Lifestyle, body mass index, diabetes, and the risk of pancreatic cancer in a nationwide population-based cohort study with 7.4 million Korean subjects

Abstract

Background

Large-scale epidemiological studies on pancreatic cancer in non-Western populations are insufficient. We investigated the risk factors for pancreatic cancer.

Methods

Using the Korean National Health Insurance database, subjects who participated in the health examination program between 2005 and 2006 were identified and followed up until 2017. Adjusted hazard ratios (HRs) for pancreatic cancer risk were estimated using a Cox proportional hazards model.

Results

During 11.5 years follow-up, 22,543 of 7,445,947 participants were newly diagnosed with pancreatic cancer. Compared with normal-weight subjects, pancreatic cancer risk was increased in those with severe obesity (BMI ≥ 28 kg/m²) (HR = 1.16; 95% CI, 1.11–1.23). Subjects with diabetes had an increased risk compared with those without diabetes (HR = 1.48; 95% CI, 1.43–1.53). Current smokers had a higher risk than never smokers (HR = 1.43; 95% CI, 1.38–1.48). Current smoking combined with diabetes increased the risk compared with never smokers without diabetes (HR = 2.13; 95% CI, 2.00–2.28). Current smoking combined with BMI ≥ 25 kg/m² had an increased risk compared with never smokers with BMI < 23 kg/m² (HR = 1.55; 95% CI, 1.46–1.65).

Conclusion

Smoking, obesity, and diabetes are significant risk factors for pancreatic cancer in Koreans. Lifestyle modifications for smoking and obesity would be beneficial for pancreatic cancer prevention.

Background

Pancreatic cancer is the fourth leading cause of cancer-related death in developed countries [1]. Since pancreatic cancer does not cause symptoms at an early stage and early detection methods have not yet been established, this cancer is usually diagnosed at an advanced stage and has a 5-year survival rate of only 9% [2]. Therefore, identifying modifiable risk factors is critical for developing adequate preventive strategies.

The etiology of pancreatic cancer remains unclear. Smoking, obesity, heavy alcohol consumption, type 2 diabetes, and dietary factors have been suggested as modifiable factors, whereas age, sex, genetic risk factors, and chronic pancreatitis are known non-modifiable risk factors [3, 4]. Smoking is the most prominent modifiable risk factor, accounting for approximately 20% of pancreatic cancer cases [5]. According to a meta-analysis, the relative risk of smoking for pancreatic cancer was 1.6–2.2 for current smokers and 1.1–1.2 for former smokers compared with never smokers [6]. Alcohol consumption has been recognized as a possible risk factor for pancreatic cancer; however, the association is still unclear [7]. In two meta-analyses, only heavy drinking, not moderate drinking, was associated with an increased risk for pancreatic cancer [7, 8]. Long-standing diabetes is a known risk factor associated with pancreatic cancer, and diabetes may also be a clinical feature of pancreatic cancer [9]. The risk of pancreatic cancer is the highest in recent-onset diabetes [10]. Recent studies have suggested that obesity is a risk factor for pancreatic cancer [6, 11–13]. Overweight and obesity are known to increase pancreatic cancer risk by 10% and 20%, respectively [14]. However, some reports in Asian populations indicated no association between obesity and pancreatic cancer [15, 16].

Most large-scale studies of risk factors for pancreatic cancer were conducted in Western populations, and studies on Asian populations, including Koreans, with different genetic and social characteristics, are insufficient. Moreover, considering the low incidence of pancreatic cancer, epidemiological research requires a very large population; however, most studies included relatively small populations. Even in meta-analyses, studies involving more than 5 million people are extremely rare. Due to the complex mechanisms of pancreatic cancer, possible interactions between risk factors may exist. However, few studies have investigated the combined effects of risk factors.

We conducted this nationwide population-based retrospective cohort study with 7.4 million people from the Korean National Health Insurance database to investigate the incidence of pancreatic cancer according to differences in lifestyle, diabetes, and body mass index (BMI). Furthermore, the combined effects of risk factors for pancreatic cancer were evaluated. We aimed to identify modifiable risk factors for the primary prevention of pancreatic cancer, and to contribute to its early diagnosis by identifying high-risk groups.

Methods

Data source

This study was based on data from the National Health Insurance Service (NHIS) database between 2002 and 2017. The NHIS is a mandatory nationwide health insurance system that covers more than 98% of the Korean population. Medical information on almost all patients in healthcare institutions is prospectively integrated into the NHIS claims database which includes extensive information on diagnoses and comorbidity codes classified according to the tenth revision of the International Classification of Diseases (ICD-10), demographic characteristics, admission and ambulatory care, medications, and procedure codes [17].

Further, we used the Korean National Health Screening (NHS) program database to obtain health and behavior information. Since 1995, the NHIS has provided the NHS program to improve the health status of Koreans through the prevention and early detection of diseases. Subjects eligible for the screening program are employment-based policyholders and household heads aged ≥19 years and regional policyholders and dependents aged ≥40 years. Health examinations in the NHS program are conducted at least every 2 years. The database of these health examinations include a health behavior questionnaire (smoking, alcohol intake, and exercise), previous medical history questionnaire, and information on physical examinations such as height and weight, laboratory studies, and chest radiography [18].

Study population

The population was selected NHS participants from 2005 to 2006. The participation rate in the NHS program among the eligible population was 51.6% in 2005 and 55.7% in 2006 [19]. Of these, subjects younger than 40 years, with a cancer diagnosis from 2002 to the year of health examination, with missing variables, and who died or developed pancreatic cancer within 1 year after health examination were excluded.

Data collection

Demographic variables were assessed. Age was classified in 10-year intervals. Health-related lifestyle data such as smoking status, alcohol intake, and physical activity were collected using standardized self-reporting questionnaires [18]. Smoking status was categorized as never smoker, former smoker, and current smoker. The current daily smoking dose was classified into <10, 10–19, and ≥20 cigarettes. Since alcohol intake was surveyed by frequency per week, not by amount, it was categorized as follows: not drinking, occasional drinking (less than weekly), and drinking 1–2, 3–4, and ≥5 times per week. Physical activity was categorized as no exercise, and exercising 1–2, 3–4, and ≥5 times per week. Exercise was defined as strenuous physical activity that was performed for at least 30 min. Bioclinical laboratory results, such as blood pressure, fasting glucose, lipid profile, hemoglobin, urine stick test, creatinine, liver enzymes, weight and height, and waist circumference, were included in the health screening database. Body weight and height were measured by trained personnel. BMI was calculated as weight divided by height squared (kg/m²) and categorized as underweight (<18.5 kg/m²), normal-weight (18.5–22.9 kg/m²), overweight (23.0–24.9 kg/m²), obese I (25.0–27.9 kg/m²), and obese II (≥28.0 kg/m²) by modifying the World Health Organization (WHO) criteria for Asian populations [20]. Using a questionnaire, previous medical histories of hypertension, diabetes, malignant diseases, heart diseases, chronic liver diseases, and stroke were included. The income level was stratified into quintiles.

Definition of disease condition

Diabetes was defined as the presence of at least two claims per year with an ICD-10 code for diabetes (E10–14) during the year preceding the health examination, the identification of diabetes in a previous medical history questionnaire or blood fasting glucose level ≥ 126 mg/dL at a health examination. Dyslipidemia was defined as the presence of at least two claims with an ICD-10 code for dyslipidemia (E78) during the year preceding the health examination or a blood total cholesterol of ≥240 mg/dL at a health examination.

Follow-up and identification of newly developed pancreatic cancer

All subjects were followed-up from the date of health examination until December 31, 2017, the date of pancreatic cancer diagnosis, or death, whichever came first. Newly developed pancreatic cancer during follow-up was defined as ICD-10 code C25 (pancreatic cancer) combined with the specialized claim code V193. Since 2005, the NHIS has been reimbursing cancer patients who are identified using the specialized claim code V193. The date of death was extracted from the qualification NHIS dataset.

Statistical analysis

Continuous variables are presented as means ± standard deviations and categorical variables as numbers of subjects and percentages. Statistical analyses were conducted using the χ² test and independent t-test to investigate the differences in variable factors. The crude incidence rate (per 100,000 person–years) was calculated in relation to person-years and number of pancreatic cancer cases. Hazard ratios (HRs) were calculated using the Cox proportional hazards model after adjusting for age, sex, income, smoking status, alcohol intake, physical activity, diabetes, dyslipidemia, and BMI. BMI was analyzed not only as a categorical variable but also as a continuous variable in 5 kg/m² increments as a risk factor for developing pancreatic cancer. Restricted cubic splines were used to detect the dose–response relationship between BMI, smoking, and pancreatic cancer risk. Population attributable fraction (PAF) for smoking (current smoker) and obesity (defined as BMI ≥ 25 kg/m²) were estimated using the standard formula [21] by age and sex category. To demonstrate the combined effects of BMI and smoking, BMI was categorized into three groups: <23 kg/m², 23–24.9 kg/m², and ≥25 kg/m². Statistical multiplicative interactions between smoking and diabetes and between smoking and BMI in relation to pancreatic cancer risk were tested using the likelihood ratio test. The cumulative HRs for different risk combinations of smoking and diabetes and those of smoking and BMI were estimated. For all risk analyses, the Cox model proportionality assumption was validated as a visual assessment of Kaplan-Meier curves. Non-intersecting lines in the log (–log) plots of Kaplan-Meier survival against the log of time indicated that the assumption is reasonable. All statistical tests were two-sided, and statistical significance was determined as p < 0.05. SAS statistical software version 9.4 (SAS Institute Inc., Cary, NC) was used for all analyses.

Results

Study population

In total, 12,248,104 participants who had undergone the annual or biennial NHS program between 2005 and 2006 were identified from the NHIS data. Among these, subjects younger than 40 years (n = 4,173,292), with a previous history of any cancer (n = 92,142), or with missing data for independent variables (n = 509,780) were excluded. Furthermore, those who died or were diagnosed with pancreatic cancer within 1 year after the health examination were excluded (n = 26,946). Finally, 7,445,947 subjects were included and followed-up until December 31, 2017 (Fig. 1).

Fig. 1. Flow chart of the study population.

NHIS National Health Insurance Service.

The mean age at the time of health examination was 53.9 ± 10.4 years and 50.6% of the subjects were men (Table 1). The mean follow-up duration was 11.50 ± 1.68 years, and the person-years of follow-up were 85,640,205 person-years. During follow-up, 22,543 patients (0.3% of the total study population) were newly diagnosed with pancreatic cancer. The crude incidence rate per 100,000 person-years was 26.3.

Table 1.

Baseline characteristics of the subjects.

	All subjects		Not developing pancreatic cancer		Developing pancreatic cancer		p value
	No.	%	No.	%	No,	%
Total	7,445,944	100.0	7,423,401	99.7	22,543	0.3
Sex
Male	3,768,191	50.6	3,755,254	99.7	12,937	0.3	<0.0001
Female	3,677,753	49.4	3,668,147	99.7	9606	0.3
Mean age (years)	53.9 ± 10.4		53.9 ± 10.4		61.9 ± 9.8		<0.0001
Ages (years)
40–49	3,043,184	40.9	3,040,339	99.9	2845	0.1	<0.0001
50–59	2,189,149	29.4	2,183,464	99.7	5685	0.3
60–69	1,487,783	20.0	1,479,273	99.4	8510	0.6
70–79	636,670	17.3	631,751	99.2	4919	0.8
≥80	89,158	1.2	88,574	99.3	584	0.7
Income level
1 quintile	1,210,474	16.3	1,206,804	99.7	3670	0.3	0.050
2 quintile	918,720	12.3	915,800	99.7	2920	0.3
3 quintile	1,269,700	17.1	1,265,914	99.7	3786	0.3
4 quintile	1,618,462	21.7	1,613,540	99.7	4922	0.3
5 quintile	2,428,588	32.6	2,421,343	99.7	7245	0.3
BMI (kg/m2)
<18.5	166,931	2.2	166,385	99.7	546	0.3	<0.0001
18.5–22.9	2,677,510	36.0	2,670,086	99.7	7,424	0.3
23–24.9	2,020,043	27.1	2,013,997	99.7	6046	0.3
25–27.9	1,928,506	25.9	1,922,285	99.7	6221	0.3
≥28	652,954	8.8	650,648	99.6	2306	0.4
Hyperlipidemia
No	6,283,180	84.4	6,264,644	99.7	18,536	0.3	<0.0001
Yes	1,162,764	15.6	1,158,757	99.7	4007	0.3
Diabetes
No	6,644,308	89.2	6,626,199	99.7	18,109	0.3	<0.0001
Yes	801,636	10.8	797,202	99.4	4434	0.6
Alcohol intake
No	4,472,872	60.1	4,458,866	99.7	14,006	0.3	<0.0001
2–3 times/month	1,047,600	14.1	1,045,122	99.8	2478	0.2
1–2 times/week	1,171,713	15.7	1,168,586	99.7	3127	0.3
3–4 times/week	474,323	6.4	472,770	99.7	1553	0.3
≥5 times/week	279,436	3.8	278,057	99.5	1379	0.5
Smoking
Never smoker	5,292,911	71.1	5,277,790	99.7	15,121	0.3	<0.0001
Former smoker	693,470	9.3	691,288	99.7	2182	0.3
Current smoker	1,459,563	19.6	1,454,323	99.6	5240	0.4
Cigarettes/day in current smoker
<10	348,278	4.7	346,809	99.6	1469	0.4
10–19	773,652	10.4	771,001	99.7	2651	0.3
≥20	337,633	4.5	336,513	99.7	1120	0.3
Physical activity
No	3,950,515	53.1	3,937,937	99.7	12,578	0.3	<0.0001
1–2 times/week	1,872,202	25.1	1,867,450	99.7	4752	0.3
3–4 times/week	858,395	11.5	856,176	99.7	2219	0.3
≥5 times/week	764,832	10.3	761,838	99.6	2994	0.4

BMI, body mass index.

Risk factors of pancreatic cancer

In all participants, after adjusting for all variables at baseline, the HR increased as the BMI increased. In categorical analysis, the HR was 0.99 (95% confidence interval [CI], 091–1.08) in the BMI < 18.5 kg/m² group, 1.02 (95% CI, 0.99–1.06) in the BMI 23.0–24.9 kg/m² group, 1.06 (95% CI, 1.03–1.10) in the BMI 25–27.9 kg/m² group, and 1.16 (95% CI, 1.11–1.22) in the BMI ≥ 28 kg/m² group. There was a significant dose-response relationship between BMI and pancreatic cancer risk (P_trend < 0.0001) (Fig. 2A). In analysis examining BMI as a continuous variable, each 5 kg/m² increment increase in BMI was significantly associated with a 6% higher risk of pancreatic cancer (4% in men and 9% in women). Diabetes was a significant independent risk factor for pancreatic cancer in the multivariate analysis (HR = 1.48; 95% CI, 1.43–1.53; p < 0.0001) (Fig. 2B). There was no association with increased pancreatic cancer risk for hyperlipidemia (Table 2). Former smokers had a slightly but statistically significantly increased risk (HR = 1.07; 95% CI, 1.02–1.12; p = 0.0079) and current smokers had a significantly greater risk (HR = 1.43; 95% CI, 1.38–1.48; p < 0.0001) than never smokers. In current smokers who smoked <10 cigarettes daily, 10–19 cigarettes daily, and ≥20 cigarettes daily, the risk was increased by 32.7%, 44.9%, and 54.9%, respectively. As the daily smoking amount increased, the HRs proportionally increased, with a significant dose-response relationship (P_trend < 0.0001) (Fig. 2C). Only subjects who consumed alcohol more than five times per week had a statistically higher HR (HR = 1.08; 95% CI, 1.02–1.14; p = 0.0126). Pancreatic cancer risk did not change depending on physical activity.

Fig. 2. Adjusted HRs for pancreatic cancer by BMI, smoking status, and diabetes.

A HRs by BMI, B HRs by diabetes, and C HRs by smoking status. The squares present HRs, and the vertical line presents 95% CI.

Table 2.

Hazard ratios of risk for developing pancreatic cancer.

	Men			Women			All participants
	No. of cases	HR(95% CI)	p value	No. of cases	HR(95% CI)	p value	HR(95% CI)	p value
BMI (kg/m2)
Continuous model per 5 kg/m2 increase		1.04(1.02–1.07)	0.0017		1.09(1.05–1.12)	<0.0001	1.06(1.04–1.08)	<0.0001
Categorical model
<18.5	337	1.01(0.90–1.12)	0.9334	209	0.97(0.84–1.11)	0.6355	0.99(0.91–1.08)	0.8532
18.5–22.9	4418	1.00(ref.)		3006	1.00(ref.)		1.00(ref.)
23–24.9	3559	0.99(0.95–1.04)	0.7704	2487	1.07(1.01–1.13)	0.0169	1.02(0.99–1.06)	0.2167
25–27.9	3545	1.04(0.99–1.08)	0.1347	2676	1.10(1.05–1.16)	0.0003	1.06(1.03–1.10)	0.0006
≥28	1078	1.16(1.08–1.24)	<0.0001	1228	1.18(1.10–1.26)	<0.0001	1.16(1.11–1.22)	<0.0001
		Ptrend = 0.0029			Ptrend < 0.0001		Ptrend < 0.0001
Diabetes
No	10,239	1.00(ref.)		7,870	1.00(ref.)		1.00(ref.)
Yes	2698	1.47(1.41–1.54)	<0.0001	1,736	1.49(1.41 –1.57)	<0.0001	1.48(1.43 –1.53)	<0.0001
Hyperlipidemia
No	11,109	1.00(ref.)		7,427	1.00(ref.)		1.00(ref.)
Yes	1828	1.04(0.99–1.09)	0.1492	2,179	1.03(0.98–1.08)	0.2764	1.03(1.00–1.07)	0.078
Smoking
Never smoker	5967	1.00(ref.)		9,154	1.00(ref.)		1.00(ref.)
Former smoker	2087	1.07(1.02–1.13)	0.0099	95	1.12(0.91–1.37)	0.2877	1.07(1.02–1.12)	0.0079
Current smoker	4883	1.42(1.36–1.48)	<0.0001	357	1.52(1.36–1.69)	<0.0001	1.43(1.38–1.48)	<0.0001
Cigarettes/day in current smoker
<10	1239	1.30(1.22–1.38)	<0.0001	230	1.48(1.30–1.69)	<0.0001	1.33(1.26–1.40)	<0.0001
10–19	2546	1.44(1.37–1.51)	<0.0001	105	1.58(1.31–1.92)	<0.0001	1.45(1.38–1.52)	<0.0001
≥20	1098	1.54(1.44–1.65)	<0.0001	22	1.63(1.07–2.48)	<0.0001	1.55(1.45–1.65)	<0.0001
		Ptrend < 0.0001			Ptrend < 0.0001		Ptrend < 0.0001
Alcohol intake
No	5659	1.00(ref.)		8347	1.00(ref.)		1.00(ref.)
2-3 times/month	1753	0.90(0.85–0.95)	0.0001	725	1.07(0.99–1.15)	0.1027	0.95(0.91–1.00)	0.0333
1-2 times/week	2760	0.95(0.90–0.99)	0.0248	367	1.07(0.96–1.19)	0.2392	0.98(0.93–1.11)	0.2509
3-4 times/week	1457	0.97(0.92–1.03)	0.3290	96	1.26(1.03–1.54)	0.0267	1.00(0.95–1.06)	0.8834
≥5 times/week	1308	1.05(0.99–1.12)	0.1201	71	1.15(0.91–1.46)	0.2403	1.08(1.02–1.14)	0.0126
Physical activity
None	6547	1.00(ref.)		6031	1.00(ref.)		1.00(ref.)
1–2 days/week	3170	0.96(0.92–1.00)	0.0705	1582	1.03(0.97–1.09)	0.3001	0.98(0.95–1.02)	0.3568
3–4 days/week	1396	0.95(0.89–1.00)	0.0617	823	1.10(0.94–1.09)	0.7028	0.97(0.93–1.02)	0.1942
≥5 days/week	1824	1.03(0.98–1.09)	0.2092	1170	1.04(0.98–1.11)	0.1793	1.04(1.00–1.08)	0.0670

Adjusted for age, incomes, smoking, alcohol intake, physical activity, diabetes, dyslipidemia, and BMI.

BMI body mass index, HR hazard ratio, CI confidence interval.

According to sex, the HRs of the risk of developing pancreatic cancer in BMI ≥ 28 kg/m² and diabetes were significantly increased in both sexes, with no overt difference between men and women (1.16 vs. 1.18, 1.47 vs. 1.49, respectively). Moreover, the HRs in former and current smokers were significantly increased in both men and women (1.07 vs. 1.12, 1.42 vs. 1.52, respectively) (Table 2). There were no significant differences in the HRs according to alcohol intake and physical activity in men and women.

The PAFs for smoking and obesity were calculated. Smoking had a PAF of 9.6% in men, 1.2% in women, and 6.0% in all participants. The PAFs for obesity were 2.9%, 4.3%, and 3.7%, respectively.

Combined effects of risk factors

The cumulative association was observed with the combination of different risk factors. For smoking and diabetes, current smokers with diabetes had a cumulative increased risk compared with never smokers without diabetes (HR = 2.13; 95% CI, 2.00–2.28, p < 0.0001) (Fig. 3A). For smoking and BMI, the pancreatic cancer risk increased significantly in current smokers with a BMI ≥ 25 kg/m² compared with never smokers with a BMI < 23 kg/m² (HR = 1.55; 95% CI, 1.46–1.65, p < 0.0001) (Fig. 3B). HRs tended to increase as BMI increased in each smoking category. However, there was no statistically significant interaction between smoking and diabetes or BMI (Table 3).

Fig. 3. Adjusted HRs for pancreatic cancer showing the combined effects of risk factors for pancreatic cancer.

A Combining smoking status with diabetes in all subjects, B combining smoking status with BMI in all subjects. The magnitude of HRs for each combination is presented on each bar. Reference groups were never smokers without diabetes (A) and never smokers with BMI < 23 kg/m² (B), with HR = 1.00.

Table 3.

Hazard ratios of risk for developing pancreatic cancer according to BMI and smoking categories.

Factors	Never smoker			Former smoker			Current smoker
	No. of cases	HR (95% CI)	p value	No. of cases	HR (95% CI)	p value	No. of cases	HR (95% CI)	p value
Diabetes
No	12,224	1 (ref.)		1718	1.06 (1.01–1.12)	0.0258	4167	1.41 (1.36–1.47)	<0.0001
Yes	2897	1.49 (1.43–1.55)	<.0001	464	1.62 (1.47–1.78)	<0.0001	1073	2.13 (2.00–2.28)	<0.0001
p interaction					0.86
BMI (kg/m2)
<23	5045	1 (ref.)		674	1.06 (0.97–1.15)	0.1978	2251	1.47 (1.39–1.55)	<0.0001
23–24.9	4055	1.03 (0.99–1.08)	0.1321	649	1.14 (1.04–1.24)	0.003	1342	1.44 (1.35–1.53)	<0.0001
≥25	6021	1.10 (1.06–1.14)	<.0001	859	1.16 (1.08–1.25)	0.0001	1647	1.55 (1.46–1.65)	<0.0001
p interaction					0.56

Adjusted for age, sex, incomes, smoking, alcohol intake, physical activity, diabetes, dyslipidemia, and BMI.

BMI body mass index, HR hazard ratio, CI confidence interval.

Discussion

This study investigated the risk factors for pancreatic cancer using the Korean NHIS data in a nationwide retrospective large cohort with 7.4 million people. We found that smoking, high BMI, and diabetes were significant risk factors for pancreatic cancer and the magnitude of this risk. Current smokers had an increased risk (42.7%) compared with never smokers. In particular, in current smokers, the risk increased in proportion to the amount of daily smoking. Diabetes increased pancreatic cancer risk by 47.8%. Subjects with a BMI of 25–27.9 kg/m² and ≥28 kg/m² had increased risks of 6% and 16%, respectively. Moreover, we presented the combined effects of smoking and diabetes or BMI. The combination of current smokers with diabetes or high BMI was associated with a higher risk of developing pancreatic cancer compared with never smokers with no diabetes or normal BMI.

Obesity is a known potential risk factor for several malignancies including pancreatic cancer. Subjects with a BMI ≥ 30 kg/m² have an increased pancreatic cancer risk compared to those with a normal BMI (HR: 1.15–1.53) [4]. A meta-analysis of 21 studies involving 3,496,981 individuals found that when the BMI increased by 5 kg/m², the relative risk of pancreatic cancer increased by 10–16% [14]. However, most findings are from Western populations, and it is unclear whether the same trends apply to Asian populations. In our study, an increase in BMI of 5 kg/m² was associated with a 6% increase in the risk of pancreatic cancer. A pooled analysis of more than 340,000 Japanese subjects reported a positive association between obesity and pancreatic cancer risk in men only [11]. In Chinese Singaporeans, BMI and pancreatic cancer were not associated in never smokers [22]. Two pooled analyses of pancreatic cancer mortality in Asian countries found no association between BMI and pancreatic cancer risk [15, 16]. In a meta-analysis of 34 global studies, with a 5 kg/m² increase in BMI, the relative risk increased 1.10 in men and 1.08 in women, but the relative ratio was 0.94 in the Asian-Pacific group only [23]. Asian populations have a higher percentage of body fat than Western populations, the WHO recommended that the cut-off point for overweight and obesity should be lower in Asian populations [20]. In Korea, obesity is defined as a BMI ≥ 25 kg/m² [24]. In our study, 34.7% of the subjects were obese and the risk of pancreatic cancer was significantly slightly increased in subjects with a BMI of 25–27.9 kg/m² (HR = 1.06) and ≥28 kg/m² (HR = 1.16). However, this risk was distinctly lower than that reported in the West [13].

Many studies have reported an epidemiological association between diabetes and pancreatic cancer. Several meta-analyses and pooled analyses have suggested that long-standing diabetes is associated with a 1.5–2.0-fold increase in pancreatic cancer risk [6, 9]. A Taiwanese population-based study found that patients with diabetes had a 1.77-fold risk of developing pancreatic cancer [25], which is consistent with our results (HR = 1.48). Pancreatic cancer risk decreases with the duration of diabetes; however, this risk exceeds 30% relative non-diabetics for more than 20 years after the diagnosis of diabetes [26]. Approximately 60–85% of patients with newly diagnosed pancreatic cancer are estimated to have diabetes or hyperglycemia at diagnosis [27, 28]. New-onset diabetes may be an early manifestation of pancreatic cancer rather than a risk factor, and is associated with a 6–8-fold increased risk for pancreatic cancer [10]. In our study, no information about the duration of diabetes was available; therefore, we were unable to analyze the risk according to the duration of diabetes.

In our study, the HR of pancreatic cancer was 1.07 in former smokers and 1.43 in current smokers compared with never smokers. According to a few previous meta- and pooled analyses, the HRs were 1.1–1.2 in former smokers and 1.7–1.8 in current smokers [5, 29, 30]. Moreover, in a pooled analysis limited to Asians, the HRs were 1.2 in former smokers and 1.6 in current smokers [5]. The HRs in our study were slightly low compared to other studies. This is presumed to be due to the differences in the racial characteristics of Asians and the study design. As in our study, smoking was associated with a lower relative risk of pancreatic cancer in Asians than Westerners [29]. Similar differences were also found with regard to lung cancer [31], bladder cancer [32], and cholangiocarcinoma [33]. Why Asian populations are less susceptible to smoking-related cancers remains unclear. Asian countries, including Korea, have the lowest levels of tar and nicotine in cigarettes [34]. The lung cancer risks have been reported to be lower with low-tar cigarettes than with high-tar cigarettes [35]. Although there is debate about the association of low-tar and nicotine concentration in cigarettes with cancer incidence, this can be a possible reason. Another possible explanation is that Asians have lower nicotine metabolism than Westerners, so they take in less nicotine from each cigarette smoked [36]. Smoking amount in current smokers in our study was determined by the amount of daily smoking, not cumulative smoking exposure. HRs were 1.33 in <10 cigarettes/day, 1.45 in 10–19 cigarettes/day, and 1.55 in ≥20 cigarettes/day, indicating that risk increases proportionally to the daily smoking amount with a dose-response relationship. This finding is consistent with those of meta- and pooled analyses, which showed that the risk increased significantly as the daily cigarette consumption increases [5, 30]. For the dose-response association between smoking and pancreatic cancer risk, a meta-analysis based on 42 observational studies reported that HRs were 1.5, 1.9, 2.0, and 2.1 for 10, 20, 30, and 40 cigarettes/day, respectively, and this association was non-linear [37]. This inconsistency between this meta-analysis and our study may be attributed to the different cutoffs used for the daily smoking categories. Sex differences in association between pancreatic cancer risk and smoking are still controversial. In our study, the risk in former and current smokers increased in both men and women with no sex differences. A Japanese pooled analysis found that a significant increase in risk in former smokers and in ever smokers with <20 pack years was observed only in women. In addition, they observed no risk reduction from smoking cessation in women, suggesting that there may be sex differences [38]. However, other meta- and pooled analyses reported no sex differences with smoking [30, 39, 40]. Further investigation is warranted on this issue.

Regarding the association between alcohol intake and pancreatic cancer risk, several pooled analyses have shown that moderate alcohol intake had no effect on risk, and high alcohol intake was associated with increased pancreatic cancer risk [7, 8, 41]. Our results revealed a slight increase in risk only in subjects who consumed alcohol at least five times per week. Since the increase in risk is relatively low and heavy drinkers make up a small proportion of the general population, alcohol appears to play a role in only a small fraction of all pancreatic cancer cases [8].

Several epidemiologic studies have demonstrated that physical activity can reduce pancreatic cancer risk; [42–44] however, a systematic review and meta-analysis indicated that evidence for this association was lacking [45]. Likewise, in our study, physical activity and pancreatic cancer were not associated.

In our study, the PAFs for smoking and obesity were 6.0% and 3.7% for all subjects, respectively, similar to other reports for Asian populations. PAFs for smoking vary from country to country, from 6.6% in China to 28.7% in the United Kingdom [46]. In Western populations, PAFs for obesity were reported to be 12–16% [47, 48], whereas in Asian populations, it was reported to be 3–4% [47, 49].

Relatively few studies have investigated the interaction of different risk factors for pancreatic cancer. A Japanese cohort study showed that obese smokers have a higher risk of pancreatic cancer than non-obese smokers, suggesting an additive effect of BMI and smoking [50]. In two Swedish cohort studies, high BMI and smoking had a cumulative association with pancreatic cancer risk [51]. The European Prospective Investigation into Cancer and Nutrition study revealed that subjects with a high healthy lifestyle index (including smoking status, alcohol intake, physical activity, obesity status, and diet) were less likely to develop pancreatic cancer, regardless of smoking status [52]. Regarding the interaction of risk factors, two case-control studies have reported synergistic interactions between risk factors for pancreatic cancer [41, 53]. However, these studies were limited to small populations. In contrast, a large cohort study and a pooled analysis reported no interactions [51, 54]. Although a test of interaction in our study was not statistically significant, current smoking was a strong risk factor for pancreatic cancer when combined with BMI or diabetes. Combined smoking and obesity further increased the pancreatic cancer risk. Current smokers with diabetes had a 2.13-fold higher risk compared with never smokers without diabetes. Smoking and obesity are primarily preventable personal traits that still significantly impact the health of the global population [55]. Smoking cessation is essential for the prevention of a myriad of diseases in all people, including pancreatic cancer, regardless of diabetes or obesity status. Moreover, strategies to effectively prevent obesity could reduce the burden of pancreatic cancer.

Patients with a diagnosis of pancreatic cancer within 1 year of their heath examination were excluded in our study. This was to clarify the causal relationship by setting a period of at least 1 year between the risk exposure and cancer development. In particular, since new-onset diabetes is an early manifestation of pancreatic cancer, this helped to reduce the error in which new-onset diabetes presents as a risk factor for pancreatic cancer.

The crude incidence of this study population was 26.3 per 100,000 person-years. The mean national crude incidence of pancreatic cancer from 2007 to 2017 in overall Korean population was 10.9 per 100,000 person-years [56]. The subjects of this study were aged above 40 years, and in 2005, those over the age of 40 years accounted for 40.3% of the total population. Since pancreatic cancer mostly occurs above the age of 40 years, assuming that the results of this study are expanded to all ages, the crude incidence was calculated to be 10.6 per 100,000 person-years. This result was almost identical to that published for the overall population.

This study has several potential limitations. Because of its retrospective population-based design, several possible biases exist, including coding bias, selection bias, and effects of confounding factors. First, because risk factors were analyzed at a single point in the health examination, it was not possible to reveal the effects of newly developed diseases such as diabetes and subsequent changes in BMI and lifestyle on pancreatic cancer development. Second, not all pancreatic cancers were histologically confirmed. Third, data on physical activity, smoking status, and alcohol intake were self-reported which may have skewed the risk estimates. Fourth, our data did not include specific histological types of pancreatic cancer. Although more than 90% of pancreatic cancers are ductal adenocarcinomas, heterogeneous histological types were included altogether. Fifth, diabetes and dyslipidemia are generally defined using prescription drug information combined with diagnostic codes. Since a large number of subjects were included, drug prescription information could not be extracted due to the data capacity limit. Diagnoses of diabetes and dyslipidemia were performed using diagnostic codes or laboratory findings only. Finally, the participation rate of the NHS was 51.5–55.7%; thus, there may be a difference between participants and non-participants. Therefore, there are limitations in the generalizability of the results.

Nevertheless, this study has several strengths. First, it included a very large number of participants. To the best of our knowledge, with 7,445,944 participants, 85,640,205 person-years of follow-up, and 22,543 incident pancreatic cancer cases, this population-based cohort study may be the largest to date to have investigated various risk factors. This made it possible to have significant statistical power for each risk factor and to analyze the combined effects in each subgroup with sufficient numbers. Second, the combined effects of important risk factors were demonstrated, which is unique among most other studies.

In conclusion, smoking, obesity, and diabetes were confirmed to be significant risk factors for pancreatic cancer in Koreans. Moreover, this risk increased in proportion to the daily smoking amount in current smokers. Smoking increased pancreatic cancer risk in subjects with and without diabetes. Furthermore, smoking combined with diabetes or obesity further increased the risk. Lifestyle modifications for smoking and obesity would be beneficial for pancreatic cancer prevention.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Author contributions

BKP and JHS designed the paper and performed the interpretation. BKP wrote the manuscript. JBC supervised the development of the work and helped in data interpretation and manuscript evaluation. JKC performed the statistical analysis. All authors read and approved the manuscript and agree to be accountable for all aspects of the research in ensuring that the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Funding

This work was supported by the National Health Insurance Ilsan Hospital grant (NHIMC 2019-20-005).

Data availability

This study’s data are available from the corresponding author, upon reasonable request.

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

This study was approved by the Institutional Review Board of the National Health Insurance Service Ilsan Hospital (NHIMC 2020-11-016). Informed consent was waived by the board because anonymous raw data were obtained from the National Health Insurance Service of Korea. The study was done in accordance with the Declaration of Helsinki.

Supplementary information

The online version contains supplementary material available at 10.1038/s41416-022-01807-5.