Introduction

Malignant pancreatic tumors, particularly pancreatic cancer (PC), are associated with notably poor prognoses, as evidenced by a 5-year overall survival rate of only 8.5% in Japan1. Conversely, patients with tumors measuring 10 mm or smaller have a significantly higher 5-year survival rate, exceeding 80%2. Nonetheless, cancer registry data reveal that only 1.7% and 4.1% of all PC patients are classified as UICC stage 0 and IA, respectively2. Therefore, early diagnosis is critical to improving the prognosis of PC. Identifying risk factors for PC is crucial for early-stage detection. The Japanese clinical practice guidelines for pancreatic cancer3 list several risk factors: familial history4, smoking5,6,7, drinking8,9, diabetes10,11,12,13, obesity14,15,16,17,18, chronic pancreatitis19,20,21, pancreatic cystic lesions22,23,24,25, gallstones/cholecystectomy26, H. pylori infection27, HBV infection28 and HCV infection29. Diabetic patients, who regularly visit the hospital, present a practical opportunity for pancreatic surveillance. However, with approximately 10 million diabetic patients in Japan30,31, it is impractical to conduct pancreatic surveillance on all of them. Effective surveillance for PC could be achieved by identifying diabetic patients at high-risk for PC who already require regular hospital visits. Nevertheless, there is a notable lack of research on pancreatic cancer risk factors within the diabetic population. This study, therefore, aimed to identify the risk factors that contribute to the development of pancreatic cancer in diabetic patients.

Materials and methods

Study design and data sources

Our study employed a cohort design utilizing the Shizuoka Kokuho database (SKDB)32. The SKDB is an administrative claims database for beneficiaries in Shizuoka Prefecture’s municipal government insurance program, including the National Health Insurance and late-stage medical care system for the elderly. Shizuoka Prefecture, with a population of approximately 3.6 million, features a climate and demographic profile representative of Japan. The SKDB covers a regional, population-based longitudinal cohort of 2,571,418 individuals from Shizuoka. All data from SKDB enrollees underwent preprocessing, which involved extensive cleaning and anonymization32. This dataset comprises basic subscriber information (e.g., sex, age, zip code, observation period, and reason for disenrollment, including death) and claims from public health insurance organizations (below 75 years for the National Health Insurance system and above 75 years for the Late-stage Elderly Medical Care System). The utility of the SKDB in real-world risk factor analysis is underscored by its comprehensive data on mortality and follow-up attrition, derived from the Basic Resident Registration System. The database has supported various studies, particularly in risk factor analysis33,34,35.

Observation period and study population

In this database-nested cohort study, we utilized the SKDB from April 1, 2012, to September 30, 2021. Figure 1 illustrates the study schema. The investigation involved examining enrollees through the individually linked data in the databases, focusing on their annual health checkups and insurance claims. For each enrollee, the data availability spanned from the later of either their insurance registration date or April 2012, whichever was later, to the earlier of either their insurance withdrawal date or September 30, 2021. Diabetic patients were selected for cohort inclusion based on identification and extraction from the database. The exclusion criteria included those diagnosed with pancreatic cancer prior to or concurrent with their initial health checkup or insurance claim, a history of any cancer (apart from nonmelanoma skin cancer), and an observation period of less than one year. Additionally, diabetic patients who developed pancreatic cancer within 1 year following their diabetes diagnosis were excluded to avoid confounding with secondary diabetes caused by pancreatic cancer.

Figure 1
figure 1

Study schema.

Full size image

Definition of diabetic patients and outcome

Diabetic patients were identified by the prescription of one or more antidiabetic agents (Supplementary Table 1), including insulin (since antidiabetic agents are not commercially available in Japan), and insurance claims indicating diabetes mellitus (E10, E11, or E14) during the baseline period. In contrast, nondiabetic patients had no such prescriptions in the same timeframe. The primary outcome of the study was the incidence of pancreatic cancer, determined using International Classification of Disease 10 (ICD-10) codes from claims data (Supplementary Table 2). To improve specificity, provisional diagnoses of pancreatic cancer and diabetes were excluded.

Variables

Our analysis incorporated several candidate confounders, including age, sex, the Charlson-Elixhauser comorbidity index, and covariates previously identified in relevant reports36,37,38,39,40,41,42. We determined the presence of individual comorbidities using standard definitions from the Charlson-Elixhauser comorbidity index43,44. In assessing specific comorbidities, we utilized a one-year period prior to the medical examination date as the search timeframe. A comorbidity was considered present if it was confirmed in the insurance claims data. The type of diabetes agent was excluded from our variables. This exclusion stems from the study’s design, which is not aligned with the Prevalent New-user Design, and therefore does not investigate the relationship between PC development and diabetes medications45,46,47.

Statistical analysis

Frequencies and percentages were calculated for categorical variables, and the mean and standard deviation were calculated for continuous variables. For comparing continuous and categorical variables across groups, we employed the t test and chi-square test, respectively. Univariate and multivariate cause-specific Cox proportional hazards regression analyses were performed to explore factors associated with pancreatic cancer onset. This method is suitable for our study as it deals with time-to-event data, allowing us to account for the varying follow-up times among participants. The hazard ratio (HR), 95% confidence interval (CI), and p-value of Wald test were calculated to estimate the relative risk of pancreatic cancer over time. In the regression analysis, we included both reported risk factors (e.g., pancreatic cysts) and potential risk factors (e.g., critical etiological and epidemiological factors). To ensure a conservative approach, no model selection was carried out, and all variables that reached statistical significance in the univariate regression analysis were entered into the multivariate regression analysis. We excluded one of two highly correlated variables (absolute Spearman’s correlation coefficient > 0.4) from the multivariable model to avoid multicollinearity. All p-values were two-sided, with a threshold of < 0.05 set for statistical significance. The 95% CI for annual incidence rate of pancreatic cancer was calculated using the Normal approximation to the Poisson distribution, following the method outlined in Rosner’s “Fundamentals of Biostatistics” (5th Ed). The Kaplan–Meier method estimated overall survival for the pancreatic cancer population, with cumulative incidence curves compared using the log-rank test. All analyses adhered to the intention-to-treat principle. For statistical analyses, we used SAS version 9.4 (SAS Institute, Cary, NC, USA), EZR version 1.55 (Saitama Medical Center, Jichi Medical University, Saitama, Japan), and R statistical software version 4.2.1 (The R Foundation for Statistical Computing).

Ethics

The data of all enrollees were anonymized to ensure participant confidentiality32. This study adhered to the principles of the Declaration of Helsinki. The study protocol received approval from the Medical Ethics Committee of the Shizuoka Graduate School of Public Health (SGUHPH_2021_001_067), and this committee waived the requirement for informed consent.

Results

Study population and baseline characteristics

The Shizuoka Kokuho database, which comprises 2,571,418 individuals, identified 212,775 as diabetic patients with a baseline period exceeding one year, as detailed in Fig. 2. Of these patients, 1755 developed pancreatic cancer during the observation period, which had a median duration of 5.1 years and a maximum of 8.5 years. The annual incidence rate of pancreatic cancer in this group was 166.7 cases per 100,000 person-years [95% CI 158.9–174.5], and incidence rates in patients newly diagnosed with diabetes and those with an existing diagnosis were 156.7 cases per 100,000 person-years [95% CI 141.9–172.8] and 169.9 cases per 100,000 person-years [95% CI 161.0–179.1], respectively. Table 1 presents the baseline characteristics of both the patients diagnosed with pancreatic cancer and those without the disease.

Figure 2
figure 2

Flowchart for inclusion of participants in the study.

Full size image
Table 1 Patients’ baseline characteristics and variables for univariable Cox model analysis.
Full size table

Identification of risk factors for pancreatic cancer in diabetic patients

Potential risk factors and significant variables were evaluated using univariate Cox regression analysis (Table 1). Identified candidate risk factors included older age, male sex, coexisting liver disease, cardiac arrhythmias, hypertension, chronic pancreatitis, and pancreatic cystic lesions, each with p-values below 0.05. These variables were subsequently incorporated into a multivariate model. The multivariable regression results, presented in Table 2, indicated that older age and male gender were associated with an increased risk of pancreatic cancer (HR: 1.267 [95% CI 1.150–1.395]). Liver disease emerged as a notable risk factor (HR: 1.13 [95% CI 1.01–1.26]). Significant correlations were found between the onset of PC and both chronic pancreatitis (HR: 1.98 [95% CI 1.48–2.64]) and pancreatic cystic disease (HR: 4.79 [95% CI 3.43–6.67]). Neither the type of diabetes nor the presence of diabetic comorbidities was associated with the onset of PC. Furthermore, the risk of pancreatic cancer onset did not significantly differ between patients newly diagnosed with diabetes and those with an existing diagnosis.

Table 2 Multivariable Cox regression analysis of pancreatic cancer onset.
Full size table

Survival curve in pancreatic cancer onset subgroup

During the observation period, among 1755 diabetic patients diagnosed with PC, 1177 (67.1%) patients died, with a median survival period of 297 days. The 6-month, 1-year, 3-year, and 5-year survival rates were 60.5%, 45.8%, 28.2% and 21.6%, respectively. The overall survival rates of diabetic patients with PC are shown in Fig. 3a. A comparison of survival rates categorized by age revealed a decline in survival with increasing age (p < 0.001, Fig. 3b). The median survival periods for the age groups < 65 years, 65–< 75 years, 75–< 90 years, and ≥ 95 years were 658 days, 620 days, 234 days and 89 days, respectively.

Figure 3
figure 3

The overall survival rates of diabetic patients with pancreatic cancer.

Full size image

Discussion

While numerous risk factors for PC have been identified in the general population, this large-scale cohort study specifically examines these factors among individuals with diabetes mellitus. Our findings indicate that among patients with a history of diabetes mellitus for more than one year, the annual incidence of PC was 166.7 cases per 100,000 person-years [95% CI 158.9–174.5]. In contrast, the incidence in the general population of Japan is 34.8 cases per 100,000 person-years1. Meta-analyses report the risk of PC in diabetic patients with a relative risk between 1.710 and 1.911. Although it is known that the risk of pancreatic cancer is elevated in patients with diabetes, the present study, which reports on the incidence of pancreatic cancer among diabetic patients in a large cohort, is of considerable value and our results suggests a higher incidence of PC in our diabetic cohort. Given this increased incidence, monitoring diabetic patients could enhance medical outcomes. Furthermore, significant risk factors identified in diabetic individuals include advanced age, male sex, a history of liver disease, chronic pancreatitis, and pancreatic cystic lesions. These findings could be instrumental in promoting early diagnosis of PC in diabetic patients, suggesting the need for abdominal screening in diabetic patients with these risk factors.

The identified risk factors for the onset of pancreatic cancer in diabetic patients, except for a history of liver disease, were consistent with those in the general population. However, the existing risk factors and their associated risk values listed in the guidelines are often a compilation of each study, reported disparately across different population backgrounds. In this study, we have re-evaluated all previously reported risk factors simultaneously in a multivariate analysis within a single large-scale cohort study. The factors thus selected are considered to independently contribute to the incidence of pancreatic cancer in the diabetic cohort, which is a strength of this research.

Pancreatic cancer typically occurs more frequently in males than in females and is more prevalent in older individuals48,49. In our study, being male was identified as a risk factor. This gender disparity may be partially attributed to the higher rates of smoking and heavy drinking among males, factors known to escalate the risk of PC5,6,7,8,9. Furthermore, the role of potential yet unidentified factors (e.g., genetic predisposition) in influencing cancer incidence between genders warrants consideration. In diabetic patients, similar to the general population48,49,50,51, age is a significant risk factor for pancreatic cancer. The disease is rare before the age of 50 years, with the median age at diagnosis for most pancreatic adenocarcinoma cases in the general population being approximately 70 years48,52. The results in diabetic patients were similar to those in the general population, suggesting that intensive PC surveillance may not be necessary for younger diabetic individuals.

Pancreatic cystic lesions including intraductal papillary mucinous neoplasm are known risk factors for PC22,23,24,25 in the general population. Tada et al. reported that in Japanese patients with pancreatic cystic lesions, the observed incidence of pancreatic cancer was 22.5 times higher (95% confidence interval 11.0–45.3) than the expected mortality from this cancer in the general population23. However, these studies do not directly compare the risk for the incidence of pancreatic cancer with those of the general population. While multiple guidelines exist for the management of pancreatic cystic lesions53,54,55,56, there is a notable variation among them, and finding the ideal balance between surveillance and intervention for specific types of lesions remains a subject of considerable debate. Buerlein et al.57 posited that prior to initiating a surveillance program for a pancreatic cystic neoplasm, it is crucial to consider various patient-specific factors and to discuss with the patient the potential risks and benefits of undergoing a surveillance program, as individuals’ tolerance for risk can vary significantly. In this study, the prevalence of pancreatic cystic disease in diabetic patients was approximately 0.4%, which is lower than rates reported in recent studies58. Notably, this study is the first to identify pancreatic cystic lesions as a significant risk factor for developing pancreatic cancer in diabetic individuals, evidenced by a high hazard ratio (HR) of 4.79 [95% CI 3.43–6.67] (Table 2). The findings of the current study suggest that pancreatic cystic lesions in diabetic patients warrant regular monitoring.

Similarly, chronic pancreatitis is recognized as a risk factor for PC in the general population19,20,21. The international consensus guidelines for chronic pancreatitis59 note that the risk of developing pancreatic cancer may differ among patients with chronic pancreatitis, depending on factors such as lifestyle and medical or surgical interventions. Ueda et al. have reported that the standardized incidence ratio (SIR) for pancreatic cancer in Japanese patients with chronic pancreatitis was 11.8 (95% CI 7.1–18.4)21; however, similar to pancreatic cystic lesions, this was not the result of an analysis of risk factors within a large-scale cohort population. Our study suggests that chronic pancreatitis may also act as a risk factor for the development of PC in diabetic patients. Consequently, patients with concurrent chronic pancreatitis and diabetes mellitus may benefit from pancreatic cancer surveillance.

Based on our findings, the coexistence of liver disease emerged as a risk factor in patients with diabetes. The underlying biological mechanisms linking liver disease and pancreatic cancer remain unclear; however, several liver diseases are known risk factors for pancreatic cancer28,29. Previous research indicates that serum levels of pancreatic enzymes rise with the progression of liver disease in diagnosed patients60,61. This suggests that liver disease may exert a biologic effect on the pancreas, particularly in those with diabetes.

Hypertension and arrythmia were identified as risk factors in univariate analysis but not as significant risk factors in multivariate analysis. Although hypertension has not been identified as a clear risk factor for pancreatic cancer, several solid tumors have been associated with hypertension62. Similarly, the evidence for a strong association between the development of pancreatic cancer and arrhythmia is limited. Nonetheless, chronic pancreatitis, a risk known factor for pancreatic cancer, can lead to arrhythmia63. Additionally, elevated levels of branched-chain amino acids, which are associated with early pancreatic cancer development, have been linked to arrhythmia64. The relationship between hypertension and arrhythmia with PC warrants further investigation.

In Japan, the five-year survival rate for pancreatic cancer in all patients is 8.5%; however, for surgical cases, it is 28.6%1. In our cohort of diabetic patients with pancreatic cancer, the five-year survival rate is 21.6% (Fig. 3a). Regular hospital visits for diabetes treatment may lead to earlier detection and surgical intervention in symptomatic pancreatic cancer. Despite this, the survival rate remains suboptimal, underscoring the necessity for innovative therapeutic strategies to enhance survival rates.

There have been reports in the literature on the relationship between diabetes medication and the risk of pancreatic cancer. Metformin, in particular, is well-known, with studies suggesting it may reduce the incidence of pancreatic cancer10,65. This study is an explorative investigation into potential predictors of pancreatic cancer incidence, rather than focusing on causal relationships45,46,47. Additionally, the New-user Design-based analysis is impractical with the current dataset, leading to a potential misinterpretation of results. Consequently, the analysis excludes consideration of various medications.

This study is not without its limitations. First, we could not gather data on family history, medication adherence, socioeconomic status, cancer stage, pathological findings, or genetic information. Notably, a family history of pancreatic cancer is a known risk factor for the disease4, but this aspect was not included in our analysis. Second, the identification of pancreatic cancer was based on ICD-10 codes, which precluded the confirmation of additional details, including cancer stage. Similarly, pancreatic cystic lesions and chronic pancreatitis were cited as risk factors for pancreatic cancer; however, details about these diseases, such as the size of the pancreatic cysts and the treatment of chronic pancreatitis, were not obtained. Therefore, the impact of disease severity on pancreatic cancer risk has not been fully evaluated. Furthermore, Due to the characteristics of pancreatic tumors, some cases are not pathologically evaluated and may be registered as pancreatic cancer under the Japanese insurance system. This can result in lower malignancy tumors being classified as pancreatic cancer, potentially affecting the observed frequency and prognosis trends in this study. Third, as the age of disease onset could not be accurately ascertained, the age at diagnosis was used as a proxy. Fourth, we defined diabetic patients based on the prescription of one or more antidiabetic agents, leaving a gap in data for diabetic patients who do not use these medications. Furthermore, due to the nature of the administrative claims data used, the duration of diabetes mellitus could not be accurately determined. Therefore, it was not possible to provide the incidence of pancreatic cancer classified by the duration of diabetes. However, we did present the incidence rates based on whether diabetes was newly diagnosed or pre-existing, and found no significant differences in the impact on pancreatic cancer onset between these groups (Table 1, p < 0.875)". Fifth, an important limitation arises from the demographic composition of the database. Derived from Japan’s National Health Insurance system, our dataset primarily comprises an older population. This demographic distribution may contribute to the high incidence of pancreatic cancer in this study and limits the representativeness of our data for younger pancreatic cancer patients. However, considering the higher prevalence of pancreatic cancer in the elderly, conducting a risk factor analysis with this dataset, which includes a significant elderly demographic, is still valuable. Finally, the study confirmed that enrollees with newly diagnosed pancreatic cancer did not have pancreatic cancer during the one-year covariate assessment window. However, if a patient who already had pancreatic cancer did not use insurance for pancreatic cancer treatment during the baseline period, he or she may have been treated as a new-onset patient. Despite these limitations, this study provides important insights into the association between pancreatic cancer and various risk factors in a large cohort of diabetic patients.

Conclusions

This cohort study has identified key risk factors for pancreatic cancer in diabetic patients. Findings indicate that advanced age, male sex, a history of liver disease, chronic pancreatitis, and pancreatic cystic lesions significantly increase the risk of pancreatic cancer in this population. These insights underscore the importance of targeted surveillance for early detection of pancreatic cancer in diabetic patients, particularly those with these risk factors. This study contributes to the understanding of pancreatic cancer in diabetics, offering a foundation for improved patient management and screening protocols.