ABSTRACT
Objective
We examined whether the association between adherence to the Japanese dietary pattern and all‐cause and cardiovascular disease (CVD) mortality is consistent across diverse residential topographical settings in Japan.
Methods
This prospective cohort study included 48 479 adults aged 40–79 years from the Japan Collaborative Cohort Study, with baseline dietary assessment in 1988–1990. Adherence was quantified by the Japanese Diet Index 8 (JDI‐8), with components dichotomized at sex‐specific medians. Residential municipalities were classified as coastal, urban inland, flat/intermediate inland, or mountainous inland. Shared frailty Cox proportional hazards models with a municipality‐level random intercept (33 municipalities) estimated hazard ratios (HRs) and 95% confidence intervals (CIs). We tested effect modification by residential area using interaction tests and stratified analyses.
Results
During a median follow‐up of 19.2 years, 9304 all‐cause deaths and 2685 CVD deaths occurred. Comparing the highest with the lowest JDI‐8 quartile, HRs were 0.929 (95% CI, 0.872–0.990; p for trend = 0.034) for all‐cause mortality and 0.862 (95% CI, 0.766–0.970; p for trend = 0.023) for CVD mortality. Interactions by residential area were not statistically significant (p for interaction = 0.246 for all‐cause and 0.185 for CVD mortality). Component‐level distributions varied by area, with urban inland residents having lower proportions of miso soup, rice, and fish intake.
Conclusions
Higher adherence to the Japanese dietary pattern was associated with lower all‐cause and CVD mortality, with directionally consistent associations across diverse residential settings. The absence of significant heterogeneity supports generalizability within Japan, although stratified estimates are exploratory.
Keywords: cardiovascular disease, dietary patterns, Japan Collaborative Cohort Study, Japanese diet, mortality
In a large prospective cohort of Japanese adults, greater adherence to the traditional Japanese dietary pattern was associated with lower all‐cause and cardiovascular mortality. This inverse association was consistent across diverse residential topographies, including urban, flat, mountainous, and coastal settings, regardless of the living environment.
Abbreviations
- BMI
body mass index
- CI
confidence interval
- CV
coefficient of variation
- CVD
cardiovascular disease
- DEM
digital elevation model
- DID
Densely Inhabited District
- FFQ
food‐frequency questionnaire
- HR
hazard ratio
- IQR
interquartile range
- JACC
Japan Collaborative Cohort
- JDI‐8
Japanese Diet Index 8
- MAFF
Ministry of Agriculture Forestry and Fisheries
- MICE
multiple imputation by chained equations
- Q
quartile
- SD
standard deviation
- SRTM
Shuttle Radar Topography Mission
- STROBE
Strengthening the Reporting of Observational Studies in Epidemiology
- STROBE‐nut
STROBE‐Nutritional Epidemiology extension
- TRI
terrain ruggedness index
1. Introduction
In aging societies, cardiovascular disease (CVD) and all‐cause mortality remain leading public health challenges. Diet is a modifiable lifestyle factor associated with both outcomes and is increasingly examined as an overall dietary pattern rather than as single foods. The traditional Japanese dietary pattern is characterized by frequent intake of rice, miso soup, vegetables, seaweed, fish, and green tea, with relatively low intake of beef and pork [1]. This pattern is low in saturated fat and rich in n‐3 polyunsaturated fatty acids, dietary fiber, and micronutrients [2]. Prospective cohort studies in Japan have reported lower risks of all‐cause mortality, CVD mortality, functional disability, and dementia with higher adherence to the Japanese dietary pattern [1, 3, 4]. A systematic review and meta‐analysis of prospective cohorts has further supported an inverse association between this pattern and CVD mortality [5].
However, whether this association is consistent across residential environments has not been adequately examined. Dietary habits are shaped by local food availability, food culture, socioeconomic background, and geographic conditions. National and regional nutrition surveys in Japan have documented within‐country differences in the intake of salt, vegetables, fruits, and other foods [6, 7]. Because it comprises a combination of foods, an equivalent adherence score may reflect different food compositions across regions. For example, fish products may contribute disproportionately in coastal areas, whereas preserved foods such as pickles may be more prominent in mountainous areas. Despite this, few studies have examined the health implications of the Japanese dietary pattern with explicit attention to such regional differences.
Examining these regional differences requires a large prospective cohort that captures diverse residential environments. The Japan Collaborative Cohort (JACC) Study is a nationwide prospective cohort with baseline assessment in 1988–1990 and includes residents from coastal to mountainous areas [8]. At that time, food distribution and transportation were more regionally distinct, and local dietary habits were more strongly reflected in the daily diet. In addition, Japan's ongoing population decline and aging may again increase the relative influence of residence‐based food environments. Demonstrating whether this association is consistent across residential environments could inform context‐aware dietary recommendations.
We therefore examined whether the association between adherence to the Japanese dietary pattern and all‐cause and CVD mortality is consistent across diverse residential topographical settings in Japan. We also described regional variation in adherence and component‐level dietary composition, and evaluated the directional consistency of the association across these areas.
2. Methods
2.1. Study Design and Population
We used data from the JACC Study, a nationwide prospective cohort of Japanese adults examining lifestyle and health outcomes [8]. Of 80 365 residents aged 40–79 years with dietary data, we excluded those with implausible or missing energy intake (< 500 or > 3500 kcal/day), insufficient information to calculate the Japanese Diet Index 8 (JDI‐8) score, or a history of CVD at baseline, leaving 48 479 participants. Reporting follows the STROBE statement and its nutritional epidemiology extension (STROBE‐nut) [9, 10].
The study protocol was approved by the Ethics Committee of Nagoya University School of Medicine and was most recently reviewed by the Ethics Committee of Hokkaido University (approval No. 14‐044). Written informed consent was obtained from most communities; community‐level procedures were used in some areas.
2.2. Dietary Assessment and Japanese Dietary Pattern Score
Usual dietary intake was assessed using a self‐administered food‐frequency questionnaire (FFQ) used in the JACC Study. The FFQ was validated in 85 participants who completed two FFQs 1 year apart and 12‐day weighed dietary records as the reference standard [11]. Spearman coefficients between the FFQ and weighed records ranged from 0.20 to 0.46 for energy and nutrients, and reproducibility between the two FFQs ranged from 0.42 to 0.86 across food groups. This indicates the FFQ suits ranking rather than estimating absolute intake, consistent with the median‐based JDI‐8 scoring. For the JDI‐8, each component was dichotomized at the sex‐specific median within the analytic cohort. One point was assigned for intake at or above the median for rice, miso soup, pickles, green and yellow vegetables, seaweed, fish, and green tea, and below the median for beef and pork. Total scores ranged from 0 to 8, with higher scores indicating greater adherence. Baseline dietary intake was assumed to reflect habitual patterns during follow‐up.
2.3. Residential Topographical Classification
Municipalities were classified into four categories: coastal, urban inland, flat/intermediate inland, and mountainous inland. Coastal designation used the National Land Numerical Information (NLNI) C23 Coastline Dataset, applied to municipalities with at least one coastline segment. Inland municipalities were classified using the Ministry of Agriculture, Forestry and Fisheries (MAFF) Agricultural Area Typology, whose first‐level classification distinguishes urban (Code 1), flat (Code 2), intermediate (Code 3), and mountainous (Code 4) areas. Urban and mountainous inland corresponded to MAFF Codes 1 and 4, and Codes 2 and 3 were combined into the flat/intermediate category. Coastal designation was prioritized over MAFF inland categories, regardless of inland characteristics. Because municipal boundaries changed, all geographic datasets were harmonized to pre‐merger municipal units before classification. The 1985 Densely Inhabited District dataset, approximating the urban/rural distribution near baseline (1988–1990), served as a sensitivity check. The classification was validated using digital elevation data from the Shuttle Radar Topography Mission [12] and the terrain ruggedness index [13]. Residence was assigned by baseline municipality; relocation during follow‐up was not modeled. Data sources and validation statistics by geographic classification appear in Tables S1 and S2.
2.4. Mortality Ascertainment and Covariates
Vital status and causes of death were obtained from death certificates provided by the Ministry of Health, Labour and Welfare. Outcomes were all‐cause mortality and CVD mortality, defined using International Classification of Diseases, 10th Revision codes I00‐I99. Person‐time accrued from baseline to death, relocation, or the end of area‐specific follow‐up.
Adjustment covariates were age, sex, total energy intake, body mass index (BMI) (< 18.5, 18.5–24.9, and ≥ 25.0 kg/m2), smoking and alcohol status (each: never, former, current), sleep duration (1 h bands from < 5 to ≥ 10 h/day), education (< 15, 15–18, and ≥ 19 years), marital status (married, widowed, divorced, single), employment status (regular, part‐time, self‐employed, homemaker, unemployed, other), sports participation (almost never; 1–2, 3–4, and ≥ 5 h/week), walking duration (almost never; < 0.5, 0.5–1, and ≥ 1 h/day), perceived stress (high vs. low), and history of cancer, diabetes, hypertension, and kidney disease (binary). Perceived stress was included as it may influence mortality risk [14]. The assumed causal structure is shown in a directed acyclic graph (Figure S1). Because BMI, diabetes, and hypertension were prevalent at baseline, preceding or coinciding with the roughly one‐year window of habitual intake captured by the FFQ, a diet‐to‐disease mediating pathway is temporally implausible for the measured exposure; these conditions were therefore treated as confounders (pre‐existing disease altering habitual diet) rather than mediators.
2.5. Statistical Analysis
Baseline characteristics were summarized using descriptive statistics. Multiple imputation by chained equations was used to handle missing covariate data, and estimates were pooled using Rubin's rules [15]. The JDI‐8 score was categorized into quartiles, with the lowest as the reference. Shared frailty Cox proportional hazards models with a random intercept for municipality (33 municipalities) estimated hazard ratios (HRs) and 95% confidence intervals (CIs) for mortality outcomes. Residential area classification was included as a fixed effect in overall models. Linear trends were evaluated by assigning ordinal integer scores to quartiles.
Effect modification was examined by adding interaction terms between JDI‐8 quartiles and residential area classification, with area‐stratified estimates reported. Proportional hazards assumptions were assessed using Schoenfeld residual tests. To address reverse causation, a sensitivity analysis excluded deaths within the first 5 years. Additional sensitivity analyses restricted the cohort to participants without selected comorbidities, excluded underweight participants, and replaced categorical area classification with continuous terrain variables.
To quantify attenuation from single‐baseline dietary measurement, we applied a regression‐dilution correction, dividing the log HR by the FFQ reliability coefficient across its reproducibility range (0.42–0.86) [16, 17]. We additionally adjusted for estimated dietary sodium (FFQ‐derived, energy‐adjusted by the residual method). We also fitted a sensitivity model omitting BMI, diabetes, and hypertension. To address residential mobility, we excluded participants who relocated during follow‐up and, because relocation is potentially informative censoring rather than a true competing event, additionally treated it as a competing event using Fine–Gray subdistribution hazards as a conservative bound. To quantify robustness to unmeasured confounding, we computed E‐values for the main HRs [18]. All analyses were performed using R version 4.3.0 (R Foundation for Statistical Computing, Vienna, Austria) with the coxme package for frailty models and the mice package for multiple imputation, with two‐sided p < 0.05 considered statistically significant, including for interaction tests.
3. Results
The analytic cohort included 48 479 participants (Figure 1). During a median follow‐up of 19.2 years (interquartile range, 11.4–20.6 years), 9304 all‐cause deaths and 2685 CVD deaths occurred. Participants in higher JDI‐8 quartiles were older, were less likely to be current smokers, reported greater physical activity and energy intake, and more often lived in flat/intermediate inland areas (Table 1).
FIGURE 1.
Flowchart of the participant selection process. Flow of participants from initial enrolment to the final analytic cohort, including inclusion and exclusion criteria. JDI‐8, Japanese Diet Index 8.
TABLE 1.
Baseline characteristics of participants according to quartiles of the JDI‐8 score.
| Q1 | Q2 | Q3 | Q4 | |
|---|---|---|---|---|
| n | 17 390 | 10 765 | 10 851 | 9027 |
| JDI‐8 score, median [IQR] | 3 [3–4] | 5 [5–5] | 6 [6–6] | 7 [7–8] |
| Age, years, mean (SD) | 54.5 (10.2) | 55.7 (9.8) | 56.5 (9.6) | 57.8 (9.3) |
| Men, n (%) | 6981 (40.1) | 4236 (39.3) | 4210 (38.8) | 3692 (40.9) |
| Body mass index, n (%) | ||||
| < 18.5 kg/m2 | 1074 (6.2) | 606 (5.6) | 634 (5.8) | 469 (5.2) |
| 18.5–24.9 kg/m2 | 12 841 (73.8) | 8043 (74.7) | 8037 (74.1) | 6688 (74.1) |
| ≥ 25.0 kg/m2 | 3475 (20.0) | 2116 (19.6) | 2181 (20.1) | 1871 (20.7) |
| Total energy intake, kcal/day, mean (SD) | 1379 (415) | 1570 (424) | 1672 (427) | 1797 (437) |
| Sleep duration, n (%) | ||||
| < 5 h | 137 (0.8) | 58 (0.5) | 47 (0.4) | 33 (0.4) |
| 5 h | 749 (4.5) | 392 (3.8) | 358 (3.4) | 271 (3.1) |
| 6 h | 3482 (20.7) | 1953 (18.8) | 1899 (18.1) | 1387 (15.9) |
| 7 h | 6345 (37.8) | 3948 (38.0) | 3915 (37.3) | 3189 (36.6) |
| 8 h | 4905 (29.2) | 3269 (31.4) | 3499 (33.3) | 3118 (35.7) |
| 9 h | 799 (4.8) | 524 (5.0) | 540 (5.1) | 477 (5.5) |
| ≥ 10 h | 376 (2.2) | 253 (2.4) | 237 (2.3) | 249 (2.9) |
| Smoking status, n (%) | ||||
| Current | 4632 (26.6) | 2519 (23.4) | 2501 (23.0) | 2086 (23.1) |
| Former | 1787 (10.3) | 1155 (10.7) | 1133 (10.4) | 996 (11.0) |
| Never | 10 971 (63.1) | 7090 (65.9) | 7218 (66.5) | 5944 (65.8) |
| Alcohol consumption, n (%) | ||||
| Current | 7875 (45.3) | 4695 (43.6) | 4664 (43.0) | 3895 (43.1) |
| Former | 568 (3.3) | 301 (2.8) | 316 (2.9) | 242 (2.7) |
| Never | 8947 (51.4) | 5769 (53.6) | 5871 (54.1) | 4891 (54.2) |
| Education level, n (%) | ||||
| < 15 years | 2001 (11.5) | 1362 (12.7) | 1454 (13.4) | 1334 (14.8) |
| 15–18 years | 12 779 (73.5) | 7947 (73.8) | 7973 (73.5) | 6563 (72.7) |
| ≥ 19 years | 2611 (15.0) | 1456 (13.5) | 1424 (13.1) | 1130 (12.5) |
| Employment status, n (%) | ||||
| Regular employment | 5050 (29.0) | 2788 (25.9) | 2598 (23.9) | 1969 (21.8) |
| Part‐time | 1455 (8.4) | 768 (7.1) | 809 (7.5) | 555 (6.1) |
| Self‐employed | 3947 (22.7) | 2837 (26.3) | 2941 (27.1) | 2555 (28.3) |
| Homemaker | 3032 (17.4) | 1992 (18.5) | 2119 (19.5) | 1812 (20.1) |
| Unemployed | 2977 (17.1) | 1743 (16.2) | 1750 (16.1) | 1564 (17.3) |
| Other | 930 (5.4) | 638 (5.9) | 634 (5.9) | 572 (6.3) |
| Marital status, n (%) | ||||
| Married | 15 174 (87.3) | 9602 (89.2) | 9742 (89.8) | 8161 (90.4) |
| Widowed | 1458 (8.4) | 860 (8.0) | 812 (7.5) | 674 (7.5) |
| Divorced | 429 (2.5) | 153 (1.4) | 167 (1.5) | 95 (1.1) |
| Single | 329 (1.9) | 151 (1.4) | 130 (1.2) | 97 (1.1) |
| Participation in sports, n (%) | ||||
| ≥ 5 h/week | 810 (4.7) | 607 (5.6) | 634 (5.8) | 637 (7.1) |
| 3–4 h/week | 1024 (5.9) | 584 (5.4) | 702 (6.5) | 556 (6.1) |
| 1–2 h/week | 2565 (14.7) | 1681 (15.6) | 1667 (15.4) | 1431 (15.9) |
| None | 12 991 (74.7) | 7893 (73.3) | 7849 (72.3) | 6403 (70.9) |
| Daily walking duration, n (%) | ||||
| ≥ 1 h | 2063 (11.9) | 1043 (9.7) | 951 (8.8) | 697 (7.7) |
| 30 min to 1 h | 3433 (19.8) | 1893 (17.6) | 1830 (16.9) | 1335 (14.8) |
| < 30 min | 3638 (20.9) | 2171 (20.2) | 2258 (20.8) | 1776 (19.7) |
| None | 8256 (47.5) | 5658 (52.6) | 5813 (53.6) | 5219 (57.8) |
| Perceived stress, n (%) | ||||
| Low/normal | 13 130 (75.5) | 8377 (77.8) | 8561 (78.9) | 7328 (81.2) |
| High | 4260 (24.5) | 2388 (22.2) | 2290 (21.1) | 1699 (18.8) |
| History of cancer, n (%) | 258 (1.5) | 156 (1.4) | 134 (1.2) | 114 (1.3) |
| History of diabetes, n (%) | 779 (4.5) | 409 (3.8) | 494 (4.6) | 389 (4.3) |
| History of hypertension, n (%) | 3171 (18.2) | 2043 (19.0) | 2096 (19.3) | 1797 (19.9) |
| History of kidney disease, n (%) | 840 (4.8) | 457 (4.2) | 458 (4.2) | 328 (3.6) |
| Residential area, n (%) | ||||
| Urban inland | 6439 (37.0) | 2919 (27.1) | 2476 (22.8) | 1859 (20.6) |
| Flat/intermediate inland | 2006 (11.5) | 1763 (16.4) | 2059 (19.0) | 2169 (24.0) |
| Mountainous inland | 7299 (42.0) | 4968 (46.1) | 5116 (47.1) | 4149 (46.0) |
| Coastal | 1646 (9.5) | 1115 (10.4) | 1200 (11.1) | 850 (9.4) |
Note: Values represent pooled estimates across 20 multiply imputed datasets. Data are presented as mean (SD), median [IQR], or n (%).
Abbreviations: IQR, interquartile range; JDI‐8, Japanese Diet Index 8; SD, standard deviation.
Urban inland residents had the lowest mean JDI‐8 score (4.58 ± 1.68) with the highest coefficient of variation (36.7%), whereas flat/intermediate inland residents had the highest mean score (5.47 ± 1.49) with the lowest coefficient of variation (27.2%). Approximately 47.0% of urban inland residents were in the lowest JDI‐8 quartile, compared with 25.1% of flat/intermediate inland residents (Figure S2). Component‐level patterns also differed: urban inland residents less often exceeded the sex‐specific median for miso soup (52.8% vs. 87.7% in flat/intermediate inland), rice (63.0% vs. 79.2%), and fish (49.4% vs. 61.2%). Coastal residents had the highest fish intake and mountainous inland residents the highest pickle intake (64.9% vs. 51.0% in coastal residents) (Figure 2).
FIGURE 2.
Proportion of participants above the sex‐specific median for each JDI‐8 component by residential area. (A) Bar chart showing the proportion of participants with intake above the sex‐specific median for each JDI‐8 component, stratified by residential area. (B) Heatmap displaying the percentage point difference compared with mountainous areas. Positive values (green) indicate higher proportions than mountainous areas; negative values (red) indicate lower proportions. Urban residents had notably lower proportions for miso soup (−19.1 percentage points), fish (−9.6), and rice (−11.1), while coastal residents showed higher proportions for fish (+4.1) and green and yellow vegetables (+14.7) but lower for pickles (−13.9). JDI‐8, Japanese Diet Index 8.
In shared frailty Cox models, higher JDI‐8 adherence was inversely associated with both all‐cause and CVD mortality (Figure 3). Comparing the highest with the lowest JDI‐8 quartile, the HR was 0.929 (95% CI, 0.872–0.990; p for trend = 0.034) for all‐cause mortality and 0.862 (95% CI, 0.766–0.970; p for trend = 0.023) for CVD mortality. Municipality‐level clustering was modest (intraclass correlation coefficients, 0.8% for all‐cause and 3.5% for CVD mortality). Schoenfeld tests indicated no major violations of the proportional hazards assumption.
FIGURE 3.
Associations between quartiles of the JDI‐8 score and mortality from all causes and cardiovascular disease. HRs and 95% CIs were estimated using shared frailty Cox proportional hazards models with a random intercept for municipality (33 municipalities) to account for within‐area clustering. Residential area (coastal, urban inland, flat/intermediate inland, mountainous inland) was included as a fixed effect. Results were obtained from multiple imputation (20 datasets, pooled by Rubin's rules). p for trend was calculated by assigning integer values (1–4) to the quartiles and modeling the variable as continuous. Quartile 1 (Q1) is the reference group. Models were adjusted for age, sex, total energy intake, body mass index (< 18.5, 18.5–24.9, and ≥ 25.0 kg/m2), smoking status (current, former, never), alcohol consumption (current, former, and never), sleep duration (< 5, 5, 6, 7, 8, 9, and ≥ 10 h/day), education level (age at last graduation: < 15, 15–18, and ≥ 19 years), employment status (regular employment, part‐time, self‐employed, homemaker, unemployed, and other), marital status (married, widowed, divorced, single), participation in sports (≥ 5, 3–4, 1–2 h/week, and none), daily walking duration (≥ 1 h, 30 min to 1 h, and < 30 min, none), perceived stress (high, low/normal), and history of cancer, diabetes, hypertension, and kidney disease (yes/no for each). CI, confidence interval; CVD, cardiovascular disease; HR, hazard ratio; JDI‐8, Japanese Diet Index 8.
Interaction tests indicated no significant heterogeneity by residential setting (p for interaction = 0.246 for all‐cause mortality and 0.185 for CVD mortality); stratified estimates are descriptive. Across strata, point estimates were directionally consistent, strongest in urban inland areas (CVD HR 0.614; 95% CI, 0.459–0.822). Other strata are shown in Figure 4.
FIGURE 4.
Associations between quartiles of the JDI‐8 score and mortality stratified by residential area. HRs and 95% CIs were estimated using shared frailty Cox proportional hazards models with a random intercept for municipality within each residential area stratum. Results were obtained from multiple imputation (20 datasets, pooled by Rubin's rules). Quartile 1 (Q1) is the reference group within each stratum. p for interaction was calculated using a likelihood ratio test comparing shared frailty models with and without the cross‐product terms between JDI‐8 quartiles and residential area. p for trend was calculated by assigning integer values (1–4) to the quartiles and modeling the variable as continuous within each stratum. Models were adjusted for the same covariates as in Figure 3, excluding residential area. Residential areas were classified based on the National Land Numerical Information C23 Coastline Dataset and the Ministry of Agriculture, Forestry and Fisheries Agricultural Area Typology (see Section 2). CI, confidence interval; CVD, cardiovascular disease; HR, hazard ratio; JDI‐8, Japanese Diet Index 8.
Sensitivity analyses supported robustness. Results remained consistent after excluding deaths within the first 5 years of follow‐up, restricting to participants without selected baseline comorbidities, and excluding underweight participants (Figures S3 and S4). Replacing categorical residential classification with continuous terrain variables (mean elevation and coastline distance) left HRs essentially unchanged for both all‐cause mortality (HR 0.929; 95% CI, 0.872–0.990) and CVD mortality (HR 0.862; 95% CI, 0.766–0.970) (Figure S5).
After correction for regression dilution, HRs ranged from 0.918 (95% CI, 0.853–0.989) to 0.840 (0.722–0.977) for all‐cause mortality and from 0.841 (0.734–0.965) to 0.702 (0.530–0.930) for CVD mortality across FFQ reliability values from 0.86 to 0.42, indicating the primary estimates are likely conservative (Table S3). Additional adjustment for estimated sodium intake did not attenuate the associations (all‐cause HR 0.919, 95% CI, 0.861–0.982; CVD HR 0.833, 0.738–0.940; Table S4). Estimates were essentially unchanged omitting baseline BMI, diabetes, and hypertension (all‐cause HR 0.939, 0.881–1.001; CVD HR 0.889, 0.790–1.000). Excluding participants who relocated (n = 1664) gave essentially identical results (all‐cause HR 0.926, 0.869–0.986; CVD HR 0.858, 0.763–0.966), and treating relocation as a competing event yielded subdistribution hazard ratios (sHRs) that were directionally concordant but attenuated toward the null, with CIs including 1 (all‐cause sHR 0.948, 0.891–1.009; CVD sHR 0.907, 0.807–1.020; Table S5). Within‐municipality variability in terrain indicators is summarized in Table S6; within‐municipality standard deviations were modest in urban, flat/intermediate, and coastal municipalities and largest in mountainous ones.
4. Discussion
In this large prospective cohort study of Japanese adults, higher adherence to the Japanese dietary pattern was associated with lower all‐cause and CVD mortality across multiple residential topographical settings. Formal interaction tests did not show statistically significant effect modification by residential area. The direction of the overall inverse association was preserved after accounting for municipality‐level clustering.
Our overall findings are consistent with prior Japanese cohort studies and a systematic review. The Ohsaki cohort reported inverse associations with all‐cause mortality, functional disability, and dementia [3]; the JPHC Study reported similar associations for cause‐specific mortality [4]; and a meta‐analysis of prospective cohorts supported an inverse association with CVD mortality [5]. Our results extend this literature by demonstrating that the association is preserved across heterogeneous topographical settings within Japan. Transportability to non‐Japanese populations cannot be inferred; international and ecological analyses generate hypotheses [19, 20] but require validation in individual‐level cohorts.
The exploratory stratified findings warrant cautious interpretation. The protective association appeared most pronounced in urban inland areas. Several non‐mutually exclusive explanations may contribute. First, urban inland residents had the lowest mean JDI‐8 score and the highest variability in adherence, providing a wider exposure contrast and greater statistical power to detect associations. National data show that Japanese dietary habits transitioned more rapidly in urban areas during the late 20th century, with rising fat intake and lower adherence to traditional patterns [21]. Second, differences in sample size and event counts across areas may have contributed to imprecision in the coastal, flat/intermediate, and mountainous strata. Accordingly, the principal interpretation should rest on the overall estimate, with stratified estimates serving as hypothesis‐generating signals.
Even when total JDI‐8 scores were similar, the underlying food combinations differed by region (Figure 2), consistent with national and regional evidence of within‐country dietary variation in Japan [6, 7]. These observations imply that a single summary score may obscure regionally distinct dietary practices.
This study has several strengths. The JACC Study is a large nationwide cohort with long follow‐up, enabling population‐level assessment of all‐cause and CVD mortality. A distinct feature is the inclusion of residents from coastal to mountainous environments, allowing examination across diverse topographies, a perspective not addressed in prior Japanese cohort studies. Shared frailty Cox models addressed within‐municipality clustering, and the robustness of the main findings was supported by multiple sensitivity analyses.
Several limitations should be considered. First, dietary intake was assessed only at baseline using a 40‐item FFQ, and dietary changes during the 19‐year follow‐up may have caused non‐differential exposure misclassification with attenuation toward the null. In a quantitative bias analysis using the FFQ reliability coefficient, the HRs moved further from the null, indicating that the reported associations are likely conservative [16, 17]. However, absolute intake estimates from a single FFQ carry residual uncertainty. Second, residence was assigned at baseline using the municipality of registration; subsequent relocation was not modeled, and within‐municipality heterogeneity in food availability and socioeconomic context was not captured. To assess the impact of mobility, we excluded participants who relocated during follow‐up and, as a conservative bound for potentially informative censoring, treated relocation as a competing event; both analyses were consistent with the main results. Third, although the analysis adjusted for sociodemographic, lifestyle, psychosocial, and clinical factors, residual confounding by unmeasured determinants of both diet and mortality cannot be excluded. Fourth, because interaction tests were not statistically significant and events per stratum were modest, the stratified estimates should be regarded as exploratory rather than definitive evidence of regional effect modification. Finally, the JDI‐8 score used median‐based cut‐offs derived within the analytic cohort, which standardizes ranking internally but limits direct comparison of score‐mortality associations across study populations with different intake distributions. This approach is common in prior Japanese cohort studies, although variants using absolute intake thresholds also exist.
Our findings provide observational evidence that adherence to the Japanese dietary pattern is associated with lower all‐cause and CVD mortality among Japanese adults. The absence of statistically significant heterogeneity suggests the overall inverse association is not driven by a single topographical context, although this does not prove a uniform causal effect across regions. The observed regional differences in component composition imply that translation of these findings into dietary recommendations would benefit from consideration of local food availability and area‐specific socioeconomic conditions. The JDI‐8 captures both cardioprotective components (fish, soy foods, vegetables, and seaweed) and higher‐sodium traditional items (miso soup and pickles); because additional adjustment for estimated sodium did not attenuate the inverse associations, the benefit does not appear to be driven by salt‐containing components, and dietary guidance should emphasize maintaining these beneficial foods while limiting salt. Regional differences in the standard and accessibility of medical care and in area socioeconomic context were partly captured by the municipality‐level random intercept, although the small estimated between‐municipality variance and the absence of direct measures of healthcare access mean residual differences cannot be excluded; E values indicated that an unmeasured confounder would need to be associated with both JDI‐8 adherence and mortality by a risk ratio of 1.36 (all‐cause) or 1.59 (CVD) to explain away the association, and by 1.11 or 1.21, respectively, to shift the CI to the null [18]. Cohort studies with repeated dietary measurements, finer‐grained residential exposure assessment, and analyses incorporating mechanistic biomarkers will help clarify whether the observed association reflects a generalizable dietary signal.
5. Conclusions
Higher adherence to the Japanese dietary pattern was associated with lower all‐cause and CVD mortality among Japanese adults, consistently across diverse residential topographical settings.
Author Contributions
Akio Shimizu contributed to conceptualization, methodology, formal analysis, and writing of the original draft. Yasutake Tomata contributed to conceptualization, data curation, supervision, and review and editing. Junji Miyazaki contributed to methodology, statistical analysis, and review and editing. Takashi Kimura contributed to data curation, methodology, and statistical analysis. Kenji Wakai contributed to investigation, supervision, validation, and review and editing. Ryo Momosaki contributed to conceptualization, supervision, funding acquisition, and review and editing. All authors contributed to interpretation of data, critically revised the manuscript for important intellectual content, approved the final version, and agreed to be accountable for all aspects of the work.
Funding
This work was supported by Grants‐in‐Aid for Scientific Research and by Grants‐in‐Aid for Scientific Research on Priority Areas of Cancer and Cancer Epidemiology from the Ministry of Education, Culture, Sports, Science and Technology of Japan (grant numbers 61010076, 62010074, 63010074, 1010068, 2151065, 3151064, 4151063, 5151069, 6279102, 11181101, 17015022, 18014011, 20014026, and 20390156). This work was also supported by the Japan Society for the Promotion of Science KAKENHI Grant Number 23K16799.
Ethics Statement
The study protocol was initially approved by the Ethics Committee of Nagoya University School of Medicine and was most recently reviewed by the Ethics Committee of Hokkaido University (Approval No. 14‐044).
Consent
Written informed consent was obtained from participants in most communities. In a minority of communities, consent was provided at the community level according to the approved study procedures.
Conflicts of Interest
The authors declare no conflicts of interest.
Supporting information
Table S1: Data sources used for geographic classification (web resources).
Table S2: DEM‐derived terrain indicators by geographic classification.
Table S3: Quantitative bias analysis for exposure misclassification.
Table S4: Sodium‐adjusted sensitivity analysis.
Table S5: Sensitivity analyses addressing residential relocation.
Table S6: Within‐municipality heterogeneity in terrain indicators by geographic classification.
Figure S1: Directed acyclic graph of the assumed causal structure.
Figure S2: Distribution of JDI‐8 scores by residential area.
Figure S3: Sensitivity analyses for associations between quartiles of the JDI‐8 score and all‐cause mortality.
Figure S4: Sensitivity analyses for associations between quartiles of the JDI‐8 score and CVD mortality.
Figure S5: Sensitivity analysis comparing categorical area classification with continuous terrain variable adjustment for associations between quartiles of the JDI‐8 score and all‐cause and CVD mortality.
Acknowledgments
We thank all participants and investigators of the Japan Collaborative Cohort Study. During the preparation of this manuscript, the authors used Claude (Anthropic) solely for language editing and proofreading. The tool was not used for generating scientific content, study design, data analysis, interpretation of results, or formulation of conclusions. All scientific content, analyses, and conclusions are the work of the authors, who reviewed and revised every output and take full responsibility for the content and integrity of the publication.
Data Availability Statement
Data sharing not applicable to this article as no datasets were generated or analyzed during the current study.
References
- 1. Shimazu T., Kuriyama S., Hozawa A., et al., “Dietary Patterns and Cardiovascular Disease Mortality in Japan: A Prospective Cohort Study,” International Journal of Epidemiology 36, no. 3 (2007): 600–609, 10.1093/ije/dym005. [DOI] [PubMed] [Google Scholar]
- 2. Tomata Y., Zhang S., Kaiho Y., Tanji F., Sugawara Y., and Tsuji I., “Nutritional Characteristics of the Japanese Diet: A Cross‐Sectional Study of the Correlation Between Japanese Diet Index and Nutrient Intake Among Community‐Based Elderly Japanese,” Nutrition 57 (2019): 115–121. [DOI] [PubMed] [Google Scholar]
- 3. Matsuyama S., Shimazu T., Tomata Y., et al., “Japanese Diet and Mortality, Disability, and Dementia: Evidence From the Ohsaki Cohort Study,” Nutrients 14 (2022): 2034. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Murai U., Ishihara J., Takachi R., et al., “Japanese Diet Index Score and Cause‐Specific Mortality in Japanese Men and Women: The Japan Public Health Center‐Based Prospective Study,” Journal of Nutrition 155 (2025): 3726–3736. [DOI] [PubMed] [Google Scholar]
- 5. Shirota M., Watanabe N., Suzuki M., and Kobori M., “Japanese‐Style Diet and Cardiovascular Disease Mortality: A Systematic Review and Meta‐Analysis of Prospective Cohort Studies,” Nutrients 14, no. 10 (2022): 2008, 10.3390/nu14102008. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Okada E., Okada C., Matsumoto M., and Takimoto H., “Regional Trends in Dietary Intake in Japan: The 2003‐2017 National Health and Nutrition Survey,” Japanese Journal of Nutrition and Dietetics 78 (2020): S16–S26. [Google Scholar]
- 7. Uechi K., Asakura K., Masayasu S., and Sasaki S., “Within‐Country Variation of Salt Intake Assessed via Urinary Excretion in Japan: A Multilevel Analysis in All 47 Prefectures,” Hypertension Research 40 (2017): 598–605. [DOI] [PubMed] [Google Scholar]
- 8. Tamakoshi A., Ozasa K., Fujino Y., et al., “Cohort Profile of the Japan Collaborative Cohort Study at Final Follow‐Up,” Journal of Epidemiology 23 (2013): 227–232. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Vandenbroucke J. P., von Elm E., Altman D. G., et al., “Strengthening the Reporting of Observational Studies in Epidemiology (STROBE): Explanation and Elaboration,” PLoS Medicine 4 (2007): e297. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Lachat C., Hawwash D., Ocke M. C., et al., “Strengthening the Reporting of Observational Studies in Epidemiology‐Nutritional Epidemiology (STROBE‐Nut): An Extension of the STROBE Statement,” PLoS Medicine 13 (2016): e1002036. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Date C., Fukui M., Yamamoto A., et al., “Reproducibility and Validity of a Self‐Administered Food Frequency Questionnaire Used in the JACC Study,” Journal of Epidemiology 15, no. 1 (2005): S9–S23. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Farr T. G., Rosen P. A., Caro E., et al., “The Shuttle Radar Topography Mission,” Reviews of Geophysics 45 (2007): RG2004. [Google Scholar]
- 13. Riley S. J., DeGloria S. D., and Elliot R., “A Terrain Ruggedness Index That Quantifies Topographic Heterogeneity,” Intermountain Journal of Sciences 23 (1999): 23–27. [Google Scholar]
- 14. Kivimaki M. and Steptoe A., “Effects of Stress on the Development and Progression of Cardiovascular Disease,” Nature Reviews. Cardiology 15 (2018): 215–229. [DOI] [PubMed] [Google Scholar]
- 15. Shah A. D., Bartlett J. W., Carpenter J., Nicholas O., and Hemingway H., “Comparison of Random Forest and Parametric Imputation Models for Imputing Missing Data Using MICE: A CALIBER Study,” American Journal of Epidemiology 179 (2014): 764–774. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Hughes M. D., “Regression Dilution in the Proportional Hazards Model,” Biometrics 49, no. 4 (1993): 1056–1066. [PubMed] [Google Scholar]
- 17. Clarke R., Shipley M., Lewington S., et al., “Underestimation of Risk Associations due to Regression Dilution in Long‐Term Follow‐Up of Prospective Studies,” American Journal of Epidemiology 150, no. 4 (1999): 341–353, 10.1093/oxfordjournals.aje.a010013. [DOI] [PubMed] [Google Scholar]
- 18. VanderWeele T. J. and Ding P., “Sensitivity Analysis in Observational Research: Introducing the E‐Value,” Annals of Internal Medicine 167, no. 4 (2017): 268–274, 10.7326/M16-2607. [DOI] [PubMed] [Google Scholar]
- 19. Abe C., Imai T., Sezaki A., et al., “Global Association Between Traditional Japanese Diet Score and All‐Cause, Cardiovascular Disease, and Total Cancer Mortality: A Cross‐Sectional and Longitudinal Ecological Study,” Journal of the American Nutrition Association 42 (2023): 660–667. [DOI] [PubMed] [Google Scholar]
- 20. Aono M., Ushio S., Araki Y., et al., “Japanese Diet Indices and Nutrient Density in US Adults: A Cross‐Sectional Analysis With NHANES Data,” Nutrients 16 (2024): 2431. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Yoshiike N., Matsumura Y., Iwaya M., Sugiyama M., and Yamaguchi M., “National Nutrition Survey in Japan,” Journal of Epidemiology 6 (1996): 189–200. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Table S1: Data sources used for geographic classification (web resources).
Table S2: DEM‐derived terrain indicators by geographic classification.
Table S3: Quantitative bias analysis for exposure misclassification.
Table S4: Sodium‐adjusted sensitivity analysis.
Table S5: Sensitivity analyses addressing residential relocation.
Table S6: Within‐municipality heterogeneity in terrain indicators by geographic classification.
Figure S1: Directed acyclic graph of the assumed causal structure.
Figure S2: Distribution of JDI‐8 scores by residential area.
Figure S3: Sensitivity analyses for associations between quartiles of the JDI‐8 score and all‐cause mortality.
Figure S4: Sensitivity analyses for associations between quartiles of the JDI‐8 score and CVD mortality.
Figure S5: Sensitivity analysis comparing categorical area classification with continuous terrain variable adjustment for associations between quartiles of the JDI‐8 score and all‐cause and CVD mortality.
Data Availability Statement
Data sharing not applicable to this article as no datasets were generated or analyzed during the current study.