As a stratified and multistage study, NHANES is designed by National Center for Health Statistics (NCHS). A nationally representative sample of the non-institutionalized civilian population of the U.S. was investigated to assess the health and nutritional status. The detailed introduction about NHANES has been described elsewhere. This study recruited people aged over 18 years with diabetes who participated in the NHANES during 2003-2014. Diabetes was defined by the following any criteria: 1) self-reported information, 2) fasting plasma glucose level (FPG) > 7.0 mmol/l, 3) hemoglobin A1c level (HA1c) > 6.5%. After excluding participants with diabetes who had 1) missing or unknown information on any dietary nutrient intake, 2) missing information of mortality. Finally, 4699 participants with diabetes (2413 men and 2286 women) were included. Research Ethics Review Board of National Center for Health Statistics approved NHANES, and all participants with diabetes had already provided written informed consent before initial exam. The NHANES data could be accessed through https://wwwn.cdc.gov/nchs/nhane.
Information about food intakes for two non-consecutive days were collected through 24-hour dietary recall interviews. In the first one, the dietary investigation was conducted in-person, and in the other one, the dietary investigation was conducted 3–10 days afterwards by calling. According to the United States Department of Agriculture’s Food and Nutrient Database for Dietary Studies, the individual energy and nutrient intake was estimated. In line with the MyPyramid Equivalents Database 2.0 for USDA Survey Foods (MPED 2.0), dietary food consumption of participants with diabetes in the NHANES was also integrated into 18 definable MyPyramid major food groups.
Align with the consumption time, we split the major food groups intake into different period, such as the forenoon (breakfast plus snack between breakfast and lunch), the afternoon (lunch plus snack between lunch and dinner), the evening (dinner plus snack after dinner).
In this research, the status of CVD and all-cause mortality was the major outcome variable. The mortality status was determined by the National Death Index (NDI) until 31 December 2015. The NDI is a highly reliable resource, which has been widely used for death identification. As an international disease classification method, the ICD-10 is used for the determination of disease-specific death. The ICD-10 codes I00–I09, I11, I13, I20–I51, or I60–I69 was defined as CVD mortality. In total, this study documented 913 deaths, including 314 deaths due to CVD death.
Assessment of covariates
Non-dietary covariates were sex (male/female), age (years), BMI (kg/m2), drink (yes/no), smoke (yes/no), regular exercise (yes/no), education ( less than 9th grade, between 9th and 11th grade, graduate from high school, GED or equivalent, Arts degree of some college or Associate, college graduate or above), race/ethnicity (non-Hispanic white/non-Hispanic black/Mexican American/other), household income per year (<$20,000, $20,000–$45,000, $45,000–$75,000, or >$100,000), family history of diabetes (yes/no), diagnosis record of hypertension or dyslipidemia (yes/no), drug use for controlling blood pressure, cholesterol and glucose. And the dietary covariates included total intakes of energy (kcal/day), carbohydrate (g/d), dietary fat (g/day), protein (g/day), major food groups and Alternate Healthy Eating Index (AHEI), which is an indicator of dietary quality.
All analyses were performed according to the guidelines analytic of NHANES. Continuous variables about demographic characteristics, anthropometric measurements, and dietary nutrient and food intakes, were showed as mean (SD) or median (P25, P75); whereas, categorical variables were showed as number (percentage). General linear models adjusting for age and x2 tests were used to compare baseline characteristics by mortality status. All statistical analyses were conducted by R 4.0.2, and two-sided P < 0.05 was considered to be statistically significant.
Cox proportional hazards models
Food groups consumed in forenoon, afternoon, and evening were transformed into categorical variables based on their distribution, respectively. Cox proportional hazards (CPH) models were performed to estimate the hazards ratio(HR) and 95% confidence interval (CI) for the association of food consumed in the forenoon, afternoon and evening with CVD and all-cause mortality. Follow-up years of participants with diabetes between interview date and death or census date (31 December 2015) was defined as Survival time. A series of covariates was also controlled, which were including age, sex, ethnics, education, income, smoking, drinking, regular exercise habits, BMI, total intake of daily energy, fat, carbohydrate, protein, family history of diabetes, hypertension and dyslipidemia, and medication. Moreover, when analyzing one food group consumed in one time period across a day, we also controlled the total intake of this food group in the whole day.
Predicted isocaloric models
Based on the CPH model developed in previous steps, we also built several isocaloric models to evaluate relative risk of deaths with altered distribution of food consumption time, which was conducted via one food group consumed at one time period theoretically replacing with the equivalent food group consumed at another time period. A key rationale of the substitution analysis is that, in the isocaloric setting, the total intake of energy, macronutrients and the food group are held constant.
We performed three kinds of sensitivity analyses. In the first analysis, we excluded the diabetes patients, whose survival time less than two years or follow-up duration less than two years to examine the impact of severe illness or accident on the results. In the second analysis, the indicator of overall dietary quality and breakfast skipping were additionally adjusted in the CPH model to evaluate whether these confounders would influence the results.