Waist circumference and risk of 23 site-specific cancers: a population-based cohort study of Korean adults

Background Large waist circumference (WC) is a risk factor for several site-specific cancers, but a large-scale systematic investigation across all common cancers adjusted for potential confounders has not been conducted. This study aimed to evaluate the possible links between WC and common cancers. Methods We prospectively examined the association between WC and the risk of cancers in a 7-year cohort study of nearly 22.9 million Korean adults. Using the claims database merged with the national health check-up data, we fitted proportional hazard models to investigate associations between WC and 23 of the most common cancers, with adjustment for potential confounders, including body mass index (BMI). We also evaluated the modification of BMI on the relationships between WC and the incidence of cancer. Results A total of 769,871 cancer cases were identified. WC was positively associated with 18 of 23 cancers, and the effects varied substantially by site in each sex. The modification of BMI on the WC-cancer association also varied across the cancer site; in most cases it mitigated the association. For cancers of the oral cavity, larynx, oesophagus, lung, and premenopausal breast, the BMI adjustment reversed the association toward being positive (all Ptrend < 0.001). Conclusions Central obesity, independent of general obesity, was associated with the risk of several cancers. The heterogeneity in the mediating effects of BMI suggests that different mechanisms are associated with different cancer sites. Based upon these findings, active strategies to monitor and prevent central obesity should be implemented.


INTRODUCTION
Obesity is an established risk factor for mortality and various chronic diseases. A pooled analysis of 57 prospective studies has documented that each 5 kg/m 2 higher body mass index (BMI) was associated with a 29% increase in mortality 1 . Furthermore, results from the EPIC (European Prospective Investigation into Cancer and Nutrition) emphasised the additive use of the central obesity index on predicting the risk of death 2 . Given the inexorable rise in obesity worldwide in recent years 3 , understanding the effects of obesity on major health outcomes is urgent and has not been sufficiently emphasised.
There is increasing evidence that obesity increases the risk of certain types of cancers 4,5 . However, most researchers have emphasised the role of overall obesity (generally assessed as BMI), rather than central obesity, in the etiologies of these malignancies 6 . Because metabolic derangement supported by insulin and insulin-like growth factor (IGF) is a possible mechanism in carcinogenesis 7 , central obesity may present a higher threat to the risk of cancer than to general obesity. Moreover, with regard to tumour angiogenesis and cell proliferation, intra-abdominal fat has been hypothesised to be biologically different from fat in other areas 8,9 . Comprehensive reviews have recently suggested that central obesity increases the cancer risk in various sites, including the gastrointestinal tract [10][11][12] , biliary tract 13 , lung 14 , breast 15 , thyroid 16 , head and neck 17 , and genital tract [18][19][20] . However, there are important limitations to these studies. Individual studies have often had insufficient power, and potential confounders have been inconsistent across studies. In addition, many studies have used self-reported WC data, which probably underestimated the true WC, and there have been few reports of the effects of BMI adjustment on the associations between WC and cancer risk 21,22 . Furthermore, cumulative meta-analyses of observational studies have inherent limitations, including publication biases and heterogeneity 23,24 .
Our aim was therefore to investigate the possible associations between WC and the most common site-specific cancers in a single population using a large-scale nationwide claims database. To better understand the role of abdominal obesity on cancer development, we additionally examined the effects of BMI adjustment on the WC-cancer association.

Design and participants
We used the Korean nationwide claims database and biennial medical examination data provided by the Korean National Health Insurance Corporation (NHIC). In the Republic of Korea, comprehensive medical care of nearly all Koreans (97%) is covered by single insurance from the NHIC. The information contained the enrollees' demographics, utilisation of medical facilities, disease code registered by clinicians, and pharmacy dispensing claims. Data from medical examinations included health-related surveys using standardised questionnaires, height, weight, blood pressure, and fasting laboratory findings, such as serum glucose and total cholesterol. The WC measurements were added in 2009. Using the survey dataset, we extracted information about the medical history and health-related habits such as smoking, alcohol consumption, and physical activity. The questionnaires, which were reviewed by a trained staff, were self-reported.
BMI and WC were used as an index of overall and central adiposity, respectively. The data on anthropometry were collected by direct measurements at medical institutions equipped with facilities and staff approved by the regulations defined by the KNHIC 28 . The categories of WC were based on quintiles in our cohort. BMI was calculated as weight in kilograms divided by the square of height in meters (kg/m 2 ). Individuals were categorised into four groups according to BMI following the World Health Organization recommendations for Asians as follows: underweight, <18.5; normal, 18.5-22.9; overweight, 23.0-24.9; and obese, ≥25.0 29 .
Menopausal status was set at the age of 50 years, representing the usual menopausal age of Korean women 30 . Socioeconomic status, estimated by the average insurance premium per month, was classified into quartiles. Smoking status was categorised into three groups as a non-, former, or current smoker. Alcohol consumption was categorised into three groups as a non-, moderate (<30 g per day), or heavy drinker (≥30 g per day). Regular physical activity was defined as exercise ≥one session per week. A comorbid condition was defined primarily based on the combination of past history and the use of ≥1 drugs for the corresponding disease, which included hypertension (ICD-10 code, I10-13/15), type 2 diabetes (E11- 14), and hyperlipidemia (E78). Individuals who had abnormal findings in the health examination were considered as patients with the corresponding parameters: ≥140/90 mmHg of blood pressure, 126 mg/dL of fasting plasma glucose, and 240 mg/dL of total cholesterol. Statistical analysis Analyses were performed separately by sex, as appropriate. Cox proportional hazards models, with attained age as the underlying time metric, estimated the hazard ratio (HR) and 99% confidence interval (CI) for the associations of WC with cancer risk, considering potential confounders such as age (continuous), three health-related habits, three co-morbidities, and the BMI (continuous). In the Cox model, smoking status and alcohol use were included as 3 categories, respectively. The follow-up period began on the date of anthropometric assessment. In addition, the WC results were presented with and without adjustment for the BMI. All Cox models were tested for and met the proportional hazards assumption. SAS software, version 9.4 (SAS Institute, Cary, NC, USA) was used for all statistical analyses. A two-sided p-value < 0.05 was considered statistically significant. Table 1 lists the baseline characteristics of participants across the WC quintiles by sex. Centrally obese individuals were more likely to be elderly and to have a high BMI and increased comorbidites. Current smoking was inversely associated with WC categories, and heavy drinkers were more common in centrally obese males and centrally lean females. Table 2 shows the HRs for developing cancers compared with individuals with the lowest WC (quintile 1), after controlling for confounding factors including the BMI. Figure 1 also depicts the forest plot for each cancer across WC quintiles before and after BMI adjustment. The data confirmed that in the Korean population, central obesity strongly increased the risk of cancer in the stomach, colorectum, hepatobiliary system, kidney, thyroid, brain, and lymphoma in a dose-dependent manner (P trend < 0.001 for both sexes). Gender specific differences in significance were notable in some cancers, with more robust associations in males for cancers of the head and neck, oesophagus, pancreas, lung, bladder and skin, and more robust correlations in females for multiple myeloma. There were also significant dose-dependent relationships between WC and cancers of the prostate (P trend < 0.001) and breast (premenopausal, P trend = 0.007; postmenopausal, inversely P trend = 0.035), but associations with other genital tract cancers were not reach the significant. The BMI did not affect or attenuated the WC-cancer association for most cancers. However, for cancers of the oral cavity, larynx, oesophagus, and lung in male (all P trend < 0.001) and premenopausal breast (P trend = 0.007), the BMI adjustment reversed the association toward being positive.

RESULTS
Selected sensitivity analyses are shown in the supplementary tables. Because smoking has an inverse relationship with obesity and is a well-known risk factor for many cancers, we analysed data for those who had never smoked. The overall trend and its significance involving the WC-cancer association remained in smoking-related cancer, while the significance diminished in some cancers (Table S1). To examine the effect of preclinical cancers that may cause weight loss and thus bias the association between obesity and cancer, we repeated the analyses after excluding cancer occurring in patients within the first 2 years of follow up, but found no significant difference (Table S2).

DISCUSSION
Continuous updates of the scientific literature have supported the association between high WC and the risk of cancer. In a single dataset, we determined the associations between WC and the incidence of cancer, when considering potential confounders including the BMI. To the best of our knowledge, this is the largest study to estimate the effects of WC-cancer associations across a range of sites. In addition, studies on this issue have been conducted primarily in Western populations. Asians, including Koreans, tend to have relatively small body frames and condensed body fat. Compared with Caucasians, they have a higher body fat percentage for a given BMI 29 . Moreover, they have higher amounts of abdominal adipose tissue 31,32 and lower muscle mass for a given BMI, leading to a greater tendency toward central obesity and to more susceptible to insulin resistance 33,34 . Thus, the effects of abdominal obesity on the development of cancer in Asians might differ from those in Caucasians.
Several mechanisms have been proposed to explain links between adiposity and increased cancer risk involving sex hormone metabolism, insulin and IFG signalling, and adipokine pathophysiology 35 . However, excess body fat is a heterogeneous condition in which individuals with similar BMIs may have distinct cancer risks. Our results provide a potential explanation for the risk differential that persists after accounting for BMI. Few studies have conducted further adjustments between the central adiposity index and BMI to clarify their independent roles in the risk of cancers.
It is currently recognised that a proportion of obese individuals might not be at an increased risk for metabolic complications of obesity. Recent studies have suggested a correlation between metabolic health and cancer outcomes 36,37 . Metabolically active visceral fat releases substantial amounts of growth factors, inflammatory markers, free fatty acids, and locally produced oestrogen and adipokines, which might contribute to the development of cancer 38 . However, WC not only distinguishes lean from adipose tissue, but reflects adipose tissues in both subcutaneous and visceral areas, which obscure any separate roles of compartments in determining the cancer risk. The most comprehensive quantitative modalities will come from largescale imaging (i.e., abdominal fat computed tomography and dual-energy X-ray absorptiometry) projects, which should result in the identification of dominant mechanistic pathways.
We identified several cancers related to central obesity that were independent of general adiposity. Overall, this finding was consistent with those from recent studies of cancers of the gastrointestinal tract 10,39,40 , lung 14 , hepato-pancreato-biliary system 13,41-43 , head and neck 17 , and kidney 40 . Cancers of the thyroid 16 , prostate 20,44 , bladder 45 , and skin 46 , which had conflicting or non-significant results, also had significantly positive dosedependent relationships with increasing WC. Non-significant results in prior studies might be simply due to relatively small sample sizes, as their associations were similar to ours, and the    associations reached significance before adjusting for the BMI. However, for prostate cancer, we were unable to explain the disagreement with previous results, because the association between body adiposity and incident prostate cancer depends mainly on the stage or subtype of prostate cancer 20 . Prostate cancer has a large clinical heterogeneity, which ranges from microscopic, well-differentiated indolent tumours to aggressive and lethal diseases. However, in the present study, we could not obtain this detailed clinical information. We did find a new association between WC and lymphoma/ brain tumours, which has not been previously reported because the BMI was not adjusted in previous studies 47,48 . In contrast, we could not find an association for leukaemia that has previously been reported to be associated with central adiposity 49 . For multiple myeloma, while a recent Mendelian randomisation study reported a non-significant (p = 0.06) reverse association of the BMI-adjusted WC 50 , we found a positive association in female (P trend = 0.003). Given the largely unknown etiology and the heterogeneous entity of these malignancies, further studies are warranted to clarify their associations with central adiposity.
We found differential associations of WC by sex with the risks of some cancers, but this might have been largely because of the differences in sample size (i.e., cancers of larynx, oesophagus, and bladder). Besides the sample size gap, a male-specific association with malignant melanoma was also found. Malignant melanoma is the most aggressive form of skin cancer, which is considered the fastest growing cancer 51 . Overall, a link between malignant melanoma and general adiposity remains unclear 52 , although a recent meta-analysis reported a male-specific effect 53 . To the best our knowledge, only a single study investigated the impact of central obesity when considering the BMI, with no convincing associations reported 46 . A possible explanation is that increased body surface in males may simply denote a larger surface at risk for sunlight exposure, thus providing an increased association with the incidences of melanoma 53 . However, our sensitivity analyses showed that the positive association between WC and melanoma in males was largely affected by the smoking status, and not by the BMI. Smoking may be at least a strong modifier in the WC-melanoma associations, even though it is not a risk factor for melanoma 54 . There has also been growing evidence that smoking is associated with abdominal fat accumulation 55,56 . In the present study, positive associations were mostly attenuated after BMI adjustment; however, notable reversions toward positive associations were found in cancers of the oral cavity, larynx, oesophagus, and lung. While the mean BMI values are generally lower in smokers than in non-smokers 57 , cigarette smoking has been positively associated with central adiposity 55,58 . Our additional analyses in non-smokers revealed an unaltered direction or significance of association. This means that metabolic derangements, represented by central adiposity, are still responsible for increased risk in such cancers, even when excluding the effect of smoking. In this regard, obtaining a WC measurement together with BMI may provide essential information that might not be feasible when assessing each parameter separately.
In a similar manner, it was noteworthy that among oestrogendriven cancers, BMI-adjusted WC predicted the risk of premenopausal breast cancer. Before menopause, plasma levels of oestrogen were not directly related to general adiposity, and obese premenopausal females had lower estradiol levels because they are more likely to have anovulatory cycles 59 . Thus, the positive association with WC in this study cannot be explained by the sex hormone hypothesis, which was supported by the finding that central obesity was associated with an increased risk of oestrogen receptor-negative breast cancer in premenopausal women 60 . Central obesity is a well-known indicator of hyperinsulinemia and higher levels of IGF-1 that are related to premenopausal breast cancer risk 61 . Overall, our results indicated that metabolic conditions may be more important than hormonal mechanisms in premenopausal breast cancer.
Occult cancers are an important type of cancer to be considered. To eliminate bias, we performed sensitivity analyses excluding an incident cancer within the initial 2 years of follow up; however, we could not find any significant differences. We assumed that this was because many of the occult cancer patients were diagnosed and excluded through their health status examination near the baseline. In addition, the general perception is that obesity does not initiate cancer, but rather promotes cancer in clinical presentations over several years. The precise time lag between development and duration of obesity and the occurrence of cancer is still not well established.
There are some limitations in this study. Our study was confined to individuals who took a health examination, so those without WC data were excluded, introducing the possibility of selection bias. Based on the 2009 Korea National Health and Nutrition Examination Survey data 62 , a prevalence of obesity/smoker was 35.8%/47% for male and 26%/7.1% for female, respectively: these are comparable to our results (37.3%/44.9% for male and 26.6%/ 4.1% for female). Moreover, our main conclusions are unlikely to have been seriously affected, because we investigated the WCcancer relation itself. A second limitation was the potential for residual confounding that we could not consider. Our outcome data were insufficiently detailed to explore potential differences between cancer subtypes, in particular for oesophageal cancer 6 , lung cancer 63 , breast cancer 64 , prostate cancer 20 , lymphoma 65 , and colon cancer 66 . We also had no detailed information on critical risk factors for some cancers, such as female reproductive factors for oestrogen-dependent tumours, viral hepatitis information for liver cancer, ultraviolet exposures for melanoma, and amounts or forms of tobacco consumption for lung cancer. Mendelian randomisation may offer a solution to the problem of residual confounding, under certain conditions 67 . A final limitation was the assumption of an unchangeable obesity index. We assessed only baseline measures, so these single measurements may not have reflected changes that occurred during the follow up. Previous studies 68,69 also emphasised the importance of time duration with a high BMI and its association with cancer development, in a similar manner to other diseases 70 . Longer follow up with sequential measurements is needed to more fully support a causal association.
Despite these limitations, our study has methodological strengths. As a longitudinal, nationwide, population-based cohort study, the size of the dataset was close to the entire population of adult Koreans. Compulsory health examinations also allowed us to use direct anthropometric measurements, which are preferred over self-reported data 71 . To date, no other study has used direct measurements of WC in such a large sample size for a wide range of cancer sites. Our results provide evidence on the independent role of central obesity as a predictor for cancer incidence, even in a relatively lean Asian population. However, this result is only indirect evidence that the interventional effect of WC reduction will decrease the risk of cancer. Heterogeneity in the effects of BMI also suggests that there are different mechanisms or combinations of mechanisms associated with different sites. Integration of experimental research on underlying mechanisms linking central obesity to the identified cancers in our study is necessary to confirm our conclusions.