Body mass index, physical activity and the risk of pancreatic cancer in relation to smoking status and history of diabetes: a large-scale population-based cohort study in Japan–The JPHC study
- First Online:
- Cite this article as:
- Luo, J., Iwasaki, M., Inoue, M. et al. Cancer Causes Control (2007) 18: 603. doi:10.1007/s10552-007-9002-z
- 291 Views
The effects of BMI and physical activity on the risk of pancreatic cancer were investigated in a large population-based cohort study in Japan (JPHC study). In particular, we explored how these effects were influenced by smoking status and a history of diabetes.
In total, 99,670 participants (47,499 men, and 52,171women) were followed for an average of 11 years through the end of 2003. Hazard ratios (HR) and their 95% confidence intervals (CI) were estimated by Cox proportional hazards models.
A total of 224 incident cases (128 men, 96 women) of pancreatic cancer were identified. A statistically significant excess risk of pancreatic cancer was associated with current smoking (HR = 1.8, CI 1.1–3.0) and a history of diabetes (HR = 2.1, CI 1.3–3.5) among men. A similar increase in risk associated with current smoking (HR = 2.0, CI 0.9–4.4) and diabetes (HR = 1.5, CI 0.7–3.5) was also seen among women. BMI was inversely associated with risk of pancreatic cancer among men, especially among current male smokers or diabetes patients, but no association was found among women. Leisure-time physical activity was not associated with risk in either men or women.
Our study confirms the association between cigarette smoking, history of diabetes and increased risk of pancreatic cancer. However, our data suggest that the association between BMI and risk of pancreatic cancer in this Japanese population may be different from that in Western populations.
KeywordsBMIPhysical activityPancreatic cancerRisk factorsSmokingdiabetes
Worldwide, more than 200,000 people die annually of pancreatic cancer. The highest incidence and mortality rates of this condition are reported in developed countries . However, while mortality rates have decreased slightly or remained stable in most developed countries over the past several decades, the age-standardized mortality rate in Japan jumped from 1.4 to 12.5 per 100,000 men between 1950 and 1995 , and although it has leveled or is leveling off since 1985, it is now the fifth most common cause of cancer death in Japan . Pancreatic cancer is one of the most deadly of all malignant neoplasms. Most patients are diagnosed at an advanced stage, and about 85% of patients die within one year . However, except for an association with tobacco smoking, chronic pancreatitis, type II diabetes [5, 6] and obesity , little is known about its etiology.
Recent emerging evidence has suggested that insulin resistance and abnormal glucose metabolism may be the risk factors in the development of pancreatic cancer [5–9]. The consistently observed association between diabetes and increased risk  supports this hypothesis, although diabetes can be a consequence of pancreatic cancer . The association between diabetes and pancreatic cancer risk may be the result of years of elevated post-load glucose concentration, hyperinsulinemia and gradual impairment of glucose tolerance . Evidence for this comes from a prospective study in nondiabetics, which reported that post-load glucose concentration was directly associated with pancreatic cancer risk in men and women , and from several large prospective studies, which identified a graded dose-response association between fasting glucose and pancreatic cancer [11, 12].
Body mass index (BMI) and physical activity, two important determinants of insulin sensitivity , have attracted growing interest. Previous epidemiological studies [14–17] have reported elevated risks of pancreatic cancer among overweight and obese individuals. Most of these studies have been conducted in Western populations, however, which differ from Asian populations, in whom the majority are relatively lean. Studies examining the influence of physical activity on the risk of pancreatic cancer are relatively few, and have provided conflicting results [14, 15, 18–23].
In this study, the effects of BMI and physical activity on the risk of pancreatic cancer were investigated in a large prospective population-based cohort study in Japan (JPHC study). In particular, we explored how these effects were influenced by smoking status and a history of diabetes. In addition, we also examined the association between smoking and a history of diabetes, the two most consistent risk factors, and the risk of pancreatic cancer in the same dataset.
The Japan Public Health Center-based Prospective Study (JPHC study) was launched in 1990, when Cohort I was established. Cohort II was added in 1993. The details of the study design have been described elsewhere . Briefly, the JPHC study consisted of 11 public health center (PHC)-based areas throughout Japan (Iwate-Ninohe, Akita-Yokote, Nagano-Saku, Okinawa-Chubu, Tokyo-Katsushika in Cohort I; and Ibaraki-Mito, Nigata-Nagaoka, Kochi-Chuohigashi, Nagasaki-Kamigoto, Okinawa-Miyako, Osaka-Suita in Cohort II), with 140,420 middle-aged residents in total. Tokyo-Katsushika PHC area was not included in the present analysis due to a lack of cancer incidence data.
The study population was defined as all registered Japanese inhabitants between the ages of 40 and 59 years (Cohort I) and 40 and 69 years (Cohort II) at the beginning of each baseline survey in all PHC areas but Osaka-Suita. In the Osaka-Suita area, two separate cohorts were established. The first was defined as all residents aged 40 or 50 years in fiscal year 1993, because these were invited to participate in a comprehensive health checkup program conducted by the city. The second cohort was defined as a part of the Suita study , in which subjects aged 30–79 years between 1989 through 1992 were randomly selected based on the population registry of the city after stratification into sex and 10 year age groups. Among these, participants who still lived in Suita City on 1 April 1993, and were aged 40–69 years at that time were used for the JPHC study. In total, 133,324 subjects were initially identified. However, during the follow-up period, 239 subjects were found to be ineligible and were excluded because of non-Japanese nationality, (n = 51), late report of emigration occurring before the start of the follow-up period, (n = 178), or ineligibility because of an incorrect birth date or duplicate identification number. (n = 10). Finally, a population-based cohort of 133,085 subjects was established.
Baseline questionnaires and exposure measurements
Self-administered questionnaires were distributed to all registered residents in 1990 (Cohort I) and in 1993 and 1994 (Cohort II). A total of 106,326 subjects responded to the questionnaires, for a response rate of 80%. These questionnaires contained items on demographic characteristics, personal medical history, family history of malignant neoplasm, smoking and alcohol drinking, dietary habits, physical activity, and other lifestyle factors.
All exposures in our analyses were collected at baseline for all subjects. Body mass index (BMI) was calculated as weight in kilograms divided by the square of height in meters. By comparing the self-reported height and weight with available data from health checkups (11,274 men, 21,196 women), we confirmed that the self-reported BMIs were slightly lower than the measured BMIs, with Spearman correlation coefficients of 0.89 in men and 0.90 in women. Physical activity was obtained in terms of the frequency of sports in leisure time (rarely, 1–3 days/month, 1–2 days/week, 3–4 days/week, and almost everyday). The Spearman’s correlation coefficients for the relationship between energy expenditure estimates (METs /day) based on this question and 24 h physical activity record using 110 volunteer samples was 0.48 [unpublished data, Fujii H et al.].
Other potential confounding variables included age at enrollment, gender, smoking status (never, past, current), history of diabetes mellitus (yes or no), history of cholelithiasis, and alcohol intake. Additional questions on smoking habit consisted of age at the start of smoking, average number of cigarettes smoked per day for both current and former smokers, and age of quitting smoking for former smokers. The cumulative effect of smoking for current smokers was evaluated by pack-years, which was defined by multiplying the number of packs of cigarettes smoked per day by the number of years the person has smoked. Alcohol intake was defined as nondrinkers, occasional drinkers (1–3 days/month) and regular drinkers (1–2 days/week or more) based on the frequency of consumption in the questionnaire. In addition, among regular drinkers, the weekly ethanol intake was calculated by multiplying the frequency per week by the usual daily amount of alcohol.
Of the responding population, we excluded 2,228 subjects who had a history of self-reported cancer at baseline, 151 subjects who were reported to have moved away before the baseline survey, and 4,277 subjects who had incomplete information on BMI, physical activity, diabetes and smoking, including 102 subjects with a BMI of less than 14 or more than 40 because of the possibility that the data were unreliable. Finally, 99,670 subjects (47,499 men, and 52,171 women) remained for analysis.
Follow-up and case ascertainment
Subjects were followed from the date of the baseline survey through 31 December 2003. In Japan, residency and death registration are required by law, and the registries are believed to be complete. Residence status and survival were confirmed annually using the residential registers kept by each municipality in each of the study areas. Inspection of the resident register is open to the general public under the resident registration law. Information on the cause of death was obtained by examining the death certificate provided by the Ministry of Health, Labor, and Welfare after permission was obtained from the Ministry of Internal Affairs and Communications.
The occurrence of cancer was identified by active patient notification by the major local hospitals in the study areas and data linkage with population-based cancer registries, with permission from each of the local governments responsible for the registries. Death certificate information was used as a supplementary information source. The site and histology of each cancer were coded using the International Classification of Diseases for Oncology, Third Edition (ICD-O-3).
During 1,063,235 person-years of follow-up (average 11 years), a total of 224 incident cases (128 men, 96 women) of pancreatic cancer were identified (no cases were diagnosed as code C254 – Islets of Langerhans cases). Of these, 47.5% of diagnosed pancreatic cancer cases were histologically confirmed, and 8.9% were identified only by death certificate. Mean (± s.d.) age at the time of diagnosis was 62.8 (7.6) years for men and 65.8 (7.3) years for women.
Person-years of follow-up were counted from the date of the baseline survey until the date of pancreatic cancer diagnosis, the date of moving out of a study area, the date of death, or the end of 2003, whichever came first. Persons lost to follow-up were censored on the last confirmed date of their presence in the study area. The age–adjusted incidence rate for pancreatic cancer was calculated by dividing the number of pancreatic cancer cases by the number of person-years standardized to the age distribution of the study population. Hazard ratio (HR) for the development of pancreatic cancer was estimated using the Cox proportional hazards model, using the time scale of attained age in place of follow-up time owing to the strong association of pancreatic cancer risk with age. PHC areas were treated as strata, which allowed a different baseline hazard for each PHC stratum. Linear trends were tested in the Cox models by treating the categories as ordinal variables. In addition, we evaluated whether the effects of BMI and physical activity on the risk of pancreatic cancer were influenced by smoking status or history of diabetes. Tests of interaction were performed by entering into the model multiplicative interaction terms between any two risk factors of interest.
In addition, to eliminate pre-clinical cases that might have experienced weight loss before completing the baseline questionnaires and to minimize the possible inverse influence of pancreatic cancer on diabetes, we also performed analyses that excluded the first four years of follow-up. Finally, to better control the strong confounding effect of smoking and history of diabetes on the effects of BMI and physical activity on the risk of pancreatic cancer, we restricted these analyses to those who were never-smokers and had no history of diabetes.
The assumption of proportional hazards was tested based on the cumulative sums of martingale residuals with a Kolmogorov-type supremum test , in which 1,000 realizations were used. We tested all exposure variables of interest and all potential confounding variables, and found the lowest p value was 0.09 for smoking status. This indicates that the proportional assumption for each variable in our study was satisfied.
Selected baseline characteristics of the study subjects in relation to body mass index (BMI) among 99,670 subjects (47,499 men and 52,171 women) in the JPHC Study, Japan, 1990–2003
Men (47,499 subjects)
Women (52,171 subjects)
No. subjects (%)
Age at baseline (mean, years)
≥1 day/weeka (%)
Regular drinkerc (%)
History of diabetes (%)
History of cholelithiasis (%)
Hazard ratios (HRs) and their 95% confidence intervals (CIs) of pancreatic cancer by exposures of interest at baseline among 47,499 men, JPHC Study, Japan, 1990–2003
No. cases (128)
Age-adjusted incidence ratea
Age-adjusted HR (95% CI)
Multivariable-adjusted HR (95% CI)b
Multivariable-adjusted HR (95% CI)c (95 cases)
p for trend
History of diabetes
p for trends
History of cholelithiasis
Ethanol < 245 g/wk
p for trend
Hazard ratios (HRs) and their 95% confidence intervals (CIs) of pancreatic cancer by exposures of interest at baseline among 52,171 women, JPHC Study, Japan, 1990–2003
Age-adjusted incidence ratea
Age-adjusted HR (95% CI)
Multi-adjusted HR (95% CI)b
Multi-adjusted HR (95% CI)c (67 cases)
History of diabetes
P for trend
History of cholelithiasis
Hazard ratios (HRs) and their 95% confidence intervals (CIs) for pancreatic cancer in relation to body mass index (BMI) and leisure-time physical activity, by smoking status and diabetes among 47,499 mena
History of diabetes
In this prospective study in a Japanese population, we confirmed the presence of positive associations between cigarette smoking, a history of diabetes and increased risk of pancreatic cancer. However, we did not observe an association between excess weight, defined by a BMI of 25 kg/m2 or more, leisure-time physical activity and risk of pancreatic cancer.
The main purpose of the present study was to examine the effects of obesity and physical inactivity on the risk of pancreatic cancer. To date, at least 26 studies have investigated the association between obesity and pancreatic cancer risk, comprising 18 prospective cohort [10, 14, 15, 17, 18, 20–23, 27–35] and eight case–control studies [36–43]. Results have been inconsistent, however, with 15 reporting a positive association [10, 14, 15, 17, 20, 21, 31–38, 43] (in men only in two [10, 37]) and 11 reporting no association [18, 22, 23, 27–30, 39–42]. When evaluated critically , the literature indicates that obesity (BMI > 30 kg/m2) plays an important role in the etiology of pancreatic cancer. Since 2000, positive associations between BMI and risk of pancreatic cancer have been consistently reported in the largest studies, whereas smaller studies have found no significant association. However, among these 26 studies, only one case–control study  in China and one prospective study in Korea  were conducted in Asian countries. In the Chinese case–control study, comparison of the top to lowest quartile of BMI showed evidence of an increasing risk of pancreatic cancer with increasing body mass index by, with no clear cutoff point, while in a prospective Korean study, no association was found on comparison of BMIs of < 25–≥ 25 kg/m2, consistent with our present findings.
The main mechanism of the link between obesity and pancreatic cancer is thought to be an effect on insulin sensitivity, similar to the plausible mechanism underlying that between diabetes and pancreatic cancer. It is hypothesized that a hyperinsulinemic state allows increased levels of insulin to pass through pancreatic exocrine cells, bind to insulin receptors, and trigger mitotic activity [8, 10, 44]. Increased insulin also can down-regulate insulin-like growth factor binding protein-I, thereby increasing the bioavailablility of insulin-like growth factor-I, a process which has been shown to stimulate cell proliferation in vitro [45–49]. An alternative mechanism for the association between BMI and pancreatic cancer may be related to lipid peroxidation-related DNA adducts. DNA adduct levels were reported to be significantly higher in pancreatic cancer patients , and positive correlations were found between obesity and levels of lipid peroxidation . Thus, an alternative explanation for the observed association with BMI may be an increase in DNA damage to the pancreas caused by increased lipid peroxidation in individuals with elevated BMIs.
We did not observe an increased risk of pancreatic cancer in subjects with a BMI of 25 or higher in the present study, which was consistent with most previous Western findings for overweight people with a BMI of 25–29.9. It is interesting that a recent Korean study , the only prospective study conducted in an Asia population to date, observed similar results to ours, namely that BMI showed no association with the incidence of pancreatic cancer but did show any significant inverse association with mortality of pancreatic cancer among Korean men. This lack of association might be accounted for by the low prevalence of obese subjects (only 2.6% for BMI ≥30 with one female case) in the lean Japanese population. In our study, the cutoff point of 25 for the top category of BMI may not have been sufficiently high to allow the detection of an association, given that risk in most of the western studies was greatest among those with a BMI of 30 or higher in both men and women. However, our lower cutoff point for this Asian population acknowledges that morbidity and mortality occur at lower BMIs among Asians, in whom obesity is defined as a BMI of 25 or more [52, 53]. Therefore, our negative association may not be explained by the lower BMI cutoff point only.
As noted, BMI was associated with smoking and diabetes in our population, which confounded and modified the relationship between BMI and pancreatic cancer. We recorded the highest risk of pancreatic cancer among lean patients with diabetes, which may be explained by differences occurring generally in the duration or stage of diabetes among different BMI categories. Diabetic patients with lower BMI may be more likely to experience a longer duration of diabetes than those with normal or higher BMI; as we had no information on the date of diagnosis of diabetes, however, we could not examine this further. Moreover, our observed inverse association between BMI and the risk of pancreatic cancer among current smokers may be due to a different intensity of smoking across different BMI categories, since body weight tends to be lower among moderate and heavy smokers than in non-smokers or light smokers . However, the small number of cases in the stratified analyses meant that the possibility of chance findings could not be ruled out. Findings derived from these cases should therefore be interpreted with caution.
We also considered various sources of error that would have tended to bias the risk estimates of pancreatic cancer by over-weighting towards the null value. One possibility was in-differential misclassification of exposure. However, the reported correlation coefficient between measured and reported BMI in this cohort was approximately 0.9 , seemingly insuring that a strong association could not have been missed. A second possibility was weight loss due to pre-clinical disease, but results did not change after data from the first four years of follow-up were excluded. A third possibility for the absence of association might have been biased ascertainment of pancreatic cancer cases, although the quality of the cancer registry system over the study period was satisfactory. However, the strengths of association between smoking and risk and between diabetes and risk were similar, making it unlikely that any such diagnostic misclassification would have affected the association with BMI only.
Previous studies on physical activity and pancreatic cancer risk are inconclusive. Of nine studies [14, 15, 18, 19, 22, 27–28, 37] that reported the association between recreational physical activity and pancreatic cancer, four found an inverse association [14, 19, 29, 37] while the other five found no or an increase but not significant association. Our study detected an association between physical inactivity and increased risk among current smokers and among diabetic patients, but no significant association between physical activity and risk in general, or among people who were never smokers, or among those who had no history of diabetes.
Given that exercise is known to improve glucose tolerance, even in the absence of weight loss , physical activity may in general be associated with a decreased risk of pancreatic cancer, and may represent a modifiable risk factor for this cancer. Here, however, we had only limited information on physical activity, with no data other than that for activity at leisure, for example. This lack of data on the intensity of activity may have increased the potential for misclassification of exposure measurement, possibly accounting at least in part for our negative association with physical activity.
In addition to these limitations, a further limitation is that exposure information was limited to that at baseline. Changes in the exposures of interest over time would have led our estimates towards the null value.
Against this, a number of strengths of this study should also be mentioned, including its prospective design, near-complete follow-up (proportion of losses to follow-up: 0.05%) , and detailed information on potential risk factors of pancreatic cancer.
In summary, our study confirms an association between cigarette smoking and history of diabetes and an increased risk of pancreatic cancer. However, our data suggested that the association between BMI and the risk of pancreatic cancer in this Japanese population may be different from that in Western populations. Further studies on BMI, physical activity and energy intake are needed, especially in lean Asian populations.
This study was supported by a Grant-in-Aid from the Cancer Research and Third-Term Comprehensive Control Research for Cancer from the Ministry of Labor, Health and Welfare of Japan. J.L. was partly supported by the SVENSKA SÄLLSKAPET FÖR MEDICINSK FORSKNING (SSMF) and the Karolinska Institutet Travel Fund.