The worldwide prevalence of diabetes has dramatically increased owing to the westernization of diet and sedentary lifestyle, and the increased incidence of diabetes mellitus is more pronounced in developing countries, including Asian countries [1-4]. This phenomenon has also been observed in epidemiological studies in Koreans [5,6]. Furthermore, an increase in the impaired fasting glucose owing to obesity and aging of the populations will accelerate this phenomenon in the foreseeable future. Many previous studies have demonstrated an increased incidence of and mortality from cancer in subjects with diabetes [7-11]. Increased endogenous insulin levels resulting from insulin resistance, hyperglycemia, chronic inflammation, and increased oxidative stress have been suggested to be the main mechanism responsible for increased cancer risk in patients with diabetes [10,12]. Given the increased risk for diabetes and cancer in obese patients, obesity is thought to play a role not only in the onset of diabetes but also in the development of cancer in patients with diabetes [12-16]. Because cancer develops over a long period, it is difficult to judge whether there is a simple association or a direct causal relationship between increased cancer incidence and diabetes [10]. De Bruijn et al. [17] suggested that detection bias based on findings, such as the highest incidence of cancer in the first 3 months after diagnosis and the gradual decrease afterward, results in a negligible difference in the risk between patients with and without diabetes. Therefore, it is necessary to determine whether the association between diabetes and cancer is causally related or whether there is detection bias or reverse causation by studying the temporal relationship between diabetes and cancer risk through population-based cohort studies [18].

Many of the existing studies have been conducted on Western populations with different genetic and environmental backgrounds from that of Koreans, and the results cannot be applied directly to Koreans. For example, the risk for prostate cancer in patients with diabetes is reduced in Western populations, while in Asian populations, including Koreans, it is known to increase [16,19,20]. This discrepancy implies that genetic and/or environmental factors should always be considered in the interpretation of the results of these studies [21]. The Korean National Health Insurance Service (KNHIS) system (i.e., program, structure, medical coverage, types of insurance benefits, reimbursement flow, and healthcare delivery system) and the population and contents of the KNHIS database have been reported previously [22]. Briefly, the Korean National Health Insurance system run by government agencies like the NHIS and the Health Insurance Review Agency covers the entire population as a social insurance benefits scheme [22]. The NHIS database contains basic data such as age, sex, living area, and income level, as well as information about diagnosis statements as determined by the International Classification of Diseases 10th revision (ICD-10), the detailed statement of prescriptions, in-hospital administration, and surgery. Because the KNHIS database provides data on prevalence or incidence of disease, comorbidity, behavioral patterns of health care utilization, medical cost, and mortality of the entire Korean population, it can be used for population-based, nationwide studies of various diseases.

In this study, we aimed to investigate site-specific cancer risk of Korean patients aged ≥ 30 years with diabetes and to estimate causal and temporal relationships by analyzing organ-specific cancer risk according to the duration of diabetes.

We found that the incidence of cancer was higher in patients with diabetes than in those without diabetes, and this phenomenon was noticeable within 6 months of the diagnosis of diabetes. The risks for cancer of the stomach, colorectum, liver, and pancreas remained higher in both sexes in the diabetes group than in the non-diabetes group for the entire duration of diabetes.

The risk for cancer in patients with diabetes varies according to the duration of the disease, location of cancer, and sex of the patient. In particular, the risk for pancreatic cancer was notably higher in patients with diabetes shortly after diagnosis (duration < 6 months; 5.94-fold) than in those without diabetes; however, it rapidly decreased over time. Nevertheless, the fact that a 1.64-fold increase in risk persisted for ≥ 3 years suggests that diabetes is an important risk factor for pancreatic cancer, despite detection bias in the early period. A similar tendency was observed in cancers involving the colorectum, prostate, and endometrium. During the entire duration of diabetes, our results were consistent with those of previous studies [11,12,18,25-27], showing increased incidences of pancreatic, liver, colorectal, prostate, and endometrial cancer in patients with diabetes. It is possible that reverse causation and detection bias may have played a role in the seemingly increased incidence of cancer immediately after the diagnosis of diabetes; however, because the risk remained high after 6 months duration of diabetes, we can interpret the result as a causal relationship.

Additionally, the increase in cancer incidence of the stomach, colorectum, liver, pancreas, prostate, and endometrium was statistically significant even after the adjustment for age, sex, underlying disease, and socioeconomic status, which supports our claim that there is a causal relationship between diabetes and cancer in certain organs. Although the risk for larynx, lung, kidney, thyroid, and bladder cancer was significantly increased in both sexes even after adjusting for sex, underlying disease, and socioeconomic status, an analysis based on the duration of the disease yielded different results: a significant increase in risk was observed in all duration groups for men, whereas a varying significance was observed in women depending on the duration, implying a difference between the sexes. The difference in the risk for larynx and lung cancer in Korea between men and women may be related to the low smoking rate of women.

Numerous debates are ongoing regarding whether the relationship between diabetes and prostate cancer is positive or negative or whether the relationship exists at all [28]. One of the prominent features of the relationship between diabetes and prostate cancer is that it appears to differ among ethnicities. In prospective studies, retrospective studies, and meta-analyses that targeted mainly Caucasians, the prostate cancer incidence was lower in patients with diabetes than in those without diabetes, whereas studies involving the Asian populations showed a significant increase in prostate cancer risk in patients with diabetes [19,20]. Our findings are consistent with previous studies performed in the Asian populations, with a significantly increased risk for prostate cancer in patients with diabetes regardless of the duration of the disease. Statistical significance was retained even after adjustments were made for age, underlying disease, and socioeconomic status. This result supports the discrepancy in the interaction between risk factors for prostate cancer in patients with diabetes and carcinogenic environmental factors associated with diabetes among the populations owing to genetic background.

For oropharyngeal and esophageal cancer, the incidence risk increased for the first 6 months after diagnosing diabetes; however, after 3 years, the risk was lower in the diabetes group than in the non-diabetes group. This phenomenon was more prominent in women than in men. Additional long-term studies would be required to address the question of whether the duration of diabetes is negatively related to the incidence risk for these two cancers. However, detection bias may cause cancer incidence to appear to be increased in patients with diabetes. Breast and ovarian cancer showed similar results. The incidence risk significantly increased for the first 6 months after diagnosis, after which the difference became negligible, further supporting the existence of detection bias.

The strength of this study is that it is a large-scale epidemiological study involving the entire Korean population aged ≥ 30 years. Thus far, this is the first study regarding diabetes and cancer incidence risk that encompasses > 25 million subjects. This strength enables our study to provide comprehensive information on cancer risk in Korean patients with diabetes and to be used as important evidence in establishing guidelines for public health policies. Our study had some foreseeable drawbacks. First, because diabetes is defined in this study using certain disease codes plus the use of anti-diabetic medications, patients with initial-stage diabetes who are on a controlled diet and exercise plan but not using medications may have been excluded. However, a previous study [29] showed that this proportion of patients is very low (1.1%), and because the results indicated a high incidence of cancer in the first 6 months after the diagnosis of diabetes, the limitation does not appear to have a profound effect on the findings of this study. Second, patients who were diagnosed with diabetes after January 1, 2008 were included in the non-diabetes group for analysis. However, because cancer incidence in patients with diabetes was the highest shortly after diagnosis and the time interval between January 2008 and December 2013 was relatively small, these factors may have caused the early stage increases in incidence to be more prominent, without further influencing the interpretation. Finally, as noted in a previous study [30], problems regarding the accuracy of diagnosis and the failure to address adjustments for risk factors related to each specific cancer owing to limited clinical and laboratory information remain. For example, cancer cases attributable to both diabetes and high body mass index were almost twice as frequent in women (especially breast and endometrial cancer) than in men (especially liver and colorectal cancer) in the independent scenario [31]. Because patients with diabetes tend to be obese, a limited portion of the increased cancer incidence in these patients can be attributed to obesity. However, we believe this factor needs to be allowed as a limitation of epidemiological studies using health insurance claim data [24,32].

In conclusion, the risk for cancer increases in patients with diabetes, and the phenomenon is more prominent over short periods, i.e., in the first 6 months after diagnosis. As the duration of diabetes increases, incidence risk for cancer varies depending on the site of cancer and the patient’s sex. Thus, in patients with diabetes, cancer screening should be individualized based on the duration of diabetes, sex, and the location of cancer.