*2.5. Statistics*

Continuous variables were presented as means and standard deviations or medians and quartiles according to the distribution of data; categorical variables were presented as percentages. Normally distributed continuous variables were tested between groups by one-way analysis of variance, for categorical variables, chi-square tests were performed for di fference across groups. To estimate the trends of nutrient intakes across DDS tertiles (T1, T2, T3), linear regression models were conducted. We assigned the midpoint values of DDS tertiles and treated the variables as continuous in the linear regression model. Values of nutrient intakes were transformed to log to reach normality.

Person years were measured from baseline survey date until the last survey date (report of ADL disability, loss to follow-up, or the phase of 2015, whichever occurred first). To assess the e ffect of DDS on the risk of ADL disability, hazard ratios (HRs) for DDS and ADL disability were estimated by Cox proportional hazard regression models. Multivariate models were conducted. In the first model, covariates including age at entry (continuous), gender (men or women), living region (southern or northern China), residency (urban or rural), income (low, middle, or high), and education level (primary school and below or middle school and higher) were adjusted. In the second model, smoking status (smoker or not), physical activity (≤100 or >100 MET-hours/week), BMI (continuous), and comorbidity (yes or no) were additionally adjusted.

Sensitivity analyses were performed by (1) excluding individuals whose follow-up time was less than 5 years and (2) the adjustment of alcohol consumption (regular consumer or not) and phase at entry (1997, 2000, 2004, 2006, 2009, or 2011). Subgroup analyses were conducted in participants with di fferent baseline characteristics (gender, living region, age at entry (≤65 or >65 years old), and comorbidity).

To assess the short-term and the long-term e ffects of DDS on ADL disability, we analyze the DDS value reported in di fferent surveys with the odds of ADL disability. Individuals involved in three or more dietary surveys were included. Logistic regression models were performed to estimate the odds ratios for the average DDS across phases, the baseline DDS, and the recent DDS prior to the end of the survey in relation to ADL disability, respectively.

All the statistics were conducted in R 4.0.2. All *p* values were two-sided, and statistical significance was defined as *p* < 0.05.
