The JPHC Study was launched in 1990 and 1993 for cohorts I and II, respectively . Participants in cohort I included residents aged 40–59 years from five Japanese public health center areas (Iwate, Akita, Nagano, Okinawa-Chubu, and Tokyo), and those in cohort II included residents aged 40–69 years from six other Japanese public health center areas (Ibaraki, Niigata, Kochi, Nagasaki, Okinawa-Miyako, and Osaka). A baseline survey questionnaire was distributed to 140,420 registered residents mostly by hand. Participants were informed of the objectives of the study, and that completion of the survey questionnaire was regarded as providing consent to participate. The 5- and 10-year follow-up surveys (second and third surveys, respectively) were conducted to update information on lifestyle habits and health conditions in 1995–1998 and 2000–2003, respectively. The present study used the second survey as the baseline.
A total of 136,163 subjects were eligible for participation in this study. Of those, 102,341 participants (75.2%) who provided valid responses formed the study cohort. We excluded 5116 participants who reported extreme total energy intake (sex-specific values outside of ± 2.5%) and 4256 participants who reported a history of disease, including cancer, stroke, myocardial infarction, and chronic liver disease. Thus 92,969 participants (42,700 men and 50,269 women) were analyzed (Fig. 1).
Exposure (8-item Japanese Diet Index score)
A semi-quantitative food frequency questionnaire (FFQ) was used to assess the average intake of 147 food and beverage items in the second survey [6, 7]. For most food items, participants were asked about consumption frequency and usual portion size. The validity and reproducibility of the FFQ have been established in previous studies [8,9,10].
Based on previous studies [2, 4], we used the JDI8 to assess the degree of adherence to the Japanese diet. The JDI8 consists of eight components: rice, miso soup, seaweeds, pickles, green and yellow vegetables (green vegetables, carrot, pumpkin, and tomato), fish (raw fish, salty fish, dried fish, seafood, canned tuna, and fish products), green tea, and beef and pork (beef, pork, and processed meat). The first seven components represent adherence to the Japanese diet, and participants received one point if their intake was more than or equal to the sex-specific median. The 8th component represents non-adherence to the Japanese diet, and participants received one point if their intake was less than the sex-specific median. Supplemental Table 1 shows the sex-specific median of intake of the JDI8 components. The JDI8 score ranged from 0 to 8, with higher scores indicating greater dietary conformity. Incidentally, the JDI8 was modified to exclude coffee from the original JDI, because lower coffee consumption was significantly associated with an increased risk of all-cause mortality . Additionally, previous studies using a meta-analysis approach reported that coffee consumption was inversely associated with all-cause and CVD mortality .
The participants’ residency and vital status were followed up from baseline to December 31, 2016, using the residential registry. Causes of death were confirmed by death certificates and were defined according to International Classification of Diseases, 10th revision (ICD-10). The major endpoints of this study were mortality from all-cause, cancer (ICD-10: C00–C97) and CVD (ICD-10: I00–I99). CVD mortality was subdivided into mortality from heart disease (ICD-10: I20–I52) and cerebrovascular disease (ICD-10: I60–I69).
This study was approved by the Institutional Review Board of the National Cancer Center of Japan (approval code; 2015-085) and the Ethics Committee of Tohoku University Graduate School of Medicine (approval code; 2018-1-321).
We counted the person-years of follow-up for each subject from the date of response to the second survey questionnaire until the date of death or the end of the study period (December 31, 2016), whichever occurred first.
The adjusted Cox proportional hazards model was used to calculate the hazard ratios (HRs) and 95% confidence intervals (CIs) for all-cause and cause-specific mortality according to the groups of the JDI8 score (G1, score of 0–2; G2, score of 3; G3, score of 4–5; and G4, score of 6–8). The lowest group was used as the reference category. Multivariable-adjusted models were adjusted for the following variables. Model 1 was adjusted for age (45–49, 50–54, 55–59, 60–64, 65–69, or ≥ 70 years), sex, and study area (11 areas). To examine whether the association between adherence to the Japanese diet and mortality was attributable to physical health status or other lifestyle factors, Model 2 was further adjusted for body mass index (< 18.5, 18.5–24.9, 25–29.9, ≥ 30 kg/m2, or missing), smoking status (current, former, never, or missing), alcohol drinking (< 1 time/month, 1–3 times/month, 1–2, 3–4, 5–6 times/week, every day, or missing), total physical activity (quartile of metabolic equivalent task-hours/day), medication (antihypertensive, cholesterol-lowering, or hypoglycemic agents [yes or no for each item]), and occupation (agriculture, forestry, fishery, office work, self-employed, specialty work, housewife, unemployed, or other [yes or no for each item]). In addition to these adjustments, Model 3 was further adjusted for total energy intake (in kcal/d; sex-specific quartile categories). To test for linear trends, categories indicating an increase in the JDI8 score (scored as 1 for G1, 2 for G2, 3 for G3, and 4 for G4), and variables using the JDI8 score as a continuous variable were entered in the corresponding Cox model. The proportional hazards assumption was checked based on the Schoenfeld residual test using Stata command "estat phtest". In this analysis, the proportional hazards assumption was not rejected with the Schoenfeld residual test (P = 0.275).
We also conducted several sensitivity analyses to test the robustness of our findings. First, considering possible reverse causality, we conducted an analysis after exclusion of deaths occurring in the first 3 years of follow-up (n = 91,589). Second, we conducted analyses stratified by sex to assess whether effects of adherence to the Japanese diet on mortality varied by sex (male = 42,700, female = 50,269). Third, to minimize potential bias attributed to total energy intake, we calculated the JDI8 score using the data after adjustment by the residual method (n = 92,969). Fourth, for the same reason (i.e., to minimize potential bias attributed to total energy intake), we also analyzed the association including participants who reported extreme total energy intake (n = 97,802). Fifth, we performed an analysis that included sodium and soybean products (tofu, fermented soybean [natto], and soymilk, etc.) intakes in the covariates because it was reported that the Japanese diet was associated with high sodium intake [12, 13], and soybean products are characteristic of the Japanese diet  (n = 92,969). In addition, we investigated the relationship between each component of the JDI8 and all-cause and cause-specific mortality. Here, we conducted an analysis for each item by including it into the Cox model with full adjustment, respectively.
The Cox proportional hazards model was performed using SAS version 9.4 (SAS Inc., Cary, NC, USA). The proportional hazards assumption was performed by Stata/MP15 (StataCorp, Texas, USA). All statistical tests described here were two-sided, and differences at P < 0.05 were considered significant.