Risk Factors for Sporadic Pancreatic Neuroendocrine Tumors: A Case-Control Study

The current study examined risk factors for sporadic pancreatic neuroendocrine tumors (PNETs), including smoking, alcohol use, first-degree family history of any cancer (FHC), and diabetes in the Han Chinese ethnic group. In this clinic-based case-control analysis on 385 patients with sporadic PNETs and 614 age- and sex-matched controls, we interviewed subjects using a specific questionnaire on demographics and potential risk factors. An unconditional multivariable logistic regression analysis was used to estimate adjusted odds ratios (AORs). No significant differences were found between patients and controls in terms of demographic variables. Most of the patients with PNETs had well-differentiated PNETs (G1, 62.9%) and non-advanced European Neuroendocrine Tumor Society (ENETS) stage (stage I or II, 83.9%). Ever/heavy smoking, a history of diabetes and a first-degree FHC were independent risk factors for non-functional PNETs. Only heavy drinking was found to be an independent risk factor for functional PNETs (AOR = 1.87; 95% confidence interval [CI], 1.01–3.51). Ever/heavy smoking was also associated with advanced ENETS staging (stage III or IV) at the time of diagnosis. This study identified first-degree FHC, ever/heavy smoking, and diabetes as risk factors for non-functional PNETs, while heavy drinking as a risk factor for functional PNETs.

Scientific RepoRts | 6:36073 | DOI: 10.1038/srep36073 The aim of our study was to complete a large case-control study of sporadic PNETs in the Han Chinese ethnic group and evaluate information on a variety of potential exposures related to PNETs risk, including a history of diabetes, alcohol consumption, smoking habits, and a first-degree FHC. Because functional (F) and non-functional (NF)-PNETs show different clinical behavior and prognosis, we present results for F-and NF-PNETs combined and separately.

Results
Patient characteristics. Three hundred and eighty-five patients with PNETs were age and sex matched with 614 control subjects, with the ratio of 1:1.59. The distribution of demographic features of cases and controls is shown in Table 1. The mean age (± SD) was 49.7 (± 11.8) years for PNET patients and 48.5 (± 9.4) years for controls. Most of the patients and controls were female (55.1% and 55.2%, respectively), lived in urban areas (80.8% and 77.7%, respectively), and were educated up to middle or high school level (56.1% and 60.1%, respectively). No statistically significant differences were found between patients and controls in terms of these variables, suggesting that the frequency matching was adequate. Table 2 summarized the clinical features of the 385 PNET patients. There were 142 (36.9%) non-functional and 243 (63.1%) functional tumors. Most (58.4%) of the functional PNETs were insulinomas. Most of the patients had well-differentiated endocrine tumors (G1, 62.9%). With respect to ENETS staging, 217 patients (56.4%) were in stage I, 65 patients (16.9%) were in stage IIa, and 41 patients (10.6%) were in stage IIb. Twelve patients with tumor-invading adjacent structures were defined as stage IIIa. Lymph nodes were involved in 27 cases (23.6%) that were defined as stage IIIb. Twenty-three patients (6.0%) who had distant metastases at diagnosis were defined as stage IV. Table 3, unconditional logistic regression analysis was used to estimate risk associations between different factors and risk of total PNET (including functional and non-functional PNETs). The univariate analyses indicated that heavy alcohol consumption, ever/heavy smoking, and first-degree FHC (yes vs. no) were significant risk factors for PNET, whereas ever alcohol drinking, regions and educational levels were not significant factors. Multivariable analyses with adjustments for risk factors showed that ever/heavy cigarette smoking and first-degree FHC were independently associated with PNET risk, with multivariate AORs (95% CIs) of 1.60 (1.10-2.33) for ever smoking, 2.07 (1.15-3.73) for heavy smoking, and 1.60 (1.01-2.40) for first-degree FHC. Furthermore, heavy drinking (≥ 30 g/day) was not associated with higher risk of PNETs development in the multivariate model (AORs = 1.31; 95% CIs, 0.74-2.31).

Risk factors for F-NET.
Because functional and non-functional tumors have evidently different clinical behavior and outcome, we conducted analyses restricting to F-(n = 243) and NF-PNET (n = 142), respectively. Univariate analysis indicated that regions, educational levels, ever alcohol drinking and first-degree FHC were not significant risk factors for the development of F-NET, whereas heavy alcohol consumption (≥ 30 g/day) and ever/ heavy smoking were significant risk factors for F-PNETs. Multivariable analyses showed that only heavy alcohol use was independently associated with F-PNET risk, with multivariate AORs (95% CIs) of 1.87 (1.01-3.51; Table 3).

Risk factors for NF-PNET.
In the analysis of NF-PNET, in addition to the variables of regions, educational levels, alcohol consumption, smoking status, and first-degree FHC, we also included a history of diabetes as a potential risk factor. Multivariable analyses with adjustments for risk factors showed that ever/heavy smoking,  Table 4, we compared PNET patients with a well WHO classification at the time of diagnosis (G1, n = 242) and those with a poor or moderate classification (G2 + G3, n = 111). None of the risk factors (regions, educational levels, ever or heavy alcohol consumption, ever or heavy smoking, first degree FHC) were associated with the WHO classification. In the analysis restricting to F-and NF-PNET, respectively, we found that none of the risk factors were associated with the WHO classification (P > 0.05; Table 4). We then assessed whether any of the identified risk factors were correlated with a more advanced ENETS staging (TNM stage III or IV). The following risk factors-regions, educational levels, alcohol consumption (ever or heavy), first degree FHC, and a history of diabetes-were not associated with ENETS staging (P > 0.05). Interestingly, ever (P = 0.014) and heavy smokers (P = 0.003) were more likely to be diagnosed with advanced ENETS staging than never smokers. Furthermore, in the analysis restricting to NF-PNET, we found that ever smokers (P = 0.035) and heavy smokers (P = 0.002) were more likely to be diagnosed as having advanced ENETS staging than never smokers. When performing the analysis restricting to F-PNET, none of the risk factors were associated with the advanced ENETS staging (TNM stage III or IV).

Risk factors and WHO classification and ENETS stage. As shown in
Additionally, we evaluated whether the effects of cigarette smoking on the ENETS staging were independent of regions and educational levels (Supplementary Table 1). For NF-PNET, neither regions nor educational levels were associated with ENETS staging in ever/heavy smokers (P > 0.05). Similar results were also shown for total PNET.

Discussion
Unlike the studies evaluating risk factors associated with exocrine pancreatic carcinomas, risk factors to date have not been systematically identified for PNETs. In this large hospital-based case-control study, we found independent associations between ever/heavy smoking, first-degree FHC, a history of diabetes and the risk of NF-PNETs. However, only heavy drinking was indicated to be independently associated with the development of F-PNETs. Interestingly, ever/heavy smoking was associated with advanced ENETS staging in NF-PNETs.
Although smoking is clearly one of the most preventable causes of pancreatic carcinoma development 18,19 , little is known about the role of smoking in the development of PNET. Several recently conducted case-control studies showed no positive association between ever smoking and the development of NETs in the pancreas 14 , rectum 20 and small intestine 21 . In another research from Italy found that although heavy smoking was associated with a slightly increased risk of PNET (OR = 1.5; 95% CI: 1-2.4) in the univariate analysis, neither smoking nor heavy smoking was associated with an increased risk in the multivariate analysis. Results of our analyses indicated that ever/ heavy smoking were related to an elevated risk of NF-PNETs in the multivariate analysis (ever smoking: OR = 1.52, 95%CI: 1.01-2.39, P = 0.046; heavy smoking: OR = 1.86, 95%CI: 1.23-3.43, P = 0.018). However, we did not find a significant association between ever/heavy smoking and F-PNET in the multivariable models, although univariable  Continued models indicated a significant association. Our research suggests that the tumorogenesis of NF-PNET was different from that of F-PNET. The effects of smoking on the risk of PNETs remain uncertain and merit further study. The association between diabetes and pancreatic carcinoma has been well examined 10,[22][23][24] . With respect to the association between diabetes and PNETs risk, several studies 14,15,17 including two meta-analyses 12,25 have consistently indicated diabetes as a potential risk factor for development of PNETs. Hassan et al. studied 160 patients with PNETs in their case-control study and observed that diabetes was associated with significantly increased risk of PNETs (OR = 2.8; 95% CI, 1.5-5.2) 14 . These results were supported by two independent studies from Capurso et al. 15 and Halfdanarson et al. 17 Similarly, our data provided strong evidence of an association between diabetes and the risk of NF-PNET, with an AOR of 1.96 (95% CI, 1.14-3.70). Because most functional tumors in this study were insulinomas (92.6%), which are scarcely diagnosed as diabetes 26 , we evaluated the role of diabetes only in the development of NF-PNETs. In addition, our study excluded all patients with an inherited syndrome (MEN-1 and VHL). One earlier study was conducted on cases diagnosed as either F-PNETs or NF-PNETs 15 , and two other studies did not describe the biological behavior of the tumors 14,17 . The mechanisms linking diabetes to PNETs development remain unknown. Several studies hypothesized that a family history of MEN-1 27 and the presence of a glucagon-producing tumor originating from the alpha cells of the pancreas 28 may result in elevated blood glucose levels. Furthermore, it is possible that diabetes may act as a mediator for chronic inflammation and oxidative stress inside the cell, which may lead to DNA mutation and the development of PNETs 29,30 .
Two previous case-control studies investigated the association between diabetes duration and PNET risk 14,15 . The effect estimate for subjects with recent onset (≤ 1 year) diabetes was higher (OR 12.80, 95%CI 2.47-66.42) than those with long-standing (> 1 years) diabetes 25 . In line with these results, our research indicated recent onset (≤ 1 year) diabetes (but not long duration of diabetes) was related to elevated risk of NF-PNET after adjustments for ever smoking and first degree FHC. This time-course characteristic strongly supported the hypothesis that DM might also be a consequence of NF-PNET, similar to the association between diabetes and pancreatic cancer 10,22 . The underlying mechanisms by which NF-PNET leads to DM might depend on the destruction of pancreatic beta cells and the development of peripheral insulin resistance 31 . Given the low rate of PNET, large multicenter studies would be necessary to explore the association between DM and NF-PNET.
In line with previous reports [14][15][16] , we observed the strong association between first-degree FHC and risk of NF-PNETs development, which was similar to the results for pancreatic carcinoma [32][33][34] . The increased risk of NF-PNET in subjects with a family history of cancer may be due to unknown genetic factors and shared environmental factors 13,35,36 . Several case-control studies have identified a possible role for apoptosis and inflammatory pathways in the etiology of NET, such as variants of the tumor necrosis factor alpha gene, interleukin 2 gene, and defender against cell death gene [35][36][37] .
Results from our study indicate no associations between ever/heavy alcohol drinking and NF-PNETs risk, which was comparable to the reports from Hassan et al. 14 , but was in contrast to the other reports 15,17 . We believe this inconsistency could be attributed to the limited size of the study sample, the different inclusion criteria for cases, and the different methods of quantification of alcohol consumption. In addition, we observed an independent association between heavy drinking and development of F-PNET, which was in line with the results from Zhan et al. 16 .
Only one study 15 presented data on the influence of risk factors on PNET patients' progression and outcome. In that report, the authors observed an association between history of diabetes and metastatic disease at the time of diagnosis (P = 0.012). No other factors were related to more aggressive disease features 15 . Interestingly, our data showed that in patients with NF-PNETs, an increased prevalence of advanced ENETS staging (stage III or IV) was associated with ever/heavy smoking (P < 0.05), but not with regions, educational levels, a history of diabetes, ever/heavy alcohol use and first degree FHC (P > 0.05). Tumors may be diagnosed at a more advanced stage in individuals who have a less favorable economic/cultural level or a "less healthy" behavior, because they may not urge to report symptoms early. In the present data, we further evaluated whether the effects of cigarette smoking on the ENETS staging were independent of regions and educational levels. Our data indicated that neither regions nor cultural levels were associated with ENETS staging in ever/heavy smokers (P > 0.05).
Research has shown genetic alterations in the lung epithelium of smokers, and increased microsatellite instability in colon tumors of smokers 38,39 . Cigarette smoke contained several carcinogens, which may reach the pancreas from the bloodstream and refluxed bile, suggesting the potential mechanism linking smoking to the development of pancreatic tumor 40 40 . However, mechanisms linking smoking to NF-PNET progression have not been explored, which should be examined in the future.
To the best of our knowledge, this case-control study is the largest to assess several risk factors for PNETs with proper adjustment for potential risk factors. The diagnosis was confirmed in each patient by individually reviewing pathology slides and reports to ensure that diagnostic inclusion criteria were met. Importantly, we explored the potential risk factors for F-PNET and NF-PNET, respectively, given the differences in clinical behavior and prognosis between the two disease entities.
We acknowledge that our study has certain limitations. First, the possibility of selection bias owing to its population (hospitalized patients with PNETs) and design (retrospective data reviewing) cannot be ruled out. Nevertheless, because both cases and controls belonged to a relatively homogeneous base population and were matched by sex, age, and sociodemographic variables, we believe that the bias would be minimized. In addition, given the rarity of PNETs and the need for confirmed pathologic diagnosis, our approach of retrospective data collection was appropriate.
Second, many potentially confounding factors could not be addressed owing to no established risk factors for PNETs. We specifically considered the association for risk factors in exocrine pancreatic cancer in the model but could not examine the influence of numerous host and/or environmental factors on risk, including chronic pancreatitis, allergies, BMI, H pylori infection, dietary factors, and commonly prescribed medications (use of statins, aspirin and hypoglycemic agents et al.). For example, the effect of obesity on the development of PNETs cannot be excluded, as obesity may be associated with type 2 diabetes development in patients with PNETs. Unfortunately, the patient records in our database contained no information about the patients' BMI before the diagnosis of PNETs. The baseline BMI may not have accurately reflected the patients' obesity history, because some PNETs patients experienced disease-related weight loss or weight gain. We did not consider the potential effects of commonly prescribed medications, such as use of statins, aspirin, which were reported to be inversely associated with risk of exocrine pancreatic cancer. Again, data on use of hypoglycemic agents (such as metformin, thiazolidinediones, insulin et al.) were not available in most of those diabetic individuals. Many researches, to date, have suggested that metformin and thiazolidinediones could exert a protective role against the development and progression of some cancers 42,43 , whereas insulin was associated with an increased risk of cancer 44,45 .   Third, the potential measurement errors could not be excluded when assessing risk factors. For instance, ever smokers/drinkers may include individuals with low level substance abuse who quit several years ago as well as patients who were heavy users and quit more recently. Furthermore, it was difficult to distinguish between type 1 and type 2 diabetes in most of our diabetic subjects. However, it was likely that the majority of diabetic individuals had type 2 diabetes because it is late onset and received treatments with only oral hypoglycemic agents.
Owing to the popularity of the endocrinology department at Ruijin Hospital of Shanghai Jiaotong University, patients with F-PNET were higher in our series than those in other studies (63.1% vs. 19.1%) 15 . We are continuing to establish a national consortium to assist in the development of a large multicenter epidemiologic study in China to examine several environmental, social, behavioral, occupational, and genetic risk factors and to assess gene-environment interactions in GEP-NETs 46 .
In summary, our study shows the different risk factors between F-and NF-PNET, suggesting different biological behavior and clinical characteristics between the two disease entities. Ever/heavy smoking and histories of diabetes and first-degree FHC may be potential risk factors for NF-PNETs, while heavy drinking may be one of the risk factors for F-PNETs. In addition, prediagnosis ever/heavy smoking may be associated with advanced ENETS staging (stage III or IV) in NF-PNETs. Although prospective studies are needed to validate these results, our preliminary findings may provide guidance in the development of PNETs surveillance programs in the future.

Material and Methods
Study design. The study design was an ongoing hospital-based case-control study conducted at Ruijin Hospital, Shanghai, People's Republic of China. The purpose of the study was to examine risk factors that contribute to the development of PNETs. The Ethics Committee of Ruijin Hospital approved the study protocol. The methods were carried out in accordance with the approved guidelines. After written informed consent was obtained, each participant was scheduled for an interview by using a structured questionnaire to collect demographic and exposure information.
Cases. Patients eligible for this study were enrolled between January 1, 2001 and June 30, 2015. There were 513 potential PNETs patients and 430 patients with pathologically confirmed primary PNETs during the study period. Of these, 45 cases were excluded, because 11 had a history of cancer, 15 missed recruitment, and 19 patients had a clinical diagnosis of inherited syndromes such as MEN-1, VHL syndrome, and neurofibromatosis type 1. The remaining 385 patients were enrolled in this study (Fig. 1).

Controls.
Subjects who were diagnosed with nonmalignant disease (including those with gallbladder polyps, polycystic kidney, breast fibroadenoma, uterine fibroids) based on discharge diagnoses in the same hospital during the same period were included as the controls. Eligible controls were age-(in 3-year age groups) and sex-matched inpatients, and underwent imaging tests and tumor marker tests (including CA19-9, CEA, AFP, etc.) to exclude potential asymptomatic common tumors. Patients with a history of malignant disease or having received any cytotoxic treatment were excluded. Conditions related to alcohol and tobacco consumption (e.g., respiratory diseases, peptic ulcer, and hepatic disease) or any chronic diseases (e.g., diabetes, cardiovascular disease) that might have resulted from substantial lifestyle modifications were excluded. Informed consent was obtained from all patients. After screening, we included 614 controls. Data collection. Cases and controls were personally interviewed for demographic characteristics (age, sex, educational level, and region); prediagnostic personal habits (smoking status and alcohol drinking); and histories of diabetes mellitus and first-degree FHC. Participants were classified as "ever-smokers" if they reported having smokers more than 100 cigarettes during their lifetime. Accordingly, never smokers were defined if they smoked less than 100 cigarettes during their lifetime. Smoking amount was recorded in terms of pack-years (pack-year = numbers of packs of cigarettes/day × years of smoking). Heavy smokers were classified if they had smoked for ≥ 21 pack-years, respectively 15 . Participants were classified as "ever-drinkers" if they had consumed > 1 serving/day (12.5 g/day) of alcoholic beverage (beer, wine or liquor) for a duration of at least 6 months 49 . For each beverage type, participants were asked to recall the number of drinks they typically consumed each week and the number of years during which they consumed that beverage. These answers were integrated into a scoring system that was used to classify alcohol consumption as "heavy drinking" (≥ 30 g/day) 50,51 . Diabetes was defined as present if the fasting serum glucose level was greater than 7.0 mmol/L or a previous diagnosis of diabetes mellitus was made based on the American Diabetes Association criteria 52 . The course of DM was calculated from the date of diagnosis of DM to the date of PNET diagnosis. As the previous published studies 14,15 , duration of DM was dichotomized at 1 year to define cases of DM as new-onset or long standing DM. For those diagnosed on admission, the course was recorded as less than 1 year. Scientific RepoRts | 6:36073 | DOI: 10.1038/srep36073 Statistical analysis. All statistical analyses were conducted using the SPSS 19.0 statistical software program (SPSS, Chicago, IL, USA). All tests were two-tailed, and a P value of < 0.05 was considered to indicate statistical significance. Pearson's χ 2 test (Fisher's exact test) was used to compare the sociodemographic and clinicopathologic data. Crude and adjusted OR and 95% CI for each variable were calculated by using unconditional logistic regression analysis. Potential confounders were included in the multivariate analysis in a stepwise manner at a significance level of P < 0.15. We divided the PNETs cases as functional and non-functional (NF) tumor, because both show different clinical behavior and prognosis. For functional PNETs, equations included terms for a first-degree FHC, smoking status, and alcohol drinking. For NF-PNETs, we also included a history of diabetes in addition to the above three variables.