Author Correction: Helicobacter pylori infection, atrophic gastritis, and risk of pancreatic cancer: A population-based cohort study in a large Japanese population: the JPHC Study

An amendment to this paper has been published and can be accessed via a link at the top of the paper.


Materials and Methods
study population. The study was conducted using the Japan Public Health Center-based Prospective Study (JPHC Study) Cohort II. This cohort was launched in 1993-1994 including with 78,825 Japanese residents (38,740 men and 40,085 women) aged 40-69 years at the beginning of the baseline survey from 6 public health center areas all over Japan 21 . Details of the study design have been described elsewhere 21 . The study protocol was approved by the institutional review board of the National Cancer Center, Japan (Approval Number: 2001-021) and The University of Tokyo (approval number: 10508). All methods used in this study were performed by the relevant guidelines and regulations.
Baseline survey. A self-administered questionnaire regarding lifestyle factors was completed at the baseline of the cohort. The participants were informed of the objectives of the study, and those who completed the survey questionnaire were regarded as consenting to participate in the study. Figure 1 shows the study particpants selection process. Of 78,825 participants at the baseline, participants from one public health center area (Suita, n = 9,747) were excluded due to the unavailability of complete cancer data. We excluded foreign nationals (n = 22), move out of the study area before the study starting point (n = 82), missing age (n = 1), duplicates (n = 4), or those with inadequate follow-up data (n = 81). We further excluded individuals who had died, moved out of the study area, or had an unknown date of diagnosis before the starting point (n = 6,969), and non-respondents to the baseline questionnaire (n = 5,706). Among those who responded to the baseline questionnaire, 38% (n = 21,329) voluntarily provided 10 mL of blood during health checkups provided by their local government. Samples were divided into four tubes for plasma and buffy layer and stored at −80 °C until analysis. Subjects who reported a history of any cancer (n = 1,213) were excluded from the study, leaving 20,116 individuals (7,316 men and 12,800 women) for the analysis.
Laboratory analysis. H. pylori infection and AG were defined using biomarkers H. pylori seropositivity (anti-H. pylori) and pepsinogen (PG) I and II respectively. Levels of Immunoglobin G (IgG) were measured using an enzyme immunoassay (E Plate "Eiken" H. pylori Antibody II; Eiken Kagaku, Tokyo, Japan) 22 . An IgG titer of anti-H. pylori ≥10 U/mL was considered H. pylori seropositivy 22 . The latex agglutination technique was used to determine the plasma levels of PG I and II (LZ test "Eiken" Pepsinogen I, II; Eiken Kagaku, Tokyo, Japan) 22 . A subject was defined as AG-positive if PG I ≤70 ng/mL, and PG I/II ratio ≤3.0 22 . Miki conducted a meta-analysis in 2006 23 for the sensitivity and specificity of the serum pepsinogen test method using results from 42 studies. The combined sensitivity and false-positive rates for PGI ≤70 ng/mL and PG I:II ratio ≤3.0 were 77% and 27%, respectively 23 . The positive predicted value ranged from 0.77% to 1.25%, while the negative predictive value varied from 99.08% and 99.90% 23 . CagA seropositivity was not examined due to limited availability of stored blood samples. Using these criteria, the study subjects were further divided into three groups according to a combination of AG status and anti-H. pylori; AG-negative and H. pylori seronegative (AG−/ anti-H. pylori−); AG-negative and H. pylori seropositive (AG−/anti-H. pylori+); AG-positive and H. pylori seropositive (AG+/anti-H. pylori+).
Follow-up and identification of pancreatic cancer cases. Subjects were followed from the date of the baseline survey until December 31 st , 2010. Residence and survival status of the subjects was confirmed through the residential registry. The incidence of pancreatic cancer was identified through active patient notification from major local hospitals in each study area and linking with population-based cancer registries. Death certificates were used to supplement the information on cancer incidence. Cases of pancreatic cancer were classified using the International Classification of Diseases for Oncology, 3 rd edition, code C25 24 . statistical analysis. Study participants were censored on the date of pancreatic cancer diagnosis, move-out from the study area, death, or December 31 st , 2010, whichever came first. Subjects' characteristics at the baseline were compared independently by anti-H. pylori and AG status, and also by a combination of the two biomarkers. Differences in baseline characteristics between anti-H. pylori and AG status were analyzed using analysis of variance or χ 2 -test. Cox proportional hazards regression models were used to estimate the hazard ratios (HRs) and their 95% CI using attained age as the time scale due to the strong association between pancreatic cancer risk and age. Participants who tested negative to the biomarkers at the baseline of the study were used as the reference group. Covariates included were based in associations found in previous studies 25 . Model 1 adjusted for public health center areas (six areas treated as strata) and gender, while model 2 further adjusted for body-mass index (BMI) calculated using measured height and weight (<25.0 kg/m 2 , ≥25.0-<27.0 kg/m 2 , ≥27.0 kg/m 2 ), self-reported history of diabetes (yes, no), self-reported physical activity (continuous, metabolic equivalent of task (METs)), self-reported alcohol consumption (never and formal, occasional, <150 g/week, ≥150 g/week), self-reported family history of pancreatic cancer (yes, no), and self-reported smoking status (never, former, current), in addition to the confounders in model 1. We conducted individual stratified analyses for each pancreatic cancer risk factors to see how H. pylori seropositivity and AG status affected the pancreatic cancer risk differently according to lifestyle habits. Interactions were considered between H. pylori infection, AG status and each of the covariates included in the analysis by running a regression model with an interaction term, then conducting a Wald test for interaction. All analyses were conducted using Stata version 13.0 (StataCorp LP).

Results
During 320,470 person-years of follow-up (mean 16 years), 119 cases (52 men and 67 women) of newly diagnosed pancreatic cancer were identified among 20,116 subjects. Table 1 shows the baseline characteristics of cohort participants by H. pylori seropositivity and AG categories. We combined never-and past-alcohol drinkers due to a small number of participants in past drinker category (n = 383, 1.9% of the total participants). Men represented 36% of the total participants. AG−/anti-H. pylori+ category had the highest percentage of men (39.1%), habitual drinkers (≥150 g of ethanol per week, 15.5%), and current smokers (18.3%). In all categories, the majority of the subjects were within normal BMI range (<25 kg/m 2 ). At the baseline, 13,752 subjects (68%) and 8,470 (42%) were found to be anti-H. pylori seropositive and AG positive, respectively. Mean PGI value for our study was 53.4 ± 29.0 ng/mL; minimum PGI value observed was 2 ng/mL and the maximum PGI value was 606.4 ng/mL. Table 2 shows the HR and 95% CI for anti-H. pylori and AG status and the risk of pancreatic cancer, using attained age as a time scale. Those who tested negative to each agent were used as a reference. AG is considered the endpoint of chronic gastritis caused by H. pylori infection 26 ; therefore, AG+/anti-H. pylori− category was combined with AG+/anti-H. pylori+ category.
No statistical association was found between AG and anti-H. pylori status and pancreatic cancer risk, even when the model was adjusted for potential confounders. A borderline decrease in risk was observed among AG−/H. pylori+ subjects (HR 0.57, 95% CI 0.31-1.03). Table 3 shows the association between H. pylori infection and the risk of pancreatic cancer by individually stratifying for smoking status, alcohol consumption, BMI, and history of diabetes. Among current smokers, we saw a statistically significant increase in the risk of pancreatic cancer for AG+ (HR 3.64, 95% CI 1.37-9.66) and AG+/anti-H. pylori− or AG+/anti-H. pylori+ subjects (HR 5.21, 95% CI 1.14-23.87). There were no statistically significant interactions between H. pylori infection, AG status, smoking status, alcohol consumption, BMI and history of diabetes.

Discussion
To best of our knowledge, this is the first prospective cohort study to investigate H. pylori infection status and the risk of pancreatic cancer incidence in a Japanese population. When stratified by smoking status, the risk of pancreatic cancer among current smokers with AG statistically increased, regardless of H. pylori infection status.
We observed a non-statistically significant decrease in the risk of pancreatic cancer for AG−/anti-H. pylori+ subjects. Although the exact mechanism of how H. pylori seropositivity lowers pancreatic cancer risk is unclear, one hypothesis proposed suggests suppression of appetite by H. pylori infection leading to a reduction of ghrelin, ultimately lowering body weight, reducing cases of pancreatic cancer caused by obesity 27 .
Another possibility is the infection by CagA seropositive H. pylori strains may be working as a protective factor for pancreatic cancer development. Risch et al. 16 showed a statistically significant risk decrease for www.nature.com/scientificreports www.nature.com/scientificreports/ pancreatic cancer among H. pylori and CagA seropositive in a Chinese population (OR 0.66, 95% CI 0.53-0.81). A meta-analysis conducted based on 2,049 cases and 2,861 controls also showed a reduction in pancreatic cancer risk among those who are infected with H. pylori (summary OR 0.62, 95% CI 0.49-0.76) and CagA seropositive (summary OR 0.66, 95% CI 0.52-0.80) in Asian population 28 . Evidence from mouse models have suggested SHP-2, a tyrosine phosphate expressed in most embryonic and adult tissues and an intracellular target of H. pylori www.nature.com/scientificreports www.nature.com/scientificreports/ CagA protein, may be regulating glucose and lipid metabolism by suppressing insulin signalling in hepatocytes 29 , suppressing tumour proliferation in the pancreas. Eradication of CagA seronegative H. pylori among patients with duodenal ulcer returns their hyperchlorhydria to normal 30,31 ; in contrast, eradication of CagA seropositive H. pylori among corpus AG patients returns the stomach environment to hypo-or achlorhydria to normal 32 . Since gastric acidity promotes secretion of bicarbonate and fluid from pancreatic ductal cells 33 , it allows CagA seronegative H. pylori to survive in pancreatic ductular epithelium 16 . An animal model has shown that excess production of pancreatic bicarbonate and fluid increased the ductular cell dysplasia and adenocarcinoma 34 . This suggests the possibility of difference in gastritis acidity caused by H. pylori stains may affect in the pancreatic cancer risk 16 . However, because we could not measure CagA seropositivity due to limited availability of stored blood samples, we were not able to test the hypothesis.
Stratified analysis by established pancreatic cancer risk factors showed a statistically significant increase in the pancreatic cancer risk among current smokers with AG. This result is inconsistent with a previous Finnish cohort study that recruited current male smokers to look at the association between AG and pancreatic cancer and found no association, whether AG was diagnosed serologically or histologically 19 . Serum PG levels are known to be affected depending on demographic characteristics such as gender, age, smoking, alcohol consumption, and dietary habits 35 , leading to various cut-off values of serum PG depending on populations. In our study, we used the threshold for defining a high-risk population in Japan 36 (PGI ≤ 70 ng/mL and a ratio of PGI and PGII ≤ 3), proposed by Miki et al. 23,37,38 , to define AG. In European nations, the often used serum PGI cut-off value is ≤25 ng/ mL, with a ratio of PGI and PGII < 3 39 . It is possible that the differences in the findings between countries are due to the difference in PGI cut-offs used to define AG.
No epidemiological evidence has been found to clarify how AG is associated with pancreatic cancer risk. Truan et al. 40 found pepsinogen expression in 38% pancreatic cancer cases, while another study found that gastrin, a gastrointestinal peptide, had a proliferative effect on pancreatic cancer cells 41    www.nature.com/scientificreports www.nature.com/scientificreports/ spreading in the presence of chronic AG [45][46][47] , resulting in a negative result in the antibody test 48 . IgG anti-H. pylori seronegative status among those who are AG positive implies prior H. pylori infection, since H. pylori cannot survive in the atrophic or intestinal metaplasia mucosa 49 . Previous studies have shown those with advanced gastritis with loss of H. pylori are at a higher risk of developing gastric cancer 48,49 . This may be the reason why the increased risk for pancreatic cancer was observed among AG positive but H. pylori negative individuals in a previous study 20 . AG therefore may have a causative effect on pancreatic cancer development, through the extra-gastric or systemic effect of H. pylori 19 .
A major strength of this study is its prospective cohort design, with subjects recruited from a large sample of the general population. The high response rate and low loss to follow-up increased the generalizability of the conclusions and reduced selection bias. We used the incidence of pancreatic cancer as an endpoint rather than death since it directly measured the risk of pancreatic cancer. Finally, because the H. pylori infection rate in the Japanese population is high for those born prior to 1950 (over 70%) 50 , the country provides a suitable setting for examining the association of H. pylori, AG, and pancreatic cancer.
There are several limitations. The cases of pancreatic cancer found during the follow up period were relatively small (n = 119). However, we believe that with an average of 16 years of follow-up, sufficient numbers of pancreatic cancer cases were identified. Furthermore, because the development of pancreatic cancer is a rare event (Age-standardized rate per 100,000: 9.7) 51 , the incidence observed during this study period is believed to be acceptable. We also had to exclude 34,965 subjects from the analysis since they did not provide a blood sample, and therefore their H. pylori and AG status could not be observed. Men comprised 36% of the study population, which may have led to a gender-biased result. H. pylori infection status was defined using serum antibody-based tests, which are relatively inexpensive, rapidly performed, and cause minimal discomfort to the subjects 52 . However, one of the limitations of using serology test is its inability to differentiate between past and present H. pylori infections 53 . Although H. pylori infection is said to be associated with low socioeconomic status (SES) 54 , due to the unavailability of the data, we were not able to include SES into our analysis. This study also did not consider blood group antigens, a potential risk factor for pancreatic cancer incidence reported by several studies including a study conducted by Wolpin et al. 55 , which reported compared to those with blood type O, blood type A and B have a significantly increased risk for pancreatic cancer. Finally, the association may have been confounded by additional unmeasured or unknown risk factors.
In summary, H. pylori infection and AG was not associated with the risk of pancreatic cancer in a general Japanese population, individually or in combination; however, among current smokers with AG, a significant increase in the risk of pancreatic cancer was observed. Further investigation in larger cohorts, especially in Asian countries, will provide a more comprehensive evaluation.