Metabolic Syndrome and Risk of Colorectal Cancer: Results of Propensity Score-based Analyses in a Community-based Cohort Study

Background: This study aimed to determine the effects of metabolic syndrome (MetS) on colorectal cancer (CRC) using propensity score (PS) methods. Methods: The study subjects were 2,417 men and 4,568 women from the Korean National Cancer Center (KNCC) Community Cohort between 2003 and 2010 who were followed up by 2016. We calculated adjusted hazard risks (HRs) and 95% condence intervals (CIs) using unadjusted and multiple Cox's hazard regression models, PS matching analysis, regression models adjusted by the PS or stratied into ve strata according to PS, and PS weighting methods. Results: In women, MetS and abnormally high triglyceride (TG) levels were associated with an increased risk of CRC (HR for MetS, 2.11 [95% CI, 1.21-3.86] using the multiple Cox's hazard regression and 2.19 [95% CI, 1.10-4.36] using the PS matching analysis; HR for abnormal TG levels, 2.09 [95% CI, 1.21-3.62] using the multiple Cox's hazard regression and 2.10 [95% CI, 1.08-4.08] using the PS matching analysis). In men, MetS did not show a signicant association with the risk of incident CRC no matter which analyses we performed (HR, 0.92 [95% CI, 0.42, 2.04] using the PS matching analysis). Conclusions: Our study might provide additional evidence that deteriorated metabolic proles increase the risk of CRC in women rather than men. Thus, it may have an important role in effective population-level interventions for deteriorated metabolic proles at an early stage.


Background
Colorectal cancer (CRC) is the third most common cancer in both sexes worldwide (1,360,602 cases, 9.7% of the total cancer burden). According to GLOBOCAN 2018, the ranking of the estimated age-standardized rates of CRC incidence for both sexes at all cancer sites were observed to be similarly high ranked in most countries, ranking fourth in the United States, third in Europe, second in Japan, and third in the Republic of Korea [1].
Worldwide, over a billion people are known to be affected with Metabolic syndrome (MetS). The MetS prevalence are increasing in low socio-economic countries as well as high socio-economic countries. In even young adults, the prevalence ranges from 5 to 7% worldwide, although it increases with age. MetS is a complex disorder characterized by a cluster of moderate levels of metabolic, anthropometric, and hemodynamic abnormalities, accepted as a modi able risk factor CRC, although the mechanism linking MetS and CRC has not been clearly elucidted [2,3]. Recent systematic reviews and meta-analyses have conclusively reported that MetS is associated with an approximately 1.3-fold increased risk of CRC in both sexes, although the risk in women was slightly higher than that in men [2,4]. A few studies have been conducted in the Republic of Korea among national health insurance subscribers or subjects who underwent colonoscopy for a health examination in a hospital. However, these studies reported inconsistent results [5][6][7]. So far, there are limited evidences on the association between MetS and CRC, especially for the Asian population. Furthermore, the ndings of previous observational studies have pointes inevitably lower causality than randomized controlled trials due to selection biases. In order to overcome this weakness of observational studies, propensity score (PS)-based methods were proposed to attenuate selection biases by balancing many covariates [8,9]. Studies using these methods have been increasingly published in a wide range of elds, including some observational studies on MetS [10,11].
In this context, this study investigated the impact of MetS on CRC incidence by conducting PS-based analyses considering age, alcohol consumption, smoking, high animal fat intake, obesity, a lack of dietary ber intake, and a lack of physical activity etc. that have been identi ed as modi able risk factors for CRC [2,3], in a community-based prospective cohort in the Republic of Korea,

Methods
Data source andstudy population The Korean National Cancer Center (KNCC) Community Cohort, as a community-based prospective cohort, was conducted by the KNCC and included 16,304 men and women who resided in Changwon-si, Chuncheon-si, Chungju-si, Sancheong-gun, and Haman-gun in the Republic of Korea [12]. The questionnaire survey was conducted by well-trained interviewers and included the following demographic information: age, sex, home region, education level, occupational history, marriage status, average household income, alcohol consumption, smoking status, physical activity, dietary intake, history of cancer, and exposure to pesticides. Additionally, the results of anthropometric measurements and clinical laboratory examinations were included. All study participants are followed through linkage to the Korean Central Cancer Registry for cancer incidence and mortality by 2016. This cohort were linked to mortality data of Statistics Korea for all participants by 2016.
After the exclusion of 9,319 participants, including those who had no history of cancer before entry, did not participate in the nutrition survey, developed cancers other than CRC, and had missing data for MetS, alcohol consumption, smoking, physical activity, diet, and education, the remaining 8,658 participants (2,417 men and 4,568 women) were nally eligible for the analysis (Figure 1). The study was approved by the KNCC Institutional Review Board (IRB No. NCC2016-0300). All participants had provided written informed consent.

De nition of CRC and MetS
The outcome of this study was CRC incidence; the type of cancer was coded as C18, C19, and C20 according to the International Classi cation of Diseases 10th edition (ICD-10).
We used the National Cholesterol Education Program-Adult Treatment Panel III (NCEP-ATP III) de nition of MetS as a condition with at least three of the following ve risk factors concurrently: abdominal obesity (waist circumference (WC) >102 cm for men and >88 cm for women), hypertension (systolic blood pressure [SBP] ≥130 mmHg and diastolic blood pressure [DBP] ≥85 mmHg), low high-density lipoprotein (HDL) cholesterol level (<40 mg/dL for men and <50 mg/dL for women), high triglyceride (TG) level (≥150 mg/dL), and abnormal fasting blood sugar (FBS) level (≥110 mg/dL) [13]. In our study, we adopted the body mass index (BMI) ≥25 kg/m2 as a measure of abdominal obesity due to the absence of WC data.

Statistical analyses
Descriptive analyses between subjects who developed CRC and those who did not were performed for continuous parameters using the Mann-Whitney U test and for categorical parameters using the Chisquare test and Fisher's exact test.
To elucidate associations between MetS and risk of CRC in both sexes and an association between abnormal TG levels and CRC in women, hazard ratios (HRs) and 95% con dence intervals (CIs) were estimated by the following statistical analyses using Cox's hazard regression models: unadjusted and multiple regression, PS matching analysis, regression adjustment with the PS, and PS weighting methods. In these analyses, we included the following covariates known as the known risk factors of CRC [2,14,15] year]), moderate-intensity physical activity (days/week), frequency of fruit or vegetable intake (days/week), frequency of intake of beef or pork (days/month), education level (illiterate, middle school or less, high school, and college or more), and study area (Changwon-si, Chuncheon-si, Chungju-si, Sancheong-gun, and Haman-gun).
To perform PS-based methods, we needed rst to calculate the PS. The PS is the probability that an individual would have MetS or abnormal TG levels based on personal demographic and lifestyle information, and it was obtained from the t of a Cox's hazard regression model adjusted with all the covariates mentioned above.
To evaluate the balance in baseline characteristics in the dataset used for different regression models, we calculated the standardized mean differences, and values less than 0.1 were considered negligible differences. First, we applied a 1:1 case-control matching to the PS technique, which is the eighth digit to rst digit greedy matching method. This method may result in a drop-out of unmatched cases for the best matching, and thus, the number of subjects for the matching analyses was 1,038 men for MetS, 2,580 women for MetS, and 2,786 women for abnormal TG levels. Second, we performed weighted Cox's hazard regression models that were considered for an adjusted MetS effect after stratifying into 5 strata according to the quintiles of the estimated PS [16]. Additionally, a Cox's hazard regression model that used the PS as a covariate (a continuous variable and a categorical variable by quintiles) was a simple PS method similar to traditional regression analysis. Lastly, there are two major PS weighting methods referred to as standardization methods that depend on the establishment of a standard population. One is the inverse probability-of-treatment weighted (IPTW) model [17]. It considers the overall study participants as the standard population, and it uses weights of (

Results
In total, 2,417 male and 4,568 female participants were included in this study ( Figure 1). Table 1 shows the baseline characteristics of the study population. Among the study subjects, there were 57 men and 54 women newly diagnosed CRC after the entry of this cohort. The mean age of the study subjects was approximately 60 years old (data not shown), which is older than the general Korean population. CRC cases were older than non-cancer cases. Male participants had a high percentage of smokers and drinkers, whereas most of the female participants were non-smokers or non-drinkers. Five hundred twenty-four men (21.7%) and 1,297 (28.4%) women had at least three components of MetS at the entry of a cohort and CRC cases in both sexes were signi cantly more likely to have MetS, and those in women might have higher TG levels.
Tables S1 and Figure S1 show male participants' characteristics according to the presence of MetS as well as the degree of imbalance among covariates. Also, Tables S2 and S3 and Figure S2 show female participants' characteristics according to MetS and TG levels. A statistically negligible difference in covariates was found in the following datasets: the matched, strati ed and IPTW datasets for MetS in men; the matched and IPTW datasets for MetS in women; the matched, strati ed and IPTW datasets for increased TG level in women.
The result of various analytical methods for the effect of MetS on CRC risk in men is summarized in

Discussion
In this community-based cohort study in the Republic of Korea, we found that an increased risk of CRC in women associated with MetS and abnormal TG level by conducting PS-based analyses. In women, both MetS and abnormal TG level were associated with an approximately 2.0-fold to 2.2-fold increased risk of CRC. However, we showed that there was no signi cant association between MetS and CRC in men no matter which analytic methods we performed.
To our knowledge, our study is the rst cohort study using PS-based methods to examine the effect of MetS on CRC incidence for both sexes in the Asian population. It seems that previous research ndings  [21]. Our nding seems to be slightly different from previous studies above, as we have estimated using PS-based methods and study subjects of our study were just community residents. Thus, we expect that this study might contribute to providing additional evidence that there is the association between MetS and an elevated risk of CRC in the Asian population and future large cohort studies using PS-based methods could provide more de nitive evidence.
In addition, our study showed that abnormal TG levels were associated with CRC risk in women. However, a previous review and meta-analysis reported that TG was not related to the risk of CRC [4]. Cohort studies in the United States and Japan also arrived at similar conclusions [19,22] [21]. Future studies are recommended as it remains controversial whether the risk of CRC or colorectal adenoma is elevated when individuals' levels of TG are 150 mg/dL or higher.
As is well known, MetS is widely known as a risk factor for CRC, but the biological mechanism underlying this association remains to be clari ed. Insulin resistance, systemic in ammation, and oxidative stress, and higher leptin levels have been suggested as potential mechanisms that may explain the association between MetS and CRC [3]. Insulin increases cell proliferation and reduces apoptosis, which may lead to tumor development. Insulin also induces the overstimulation of receptors of insulin-like growth factor-1 and 2 (IGF-1 and IGF-2), a key promoter of tumor development. Besides, deteriorated metabolic status in uences elevated levels of in ammatory cytokines such as interleukin 6 (IL-6), tumor necrosis factor alpha (TNF-α), C-reactive protein (CRP), and leptin hormone, which may be implicated in insulin resistance and tumor development. Further studies are required to elucidate the mechanism underlying the effect of each component of MetS on CRC.
The strengths of this study include its prospective nature (i.e., cohort design), the strong causality between MetS and CRC by using PS-based methods. The selection biases that are present in most observational studies may lead to a lack of causality in this study. To improve the causality between MetS and CRC incidence, we performed PS-based analyses in a community-based cohort. A major advantage of using PS-based methods in observational studies can minimize selection biases by balancing nonrandomized individuals' data to reach the level of causality determined by randomized controlled trials. Recently, there have been some well-designed studies that have revealed associations between MetS and non-communicable diseases, including cancers, using PS-based analyses [23, 24].
There were several limitations to this study. We could not observe the following confounders: the history of MetS before cohort entry, individuals' stressful events, menopause, the consumption of carbohydrate and starchy foods, etc. Also, there is the possibility of information bias due to the use of a self-reported questionnaire. Also, we measured MetS only at the entry of this cohort study, so we could not estimate the risk of CRC due to changes in MetS over time. Furthermore, we used BMI as a measure of abdominal obesity of MetS due to the absence of WC data, although BMI and WC have slightly difference in the pathological meaning of MetS. Last, we could not obtain mortality data between 2014 and 2016 for noncancer cases among the study population from Statistics Korea, so we could not calculate a time-to-event outcome of performing survival analyses.
In conclusion, this study may provide additional evidence that deteriorated metabolic pro les increase the risk of CRC in women. We highlight the importance of effective population-level interventions for deteriorated metabolic pro les at the early stages.  SMRW, standardized mortality ratio weighted. a) Adjusted by age, education, smoking status, alcohol consumption, physical activity, frequency of intake of fruits or vegetables, frequency of intake of red meats, and residential area. HR, hazard ratio; CI, confidence interval; PS, propensity score; IPTW, inverse probability-of-treatment weighted; SMRW, standardized mortality ratio weighted. a) Adjusted by age, education, smoking status, alcohol consumption, physical activity, frequency of intake of fruits or vegetables, frequency of intake of red meats, and residential area. Figure 1 Flow Diagram of the Derivation of the Study Population.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.