*2.4. Statistics*

To represent the overall US population, all analyses incorporated oversampling, clustering, and stratification as recommended by the NHANES data analysis guideline [31]. Continuous variables are listed as the weighted mean and 95% confidence interval (CI), while categorical variables are presented as weighted proportions. Basic characteristics are compared by baseline glycemic status using the adjusted Wald test for continuous variables and Rao-Scott χ<sup>2</sup> test for categorical variables.

The weighted Cox proportional hazard regression models were adopted to assess the impact of DII on participants' long-term mortality, which were adjusted for age, sex, educational level, BMI, smoke, hypertension, hyperlipidemia, glycemic status, recreational activity, and alcohol consumption. In addition to estimating DII as a continuous variable, we equally classified participants into three groups: low DII, medium DII, and high DII. Similar Cox regression models as well as weighted Kaplan-Meier curves were adopted to estimate the correlation between all-cause and cardiovascular mortality and different DII groups.

Furthermore, to test whether the impact of DII on prognosis is different across patients with different glycemic statuses, the weighted Cox regression model and interaction *p* value were used to estimate the relationship between DII (continuous/categorical variable) and participants' long-term mortality in participants with normal glucose status, prediabetes, and type 2 diabetes. The regression model was adjusted for age, sex, educational level, BMI, smoke, hypertension, hyperlipidemia, recreational activity, and alcohol consumption.

All analyses were conducted by the R software (version 4.1.2, R Foundation for Statistical Computing, Vienna, Austria) and Stata (version 16.0, StataCorp, College Station, TX, USA). A two-sided *p* value < 0.05 was considered statistically significant.
