in women with high BMI in early adulthood. In fact, we found no association between BMI and early or very early onset of CRC (diagnosis before age 30) (data not shown).
The overall findings of increased risk for developing CRC with increasing BMI for men but not for women are consistent with previous reports [5,7,[9][10][11]. However, the pathophysiology linking obesity to CRC remains obscure [11,28]. Suggested mechanisms include increased plasma insulin and hence insulin-like growth factor 1, low grade chronic inflammation, or increased levels of sex hormones and the hormone leptin [29,30]. These mechanisms may be age and gender dependent. For example, oestrogen may protect against cancer development and is lower in obese than non-obese premenopausal women [29,31,32]. After menopause, higher levels of oestrogen are observed in obese compared to non-obese women [33]. This might partially explain why obesity does not increase the risk of CRC in postmenopausal women.
Furthermore, several studies show differences between right-sided and left-sided colon cancer regarding microsatellite instability and mutations in several cancer-related genes, including BRAF and KRAS [34,35]. A difference in these markers between obese and non-obese individuals has been shown and may in part explain our finding of a stronger relationship between obesity and right-sided cancer in men [36,37]. A recent publication has also implicated obesity-related alterations in gut hormones in cancer pathogenesis [38]. However, future studies are needed to explore these and other potential interactions.

Cancer stage
A hypothesis of an increased risk for higher UICC stages of CRC with increasing BMI was made, with BMI acting as a promoter speeding up the cancer process. However, our findings of the same risk across all UICC stages did not support this hypothesis. BMI may rather be an initiator, increasing the probability of cancer development, a hypothesis that should be explored further.

Cancer-specific death
Several studies indicate favourable survival in cancer patients with higher BMI in parallel with a suggested increased risk of developing F I G U R E 3 Age-adjusted hazard ratio (HR) for colorectal cancer by subsite according to body mass index (BMI). Curves are plotted for men (A-D) and women (E-H) with 95% confidence intervals, allowing for nonlinear effects modelled using restricted cubic splines. The reference BMI fitted as a spline with HR fixed as 1.0 is 22 kg/m 2 .
cancer [39,40]. The 'obesity paradox in cancer' has been related to methodological limitations like confounding, reverse causation and stratification biases and needs further clarification. The inadequacy of BMI as a measure of body fatness and body composition may also contribute [15,16].
Our study did not support the 'obesity paradox in cancer' as it is typically defined. In men, we found an increased risk of developing CRC but no significant risk of CRC-specific death with increasing BMI. In women we did not observe an increased risk of developing CRC with increasing BMI at large; however, we did find an increased risk of colon cancer-specific death. Women with high BMI in early adulthood had an increased risk both for developing colon cancer and for colon cancer-specific death.

Public health relevance
In this study cohort, 8.9% were defined as obese at the time of their initial BMI measurement ( Table 1). Between 1975 and 2016, the mean BMI in the Norwegian population increased from 24.6 to 27.5 kg/m 2 [2]. This was similar to our finding of increase in BMI in a subgroup of our cohort. Currently, 23% of the Norwegian individuals are obese and estimates suggest that these numbers are increasing [2,41]. In parallel with the increased proportion of overweight and obese adults, there has been a 2.2% increase in CRC and 2.8% increase in BMI-related colon cancer in men in Norway.
Our findings are in line with earlier publications [42]. These findings are relevant for strategies aimed at reducing the adverse public F I G U R E 4 Forest plots showing hazard ratio (HR) for colorectal cancer by subsite and age group. Plots are shown for men (A-C) and women (D-F) with 95% confidence intervals where colon cancer subsites are stratified according to the UICC classification* in Stages I/ II, Stage III and Stage IV, per 5 kg/m 2 increase in body mass index (BMI). Number of cases is given in parentheses. *Union for International Cancer Control (UICC), 8th edition, definition of colorectal cancer staging: Stage I/II, no metastasis; Stage III, regional lymph node metastasis; Stage IV, metastases to distant organs.
health effects of obesity. Such strategies could include the use of bariatric surgery, where a recent paper showed that obese patients had the same risk of CRC as the general population following bariatric surgery [43].

CON CLUS ION
In this unprecedented, large, single cohort study, we showed that the association between BMI and CRC varied between genders.
We found an increased risk for CRC in men of all age groups with increasing BMI, with a probable highest risk for right-sided colon cancer. In women, an association was found between higher BMI in young adulthood and risk of colon cancer. Women of all age groups appeared to have an increased risk of CRC death with higher BMI. The manuscript has not been and will not be a podium or poster meeting presentation.

CO N FLI C T O F I NTE R E S T
No conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the Cancer registry of Norway and Norwegian Institute of Public Health. Restrictions apply to the availability of these data, which were used under license for this study. Data are available from corresponding author with the permission of a third party.

E TH I C A L A PPROVA L
The study was approved by the Regional Committee for Medical and Health Research in Southeastern Norway.