Effect of Metabolic Health and Obesity Phenotype on the Risk of Pancreatic Cancer: A Nationwide Population-Based Cohort Study

This article is featured in Highlights of This Issue, p. 427

Pancreatic cancer is a major cause of cancer-related death following lung and colorectal cancers in the United States (1) due to late diagnosis and early metastasis (2). The worldwide 5-year survival rate in patients with pancreatic cancer remains low (approximately 6%; ref. 3). In the United States, pancreatic cancer incidence is rapidly increasing, and pancreatic cancer is expected to be the second major cause of cancer-related death in 2030 after lung cancer (4). Thus, identifying and controlling the potential risk factors are necessary to decrease pancreatic cancer incidence and improve its prognosis.

Obesity, represented by a high body mass index (BMI), is suggested to be significantly associated with pancreatic cancer (5–7). Although obesity is associated with incident pancreatic cancer, an important question remains: is obesity a direct causal factor of pancreatic cancer? Obesity is often accompanied by metabolically unhealthy status, defined as metabolic syndrome, which has similar pathophysiologic mechanisms including inflammation, oxidative stress, insulin resistance, and hyperinsulinemia (8). Previous studies have suggested that these mechanisms play a pivotal role in pancreatic cancer development and prognosis (7, 9). Recent studies have identified a unique subset of metabolic obesity phenotypes termed metabolically healthy obese (MHO), which represents obesity without cardiometabolic risk profiles. On the other hand, metabolically unhealthy normal weight (MUNW) displayed increased insulin resistance, inflammation, and other metabolic syndrome features despite normal body weight. A recent study demonstrated an inverse relationship between the metabolically healthy overweight/obese phenotype and breast cancer incidence (10). Individuals with the metabolically unhealthy obese (MUO) phenotype were reported to have an increased gastric cancer risk, but not those with the MHO phenotype (11). However, another study reported that obesity itself plays a more important role in colorectal carcinogenesis, although men in the MUNW group had a high colorectal cancer risk (12). Obesity was also associated with an increased risk of incident thyroid cancer in both metabolically healthy and unhealthy men (13).

Despite the pathophysiologic importance and potential clinical implications of the associations between metabolic health and obesity phenotype and incident pancreatic cancer, no previous studies on this topic were identified. Therefore, using the longitudinal National Health Insurance Service-National Health Screening Cohort (NHIS-HEALS) database in Korea, we investigated the effects of metabolic health and obesity phenotypes on the risk of incident pancreatic cancer in the general population.

In South Korea, the NHIS was established as a government-managed, mandatory, public health insurance program in 2000, covering approximately 98% of the entire Korean population ages ≥40 years (∼50 million Koreans). Under this program, the entire Korean adult population ages ≥40 years was requested to undergo a health examination at least once every 2 years (14). The NHIS database contains eligibility-related data (e.g., age, sex, and income level), health examination data (e.g., anthropometric measurements, including height, weight, and waist circumference; blood pressure (BP) values; laboratory results; self-reported standard questionnaires on lifestyle and behavior), and inpatient and/or outpatient records [e.g., diagnoses based on the International Classification of Diseases, Tenth Revision (ICD-10) codes, medical treatment, and death]. National health examinations, including anthropometric and laboratory tests, were performed after overnight fasting. The Korean Association of Laboratory Quality Control administered the quality of the procedures. This study used the NHIS-HEALS cohort database, which made publicly available to facilitate the use of health screening data and which was chosen at random to include nearly 10% of the total population ages between 40 and 79 years in the NHIS database. Our sample initially included 362,285 individuals who underwent the health examination in 2009–2010. We excluded participants with missing data for at least one variable including laboratory results, BP values, and anthropometric measurements (n = 12,926) and participants with a pancreatic cancer diagnosis prior to 2009 (n = 1,925). In total, 347,434 individuals were included in the analysis (Fig. 1). Among them, 182,009 (52.4%), 45,093 (13.0%), 65,983 (19.0%), and 54,349 (15.6%) participants were categorized into the metabolically healthy normal weight (MHNW), MUNW, MHO, and MUO groups, respectively. Our protocol was approved by the NHIS review committee. The Hallym University (Seoul, Republic of South Korea) institutional review board approved the study protocol in accordance with the Declaration of Helsinki of the World Medical Association. The need for informed consent was waived because anonymous and deidentified information was used for the analysis.

We examined newly incident pancreatic cancer as outcomes from January 1, 2011 to December 31, 2015. Pancreatic cancer diagnosis was defined on the basis of the use of the ICD-10 code C25.0-C25.3, C25.7-C25.9 during hospital admission.

Obesity was defined on the basis of the Asia-Pacific BMI criteria (15). BMI was calculated as weight divided by the square of height (kg/m²). Participants were categorized into two groups: normal weight (BMI < 25 kg/m²) and obesity (BMI ≥ 25 kg/m²).

Metabolically unhealthy status was diagnosed using the National Cholesterol Education Program-Adult Treatment Panel III criteria (16) based on the presence of ≥3 of the following factors: (i) fasting glucose levels ≥5.6 mmol/L (100 mg/dL) or the current use of glucose-lowering agents under the ICD-10 codes E10–E14; (ii) BP ≥130/85 mm Hg or the use of antihypertensive agents under the ICD-10 codes I10–15; (iii) serum triglyceride levels ≥1.7 mmol/L (≥150 mg/dL) or the current use of lipid-lowering agents under the ICD-10 code E78; (iv) high-density lipoprotein (HDL)-cholesterol levels <1.0 mmol/L (40 mg/dL) in men or <1.3 mmol/L (50 mg/dL) in women or the current use of lipid-lowering agents under the ICD-10 code E78; and (v) waist circumference (WC) >90 cm for men or ≥85 cm for women, based on the International Diabetes Federation criteria for the Asian population (17). Alternatively, metabolically healthy status was defined as the presence of ≤2 of these factors. On the basis of BMI and metabolic health status, we divided our study population into four groups: MHNW, MUNW, MHO, and MUO.

Data on alcohol consumption and smoking status were obtained from a questionnaire completed during the health examination. Smoking status was categorized as follows: never smoker, former smoker, and current smoker. Heavy smokers were defined as participants who smoked ≥20 pack-years; heavy alcohol consumers were defined as participants who consumed ≥30 g of alcohol per day. Regular exercise was defined as strenuous physical activity for at least 20 minutes ≥5 times per week. Income levels were dichotomized at the lower 20%. History of pancreatitis (acute and chronic) was defined on the basis of the ICD-10 codes K85–K86. Type 2 diabetes was defined as ICD-10 code E11 with use of glucose-lowering agents. Data on previous cholecystectomy and gastrectomy were extracted in the health insurance payroll codes.

Baseline characteristics are presented as means ± SDs for continuous variables and numbers (percentages) for categorical variables. Differences in the distribution of baseline characteristics according to metabolic health and obesity status were identified using ANOVA or Pearson χ² test, as appropriate.

The incidence rates of outcomes were calculated by dividing the number of incident cases by the total follow-up period (person-years). HRs and 95% confidence intervals (CI) for incident pancreatic cancer were analyzed using Cox proportional hazards models for the four groups. We tested the proportional hazards assumption tests based on scaled Schoenfeld residuals and found no evidence of violating the proportional hazards assumption. Furthermore, to determine whether the association between metabolic health and obesity phenotypes and incident pancreatic cancer was independent of other potential confounders, we performed analyses using stepwise models adjusting for the following: (i) age and sex; (ii) age, sex, smoking status, alcohol intake, physical activity, income level, and hemoglobin, creatinine, alanine aminotransferase (ALT), and total cholesterol levels. We also performed sensitivity analyses by excluding participants with a history of pancreatitis, type 2 diabetes mellitus, those who underwent cholecystectomy and/or gastrectomy, those who developed pancreatic cancer within 3 years of examination, current smoker, and heavy alcohol consumers and/or heavy smokers. Kaplan–Meier curves for the probability of incident pancreatic cancer were obtained for the four phenotypes according to age.

To evaluate the risk of incident pancreatic cancer according to the number of metabolically unhealthy components (or three groups of BMI), we used Cox proportional hazards models adjusted for multiple confounders with and without BMI (or metabolically unhealthy status). Likelihood ratio tests were used to examine linear trend for HR by comparing −2 log likelihood χ² between nested models with and without the number of metabolically unhealthy components or three groups of BMI, respectively. These variables were calculated by treating category scales as continuous values.

We also examined the association between metabolic health and obesity phenotypes and pancreatic cancer in separate subgroups for age, sex, smoking status, alcohol intake, and physical activity. All statistical analyses were performed using SAS 9.4 (SAS Institute Inc.). A P < 0.05 was assumed to indicate statistical significance. All statistical analyses were performed by an experienced professional statistician (to J.S. Lee).

Table 1 presents the different anthropometric, laboratory, and behavioral characteristics based on metabolic obesity phenotypes. In each obesity-status category, metabolically unhealthy participants were older and had a higher BMI, were more likely to be male with a low income level, and had less healthy behaviors, including smoking, heavy alcohol consumption, and lack of exercise, than metabolically healthy participants. Moreover, the metabolically unhealthy groups showed increased levels of systolic BP, diastolic BP, fasting plasma glucose, triglycerides, aspartate aminotransferase, and ALT and decreased levels of HDL cholesterol.

During a median follow-up of 6.1 (interquartile range: 5.5–6.5) years, there were 886 cases of pancreatic cancer. Table 2 presents the associations between the metabolic health and obesity phenotypes and risk of pancreatic cancer. The pancreatic cancer incidence values in the MHNW, MUNW, MHO, and MUO groups were 0.36, 0.72, 0.34, and 0.53 per 1,000 persons-years, respectively. The risk of pancreatic cancer was higher in the MUNW and MUO groups than in the MHNW group in all models. After adjusting for age, sex, smoking status, alcohol intake, physical activity, income level, and levels of hemoglobin, creatinine, ALT, and total cholesterol, the adjusted HRs of incident pancreatic cancer for the MUNW and MUO phenotypes were 1.52 (95% CI, 1.27–1.81) and 1.34 (95% CI, 1.12–1.61), respectively (comparison with the MHNW phenotype). However, compared with the MHNW group, the MHO group showed no significant differences in incident pancreatic cancer (adjusted HR, 1.07; 95% CI, 0.88–1.31) before and after adjustment.

Figure 2 shows the Kaplan–Meier curves of the probability of incident pancreatic cancer during 6.5 years for the metabolic health and obesity phenotypes according to age. Over the follow-up period, the rates of incident pancreatic cancer were significantly higher in the MUNW and MUO groups than in the MHNW group within both the younger (<65 years) and older (≥65 years) groups. Particularly, in the older group, the MUNW phenotype showed significantly increased risks of pancreatic cancer.

In the subgroup analyses, most subgroups showed similar relationships after adjusting for multiple variables (Fig. 3). Particularly, in the age ≥65 years, women, never smoker, and no regular exercise subgroups, the HRs of pancreatic cancer were greater for the MUNW phenotype than for the MHNW phenotype. However, in the former smoker, heavy alcohol consumption, and regular exercise subgroups, the risks of incident pancreatic cancer were attenuated.

In the sensitivity analyses, the relationships between metabolically unhealthy phenotypes and pancreatic cancer persisted after excluding participants previously diagnosed with acute or chronic pancreatitis (Supplementary Table S1), type 2 diabetes mellitus (Supplementary Table S2), current smoker (Supplementary Table S3), heavy alcohol consumers and/or heavy smokers (Supplementary Table S4), or those who underwent cholecystectomy and/or gastrectomy (Supplementary Table S5). Furthermore, to alleviate the possibility of reverse causation, we performed analyses after excluding patients who developed pancreatic cancer within 3 years of follow-up and obtained similar results (Supplementary Table S6).

Table 3 presents the associations between the number of metabolically unhealthy components individuals had and pancreatic cancer. The HR for pancreatic cancer compared to people without any metabolically unhealthy components gradually increased with the number of components (P_trend < 0.001; Table 3). These associations persisted even after adjusting for multiple confounders including BMI. Individuals with four metabolically unhealthy components were at 86% higher risk of pancreatic cancer, and those with all five components were at 78% higher risk, compared with those without any components (Model 2). However, there was no significant associations between BMI category and pancreatic cancer after adjusting for multivariables with/without metabolically unhealthy status (Table 3).

This nationwide population-based cohort study reported an association between metabolically unhealthy phenotypes, independent of obesity, and pancreatic cancer. However, metabolically healthy individuals with obesity did not show increased risks of pancreatic cancer. Moreover, people with a higher number of metabolically unhealthy components were at higher risk of incident pancreatic cancer even after adjusting for multiple variables with and without BMI.

There is widely supported hypothesis that obesity is linked with an increased pancreatic cancer risk. According to the World Cancer Research Fund, 19 of 23 studies on pancreatic cancer incidence and 4 of 7 studies on pancreatic cancer mortality identified increased risks in high BMI groups (18). In a dose–response meta-analysis of these studies, the risks of both pancreatic cancer incidence and mortality increased by 10% for every 5-unit increase in BMI [relative risk (RR), 1.10; 95% CI, 1.07–1.14 and RR, 1.10; 95% CI, 1.02–1.19, respectively; ref. 18]. A prospective study including Swedish adults reported that an increased pancreatic cancer risk was associated with both high BMI and WC. In addition, Russo and colleagues demonstrated a relation between metabolic syndrome and pancreatic cancer risk (19). In the Metabolic Syndrome and Cancer Project involving 580,000 adults, BP and fasting glucose levels were significantly associated with the incident pancreatic cancer risk (20). A dose–response meta-analysis demonstrated that pancreatic cancer incidence increases by 14% with every 0.56 mmol/L (10 mg/mL) increase in fasting blood glucose levels (21). A meta-analysis demonstrated that metabolic syndrome is associated with pancreatic cancer risk, and hyperglycemia is the key component (22). In this study, we demonstrated the lack of association between metabolically healthy status with obesity and pancreatic cancer incidence. This suggests that obesity itself may not have a significant impact on pancreatic cancer risk if it is not accompanied by a metabolically unhealthy phenotype. Furthermore, the MUNW group showed greater HRs for pancreatic cancer than the MUO group, especially in the older population. These results are consistent with those of previous studies on the “obesity paradox,” which have shown that a high BMI is associated with low mortality (23). We previously reported that elderly individuals with MUNW phenotype showed higher all-cause mortality compared to those with MHO phenotype during a median follow-up of 10.3 years (24). BMI may be limited with respect to the representation of body composition. Atkins and colleagues reported that sarcopenic obesity, combination of obesity, and loss of muscle mass and strength, had the highest risk of all-cause mortality in a population-based cohort study (25).

In this study, higher risks of pancreatic cancer in the metabolically unhealthy phenotypes were observed in the older (≥65 years) people and women than in other subgroups. For the elderly and women, the risks of pancreatic cancer was 57% and 68%, respectively, greater in the MUNW group than in the MHNW group. Older people and women may be more affected by a metabolic disorder with regards to pancreatic cancer development and prognosis. Age- and sex-based differences may be linked due to the interactions of and alterations in various hormones. Insulin resistance-related cancers have been suggested to be directly and indirectly affected by sex hormones (26). Furthermore, a study on the Japanese population reported that pancreatic cancer incidence increased in relation to metabolic factors in women but not men (27).

Some potential mechanisms are related to the association between reduced metabolic health or insulin resistance and pancreatic cancer incidence and prognosis. One such mechanism is chronic inflammation, with changes in inflammatory cytokine concentrations and infiltration of immunosuppressive cells in the pancreas (28, 29). Ina ddition, insulin resistance related to reduced metabolic health, beyond general obesity, could increase insulin and insulin-like growth factor I levels, known to inhibit apoptosis and promote the growth of cancer cells (28, 29). Future studies are needed to clarify the underlying pathophysiological mechanisms of this association.

Our study has some limitations. First, although we tried to eliminate confounding factors using multivariate or sensitivity analyses, we cannot exclude the possibility of residual confounding variables or bias. It was impossible to identify causality and to remove reverse causation completely, because this study was designed retrospectively. In addition, family history and specific genetic mutations are regarded as powerful risk factors for pancreatic cancer. However, we could not exclude the influence of these risk factors due to the use of unidentified personal information and the characteristics of the database. Second, we defined as pancreatic cancer diagnosis based on medical claim codes (ICD-10 code C25.0-C25.3, C25.7-C25.9) during hospital admission. Therefore, the possibility of coding errors could not be controlled. However, the accuracy of pancreatic cancer diagnosis using the same algorithm has been validated by comparing NHIS claims data and data from the National Cancer Registry of Korea (30). Furthermore, Hwang and colleagues reported that ICD-10 codes of pancreatic cancer in the NHIS database are a reliable method of diagnosis in population-based cohort studies (31). Several previous studies have used this method for defining a pancreatic cancer diagnosis (32–35). Third, this study only included subjects from Korea; therefore, the results need to be confirmed in other ethnic groups.

Nevertheless, this study has some strengths. This study was based on a longitudinal, large, standardized database managed by the Korean government, with credible information containing measurement for anthropometric factors and fasting blood samples for all participants. Previous studies on pancreatic cancer involving anthropometric factors such as BMI and WC have mostly depended on self-reported records, which could be biased, as weight tends to be underestimated by people with obesity (36). In addition, because pancreatic cancer is associated with poor survival and a relatively low incidence, the use of large administrative databases can provide statistical power. Moreover, because this study included a sample taken from the general population, the findings could be representative of the general Korean population. Finally, the main findings were consistent with those of sensitivity analyses, excluding individuals with a history of pancreatitis, type 2 diabetes, heavy drinking and/or smoking, or who had undergone cholecystectomy and/or gastrectomy. Furthermore, a sensitivity analysis was performed excluding those who developed pancreatic cancer in the first 3 years of follow-up to attenuate reverse causality, and similar results were observed, further supporting our results.

In conclusion, our study provides evidence for an association between metabolically unhealthy phenotypes and the risk of pancreatic cancer independent of obesity. Compared with the MHNW phenotype, the MUNW and MUO phenotypes had elevated risks of pancreatic cancer but the MHO phenotype did not. As a modifiable risk factor of pancreatic cancer, a metabolically unhealthy phenotype may be controlled using a multidisciplinary approach involving diets, physical activity, lifestyle modification, and pharmacologic management. Further studies with larger populations and other ethnic populations are needed to confirm this association because pancreatic cancer has a relatively low incidence rate with geographical variation.

No disclosures were reported.

The authors assume full responsibility for analyses and interpretation of these data.

H.S. Chung: Conceptualization, resources, data curation, formal analysis, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. J.S. Lee: Conceptualization, resources, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. E. Song: Software, formal analysis. J.A. Kim: Data curation, software, formal analysis. E. Roh: Data curation, software, formal analysis. J.H. Yu: Data curation, software, formal analysis. N.H. Kim: Software, formal analysis. H.J. Yoo: Data curation, formal analysis. J.A. Seo: Supervision. S.G. Kim: Supervision. N.H. Kim: Supervision. S.H. Baik: Supervision. K.M. Choi: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

This study used data (NHIS-2020-2-054) from by National Health Insurance Service (NHIS). The authors thank the National Health Insurance Sharing Service.

This study was supported in part by the Korea University Research Fund (K2020461, to K.M. Choi) and the National Research Foundation of Korea (NRF-2018R1D1A1B07049605, to H.S. Chung).

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.