Nutrients, Vol. 18, Pages 506: Differential Network-Based Dietary Structure and Type 2 Diabetes Risk: A Prospective Cohort Study Using Food Co-Consumption Networks
Nutrients doi: 10.3390/nu18030506
Authors:
Hye Won Woo
Yu-Mi Kim
Min-Ho Shin
Sang Baek Koh
Hyeon Chang Kim
Mi Kyung Kim
Background/Objectives: Current data-driven dietary pattern methods have limitations in identifying disease-specific dietary structures. We developed network-derived dietary scores based on type 2 diabetes (T2D)-differential food co-consumption networks and examined their associations with incident T2D risk. Methods: Using the Korean Genome and Epidemiology Study-CArdioVascular disease Association Study (KoGES-CAVAS, n = 16,665), we constructed food co-consumption networks from cumulative average intakes stratified by incident T2D status. The network centrality scores from edges appearing exclusively in either T2D or non-T2D networks were used to generate a differential co-consumption network-derived (D_CCN) score, with higher scores indicating a greater alignment with diabetes-specific structures. CAVAS-derived scores were applied to the Health Examinee Study (KoGES-HEXA, n = 51,206) for cross-cohort validation. Incidence rate ratios (IRRs) were estimated using modified Poisson regression with robust error estimation. Results: During follow-up, 953 and 2190 new cases of T2D were identified in CAVAS and HEXA, respectively. Rice and vegetable dishes were primary hub foods in both networks, with rice showing exclusively negative correlations. Non-T2D networks were more complex, whereas T2D networks were simpler and centered on refined flour-based foods. The D_CCN score was associated with a higher T2D risk in CAVAS (IRR = 1.45, 95% CI: 1.21–1.74), and this association was validated in HEXA (IRR = 1.58, 95% CI: 1.40–1.78), with consistent dose–response relationships (both p-trend < 0.0001). Conclusions: Differential network analysis identified T2D-specific co-consumption structures, and the D_CCN score consistently predicted T2D risk across cohorts. This approach highlights the utility of network-based methods for capturing disease-relevant dietary structures beyond traditional approaches.
Background/Objectives: Current data-driven dietary pattern methods have limitations in identifying disease-specific dietary structures. We developed network-derived dietary scores based on type 2 diabetes (T2D)-differential food co-consumption networks and examined their associations with incident T2D risk. Methods: Using the Korean Genome and Epidemiology Study-CArdioVascular disease Association Study (KoGES-CAVAS, n = 16,665), we constructed food co-consumption networks from cumulative average intakes stratified by incident T2D status. The network centrality scores from edges appearing exclusively in either T2D or non-T2D networks were used to generate a differential co-consumption network-derived (D_CCN) score, with higher scores indicating a greater alignment with diabetes-specific structures. CAVAS-derived scores were applied to the Health Examinee Study (KoGES-HEXA, n = 51,206) for cross-cohort validation. Incidence rate ratios (IRRs) were estimated using modified Poisson regression with robust error estimation. Results: During follow-up, 953 and 2190 new cases of T2D were identified in CAVAS and HEXA, respectively. Rice and vegetable dishes were primary hub foods in both networks, with rice showing exclusively negative correlations. Non-T2D networks were more complex, whereas T2D networks were simpler and centered on refined flour-based foods. The D_CCN score was associated with a higher T2D risk in CAVAS (IRR = 1.45, 95% CI: 1.21–1.74), and this association was validated in HEXA (IRR = 1.58, 95% CI: 1.40–1.78), with consistent dose–response relationships (both p-trend < 0.0001). Conclusions: Differential network analysis identified T2D-specific co-consumption structures, and the D_CCN score consistently predicted T2D risk across cohorts. This approach highlights the utility of network-based methods for capturing disease-relevant dietary structures beyond traditional approaches. Read More