Introduction

Pancreatic cancer is one of the most fatal cancers [1], and the ninth most common cause of cancer deaths worldwide [1]. Although pancreatic cancer is rare in Asia (age-adjusted rate <1/100,000) compared with Western countries (age-adjusted rate 10–15/100,000), its incidence is higher in some high-income countries such as Japan (age-adjusted rate in 2011 8.7/100,000), similar to Western countries [1, 2]. Pancreatic cancer is the fourth most common cause of cancer death in Japan [3]. This tendency is similar to that in other developed countries [4]. It accounted for 8.4% of all cancer deaths in 2013 [3], with a 5-year relative survival rate of 7.0% in participants diagnosed in 2003–2005 in Japan [5].

Previous studies have shown a clear association between diabetes and an increased risk of pancreatic cancer [6–9]. Furthermore, a linear dose–response relationship between fasting blood glucose levels and pancreatic cancer risk has been reported in a meta-analysis [10]. A similar relationship was observed in Asian populations in a pooled analysis including Japanese and Korean cohorts [11, 12]. However, insulin sensitivity and insulin response differ among East Asians, Caucasians, and Africans [13], and the impact of obesity on diabetes is consequently also different among Asian, White, Hispanic, and Black populations [14]. Raimondi et al. noted that the incidence of pancreatic cancer varied greatly between regions because of differences in lifestyle factors such as diet, or environmental factors such as vitamin D exposure [4]. The relationships among diabetes, blood glucose level, and pancreatic cancer might thus be different in Japanese individuals. Although the impact of diabetes on pancreatic cancer risk has also been observed in Japanese individuals [8, 9], the dose–response relationship between fasting blood glucose levels and pancreatic cancer mortality and incidence has not been investigated.

A systematic review reported an association between glycosylated hemoglobin (HbA1c) and pancreatic cancer [15]. It is therefore possible that casual blood glucose levels, as well as fasting levels, may also be associated with pancreatic cancer, because HbA1c is associated with 24 h mean blood glucose [16]. However, such an association between casual blood glucose levels and pancreatic cancer has not been examined.

The purpose of this study was to examine the association between casual and fasting blood glucose levels and death due to pancreatic cancer in Japanese individuals, based on a large pooled analysis, including well-established cohorts enrolled in the Evidence for Cardiovascular Prevention from Observational Cohorts in Japan (EPOCH-JAPAN).

Materials and methods

Study population

The EPOCH-JAPAN database has been described in detail elsewhere [17]. Briefly, EPOCH-JAPAN is a pooling project for several well-qualified cohort studies in Japanese individuals. The database examining cancer deaths includes 12 cohorts (Tanno–Sobetsu, Ohsaki, Ohasama, Oyabe, YKK workers, Suita, RERF cohort, Hisayama, JACC, NIPPON DATA80, NIPPON DATA90, and Osaka), with a total of 99,951 participants.

We excluded two cohorts (Oyabe and JACC; 35,462 participants) because information of blood glucose was not pooled in the database. Additional exclusions were made in the case of participants who were aged <40 or ≥80 years (n = 12,000), had a reported history of cancer (n = 616), or had missing information about blood glucose (n = 5,486). A total of 46,387 participants (20,426 men and 25,961 women) were finally included in the study analysis.

Blood glucose level

Blood glucose levels were measured in the plasma in all but two cohorts (RERF and Osaka cohorts), in which they were measured in the serum. Blood collection time was pooled as dichotomous variable (fasting or not). If information on blood collection time was not fasting or missing, we treated the blood glucose level as casual.

Endpoint

In accordance with Family Registration Law in Japan, all death certificates are forwarded to the Ministry of Health, Labour and Welfare via the public health center in the area of residence [18]. Registration of death is required and completed by law. Some studies have used other sources, such as autopsy reports [19, 20], medical records [19–21], health examinations [21, 22], and questionnaires. The underlying causes of death were coded according to the International Classification of Diseases and Related Health Problems, the Ninth Revision (ICD-9) until the end of 1994, and the Tenth Revision (ICD-10) from the beginning of 1995 [23, 24]. Pancreatic cancer deaths were defined by codes 157 or C25.

Person-years of follow-up were counted for each participant, until the date of pancreatic cancer death, withdrawal from each cohort, or the end of the study period, whichever occurred first.

Statistical analysis

Participants were classified into three groups according to blood glucose level and anti-diabetic drug use: normal (casual blood glucose level <140 mg/dl, or fasting blood glucose level <110 mg/dl), prediabetes (casual blood glucose level 140–199 mg/dl, or fasting blood glucose level 110–125 mg/dl), and diabetes (casual blood glucose level ≥200 mg/dl, fasting blood glucose level ≥126, or anti-diabetic drug use) [25]. HbA1c was not included in the definition of diabetes because its information was not pooled in the database. Participants in the normal category were further divided in tertiles: low normal (casual blood glucose level <94 mg/dl, or fasting blood glucose level <90 mg/dl), middle normal (casual blood glucose level 94–107 mg/dl, or fasting blood glucose level 90–96 mg/dl), and high normal (casual blood glucose level 108–139 mg/dl, or fasting blood glucose level 97–109 mg/dl). Participants were thus classified into five groups in the present study.

Hazard ratios (HRs) and 95% confidence intervals (CIs) for pancreatic cancer mortality were estimated after stratifying casual and fasting blood glucose levels by cohort-stratified Cox proportional hazards regression analysis, which accounted for the variability in baseline hazards among cohorts, with low normal as a reference. The following variables were considered to be potential confounding factors in multi-adjusted 1: sex, age (continuous variable), body mass index (BMI) (<18.5, 18.5 to <25.0, and ≥25.0 kg/m2), cigarette smoking (never smoker, past smoker, and current smoker), and alcohol drinking (never drinker, past drinker, and current drinker). Tests for trends were calculated from categorical variable (ordinal scale). In addition, the analyses were repeated after excluding 2,950 participants (pancreatic cancer death n = 61) who were censored for the first 5 years after each baseline survey in multi-adjusted 2.

As a source of potential interaction, analyses were repeated after stratifying by sex, age (<65 or ≥65 years), BMI (<25.0 or ≥25.0 kg/m2), cigarette smoking (never smoker, or past, or current smoker), and alcohol drinking (never drinker, or past, or current drinker).

Statistical analyses were conducted using SAS software (version 9.3, Cary, NC, USA). All hypothesis tests were two-sided and based on a level of significance of 0.05.

Results

The characteristics of the ten cohorts are shown in Table 1. Each baseline survey was performed between 1977 and 1994, with the number of participants ranging from 791 in Ohasama to 12,789 in Ohsaki. The mean (standard deviation) follow-up period of the overall study population was 14.6 (±5.6) years, ranging from 11.1 years in Ohsaki to 23.0 years in Tanno–Sobetsu. There were 258 pancreatic cancer deaths during 678,001.0 person-years of follow-up.

Table 1 Characteristics of the cohort studies
Full size table

The baseline characteristics of the study participants according to blood glucose level are shown in Tables 2 and 3. Compared with low normal, participants in the diabetes group were older and more prevalent of men, BMI ≥25.0 kg/m2, past smokers, past drinkers, and current drinkers regardless of blood collection times. Current smokers were more prevalent according to only fasting blood glucose.

Table 2 Baseline distribution of the study participants according to casual blood glucose level
Full size table
Table 3 Baseline distribution of the study participants according to fasting blood glucose level
Full size table

Tables 4 and 5 show person-year totals, number of pancreatic cancer deaths, and HRs of pancreatic cancer mortality with 95% CIs according to casual and fasting blood glucose levels. In multi-adjusted analysis, compared with low normal, pancreatic cancer mortality increased significantly with increasing fasting blood glucose levels, though no similar significant association was observed for casual blood glucose levels. In multi-adjusted 1, the HRs for pancreatic cancer mortality according to casual blood glucose were 1.40 (95% CI 0.78–2.52) for prediabetes and 1.63 (0.87–3.05) for diabetes (p for trend = 0.202). The corresponding values for fasting blood glucose level were 2.83 (1.18–6.76) for prediabetes and 3.96 (1.56–10.08) diabetes (p for trend = 0.005). Similar non-significant and significant associations were observed in multi-adjusted two after exclusion of participants who were censored for the first 5 years of follow-up. After stratification according to sex, prediabetes and diabetes were associated with pancreatic cancer mortality in men, but there was no association between either elevated casual or fasting blood glucose levels in pancreatic cancer in women.

Table 4 Hazard ratios and 95% confidence intervals of pancreatic cancer mortality according to casual blood glucose level
Full size table
Table 5 Hazard ratios and 95% confidence intervals of pancreatic cancer mortality according to fasting blood glucose level
Full size table

We observed similar tendencies between blood glucose level and pancreatic cancer mortality in multi-adjusted 2 after stratification by age, BMI, smoking, and drinking (Supplemental Tables 1 and 2). All the above-stratified analyses showed non-significant interactions for casual and fasting blood glucose.

Discussion

The results of the present study revealed a dose–response relationship between fasting blood glucose levels and pancreatic cancer mortality, but no such association between casual blood glucose levels and pancreatic cancer mortality. The risk of pancreatic cancer mortality was particularly increased in individuals with not only diabetes but also prediabetes according to fasting blood glucose levels. This was the first study to investigate the risk of pancreatic cancer death according to fasting blood glucose levels in Japanese individuals, who have a unique dietary pattern [26–28] and casual blood glucose levels in the world.

Present study showed a dose–response relationship between fasting blood glucose levels and pancreatic cancer risk in Japanese individuals that was in accordance with a previous meta-analysis [10]. In general, various metabolic and hormonal factors related to glucose metabolism, including hyperglycemia, insulin resistance, and hyperinsulinemia, are proposed to be involved in the mechanisms underlying the onset of cancer [7, 8, 29, 30]. For example, hyperglycemia provides more glucose and induces pancreatic cancer cell growth [7, 29, 30], and is also associated with increased free radical formation, and may lead to the development of advanced glycation end products that can increase inflammation [31]. In addition, insulin resistance promotes the overproduction of reactive oxygen species [32, 33]. Chronic inflammation and oxidative stress caused by hyperglycemia or insulin resistance are known to cause DNA damage and are linked to the development of cancer [33, 34]. Although these mechanisms are not organ-specific in pancreatic cancer [33, 35], it has been suggested that hyperglycemia and insulin resistance may play important roles in the development of pancreatic cancer. Fasting blood glucose levels were positively associated with insulin resistance in diabetic and non-diabetic participants [36], which is in accordance with the current results. Our results suggested that the risk of pancreatic cancer death increased notably from a condition of prediabetes assessed by fasting blood glucose levels.

In contrast, there was no association between casual blood glucose and pancreatic cancer mortality. Casual blood glucose levels differ markedly depending on the measurement time after eating. Glucose levels measured at unknown times were defined as casual blood glucose levels in the present study, and these non-differential misclassifications will weaken any true association toward the null and induce the no association. Alternatively, previous reports showed that fasting blood glucose levels were associated with hepatic insulin resistance, while postprandial blood glucose levels were mainly linked to insulin resistance in skeletal muscle [37, 38]. Our results suggest that hepatic insulin resistance, not insulin resistance in skeletal muscle, may be an important factor in the pancreatic cancer development. Furthermore, blood glucose elevations during fasting, or whole day rather than transient postprandial hyperglycemia, might also increase the risk of pancreatic cancer given that HbA1c levels, as an index of 24 h mean blood glucose, were shown to be associated with pancreatic cancer in a previous report [15].

Stratification by sex showed that the above associations only occurred in men. However, Jee et al. showed that fasting blood glucose levels were associated with pancreatic cancer risk in both men and women [12]. It is possible that the lack of any association between fasting blood glucose levels and pancreatic cancer mortality in women in the current study was the result of the small number of events. Notably, there were only two pancreatic cancer deaths in women in the prediabetes and diabetes fasting blood glucose groups, respectively, and no pancreatic cancer deaths in the diabetes group after exclusion of a participant who was censored for the first 5 years after baseline. We conducted additional stratified analyses to examine the interactions with age, BMI, smoking, and drinking. Obesity is associated with both metabolic disorders and pancreatic cancer death [1, 39]. However, even though Lin et al. showed a significant increase in the risk of pancreatic cancer death in only underweight individuals with a history of diabetes, there was no interaction between BMI ≥18.5 kg/m2 and history of diabetes on pancreatic cancer in Asians (p for interaction = 0.07) [40]. In the present study, the association between blood glucose level and pancreatic cancer deaths was not modified by BMI <25 kg/m2 or ≥25 kg/m2. Meanwhile, never smokers and never drinkers showed no association between fasting blood glucose levels and pancreatic cancer mortality. The prevalence of women in whom fasting blood glucose level was not associated with pancreatic cancer mortality was higher in these strata.

This study had some limitations. First, information on the history of cancer was only pooled in a few cohorts, and it is possible that the present results were affected by individuals’ cancer history. However, we conducted analyses excluding participants who were censored for the first 5 years after each baseline, and the observed association was robust. We considered that the possible effect of reverse causation, whereby raised blood glucose levels could be due to a history of pancreatic cancer or undiagnosed pancreatic cancer, would be eliminated because of the very low 5-year survival of pancreatic cancer [1]. After exclusion, most analyzed participants would not have any history of pancreatic cancer, or any undiagnosed pancreatic cancer at each baseline. Second, blood glucose was measured at the beginning of the follow-up period. The distribution of single baseline glucose measures is therefore wider than the distribution of true usual glucose values because of regression to the mean, resulting in potential regression dilution bias [41]. This bias weakens any true association towards the null. Participants might have been reclassified when they were examined during the follow-up period, and this bias may have masked any association between casual blood glucose levels and pancreatic cancer mortality. Third, we observed few pancreatic cancer deaths in women in the prediabetes and diabetes, respectively, therefore the statistical model and result of women in stratified analysis might be instable.

In conclusion, fasting, but not casual blood glucose levels showed a linear association with the risk of pancreatic cancer death in Japanese individuals. The risks of pancreatic cancer mortality were particularly significantly increased in individuals with prediabetes and diabetes according to fasting blood glucose levels. The detailed mechanisms responsible for the different impacts of casual and fasting blood glucose on pancreatic cancer remain unknown, but fasting blood glucose appears to be a more reliable predictive index of pancreatic cancer death than casual blood glucose.