*2.6. Statistical Analysis*

Descriptive statistics were calculated for all outcomes and exposures. Pearson's correlations were calculated between the four diet scores at each visit separately and between time points. Independent two-sample *t*-tests, chi-square tests, or Fisher's exact tests were used to test for differences in participant demographics between the baseline cohort and follow-up cohort. Paired *t*-tests or McNemar–Bowker tests were used to test for differences in exposures and outcomes between visits. Due to the small numbers of participants with values meeting the diagnostic criteria for type 2 diabetes, prediabetes and diabetes were combined into one category (prediabetes/T2D) for analysis. Primary outcomes of interest were those related to glucose regulation: prediabetes/T2D, fasting glucose, 2-h glucose, glucose AUC, and HbA1c. Body composition measurements were secondary outcomes: BMI, body fat percent, FFMI, fat mass to height ratio, android to gynoid ratio, trunk to leg ratio, trunk to limb ratio, and VAT.

Cross-sectional analyses were performed for both baseline and follow-up visits, using multivariable linear regression for continuous outcomes and logistic regression for prediabetes/T2D. For longitudinal analyses, change in diet indices from baseline to follow-up was modeled against change in outcome using linear regression for continuous outcomes, or against diabetes at follow-up using logistic regression. Longitudinal models also adjusted for baseline diet score. Beta coefficients for exposures were scaled to one standard deviation (SD) of the exposure to account for the differing scales.

All analyses included the following covariates: age, ethnicity, physical activity, energy intake, and parental education. Because these factors were not accounted for in the scoring system, analyses with HEI, DASH, and DII scores additionally controlled for sex, and analyses with MDS additionally controlled for energy intake. BMI and body fat percent were presumed to be on the causal pathway between diet and prediabetes and T2D and were not included as covariates in the main analyses to avoid overadjustment [42].

#### *2.7. Sensitivity Analyses*

For all diet indices and glucose outcomes, two additional analyses were performed. The first did not include physical activity in as a covariate to determine if it had the potential to confound the relationship between diet and glucose regulation and if it was necessary to control for this variable in the main analysis. The second analysis controlled for body fat percent. Though we expect that body fat (or BMI) is on the causal pathway between diet and T2D, we included it as a covariate to examine the possibility that body fat mediates the relationship between diet and T2D.

We also performed additional logistic regression analyses to examine the association between each adiposity measure and risk for prediabetes/T2D at each visit and to examine the associations between changes in these measures between visits and risk for prediabetes/T2D at the follow-up visit. Models were adjusted for age, sex, ethnicity, parental education, energy intake, and physical activity as in the main analyses.

### **3. Results**

Average length of follow-up was 4.1 years (SD = 1.1 years). There were no differences in participant demographics at each visit (Table 1). HEI, DASH, and DII scores significantly decreased from baseline to follow-up (Table 2), and mean fasting glucose and glucose AUC increased (Table 3). Mean BMI and body fat percentage also increased between visits (Table 4).


**Table 1.** Descriptive statistics for participant demographics at baseline and follow-up.

<sup>1</sup> Includes three participants who did not complete the baseline visit. <sup>2</sup> *p*-values calculated using chi-Square or Fisher's exact tests. <sup>3</sup> Response to the question "Do you exercise?". SD: standard deviation.

**Table 2.** Descriptive statistics for diet at baseline, follow-up, and change between visits.


<sup>1</sup> Three additional CHS participants participated in the second visit without having completed the first. <sup>2</sup> *p*-values calculated using paired *t*-tests. Abbreviations: HEI: Healthy Eating Index—2015; MDS: Mediterranean Diet Score; DASH: Dietary Approaches to Stop Hypertension; DII: Dietary Inflammatory Index.


**Table 3.** Descriptive statistics for glucose outcomes at baseline, follow-up, and change between visits.

<sup>1</sup> Three additional CHS participants participated in the second visit without having completed the first. <sup>2</sup> *p*values calculated using paired *t*-tests for continuous variables and McNemar–Bowker test for diabetes categories. Abbreviations: SD: standard deviation; HbA1c: hemoglobin A1c; AUC: area under the curve.

**Table 4.** Descriptive statistics for body composition at baseline, follow-up, and change between visits.


<sup>1</sup> Three additional CHS participants participated in the second visit without having completed the first. <sup>2</sup> *<sup>p</sup>*-values calculated using *<sup>t</sup>*-tests for continuous variables and McNemar–Bowker test for BMI category. <sup>3</sup> Fifty-seven participants completed the DEXA scan on a machine that provided additional body composition indices. Abbreviations: BMI, body mass index; FFMI, fat free mass index; VAT, visceral adipose tissue; SD, standard deviation.

#### *3.1. Prediabetes/T2D*

Positive change in HEI and DASH scores between the baseline and follow-up visits was associated with decreased risk for prediabetes/T2D at follow-up (Figure 1). A one-point increase in DASH score over the follow-up period was associated with a 64% (OR = 0.36, 95% CI: 0.17, 0.68) reduction in risk for prediabetes/T2D at follow-up, while a one-point

increase in HEI between visits was associated with a 9% decrease in risk (OR = 0.91, 95% CI: 0.85, 0.96). When scaled by standard deviation of diet index, improvements in DASH diet score reduced the risk for prediabetes/T2D by a greater extent than the HEI (OR = 0.14, 95% CI: 0.03, 0.46; OR = 0.83, 95% CI: 0.72, 0.93, respectively). In the cross-sectional analysis of the follow-up visit, higher HEI and DASH scores were also associated with reduced risk for prediabetes/T2D. At baseline, only MDS was associated with reduced risk for prediabetes/T2D.

**Figure 1.** Coefficient plots for the effects of diet quality on prediabetes. "Baseline" and "follow-up" values are the result of cross-sectional analyses of diet quality score and risk of prediabetes/T2D at the same visit. The value for "change between visits" represents the risk of prediabetes/T2D at the follow-up visit associated with change in diet score between the baseline and the follow-up visit. Effects are standardized to one standard deviation of exposure. Covariates: *Baseline and follow*-*up models.* HEI, DASH, and DII models adjusted for age, sex, ethnicity, physical activity, and parental education. MDS models adjusted for energy intake, age, ethnicity, physical activity, and parental education. *Change between visits models.* Baseline and follow-up model covariates + baseline diet score. Abbreviations: DASH: Dietary Approaches to Stop Hypertension; DII: Dietary Inflammatory index; HEI: Healthy Eating Index—2015; MDS: Mediterranean Diet Score.
