Results
Variance partitioning of the kcal data attributed 3.65% of variance to SS_a (individuals – Mode A), 17.02% to SS_b (macronutrient – Mode B), 29.15% to SS_c (timeslot- Mode C), 3.69% to SS_ab (individual-nutrient interaction), 18.83% to SS_ac (individual-time interaction), 10.62% to SS_bc (nutrient-time interaction), and 16.67% to SS_abc (individual-nutrient-time interaction). The timeslot, nutrient and the individual-time interaction were the dominant variance sources, followed by the three-way interaction, confirming individual-nutrient-time heterogeneity and justifying the use of Tucker3. The rotated individuals (A) mode had two components: A2, captures the population average pattern, and A1, captures the individual deviation from the baseline pattern. The rotated macronutrient (B) had also two: B1, contrasting fat (0.61) and protein (0.24) against carbohydrate (-0.73), and B2, capturing an average sustained carbohydrate and fat intake (0.68 and 0.67 respectively). The rotated temporal (C) mode had four components: C1, late-night snacking (h6 = 0.96); C2, both lunch and dinner eating (h3 = 0.67, h5 = 0.69); C3, morning eating (h1 = 0.78, h2 = 0.48); and C4, a contrast between lunch (h3 = 0.71) and dinner (h5 = −0.69). The dominant core interaction, A2 × B2 × C2 = 20,222.76, described the variation from average individual pattern of carbohydrate-fat intake concentrated at lunch and dinner. A carbohydrate-dominant, fat-depleted morning meal was captured by A2 × B1 × C3 = −1,956.53. The A1 subject component identified individuals who concentrate their caloric intake disproportionately at lunch (A1 × B2 × C4 = 6,119.29; A1 × B1 × C4 = 1,350.24). Demographic profiling indicated that the A1 pattern was significantly associated with Southern Italian residence (p = 0.006) and employment status (p = 0.027 for retirement), while body mass index showed no association.
