This was a secondary analysis of data from LiveWell, a cohort study to evaluate the impact of a new state Medicaid program to provide food and housing support for eligible patients (5). Participants were recruited between December 2019 and December 2020 from 5 community health centers affiliated with a large health system in Boston, Massachusetts. Eligible patients were 21–62 years old, had ≥ 2 health center visits in the prior 2 years, and spoke English or Spanish; 846 patients with Medicaid insurance were enrolled. This analysis included 188 participants with prediabetes or T2D at enrollment. Diagnoses were determined using the electronic health record (EHR) to identify an International Classification of Diseases-10 code, problem list diagnosis, use of T2D medication(s), or a laboratory value consistent with prediabetes or diabetes within 2 years prior to enrollment. When the diagnosis was unclear (i.e., EHR diagnosis of prediabetes or T2D but no prior abnormal A1c), a physician (K.D.G.) performed chart review to adjudicate the diagnosis.
Baseline and annual follow-up surveys collected data on gender, race/ethnicity (baseline only), marital status, number of dependents, household income, health insurance, height and weight, Patient Health Questionnaire-8 (6), Generalized Anxiety Disorder-7 (7), housing instability, financial stress, cost-related medication underuse (9), and FS status. FS was assessed with the USDA 10-item Adult FS Survey Module and was dichotomized (≤ 2 = FS; 3–10 = FI) (8). Participants who indicated FS at all 3 timepoints (baseline, 1-year, and 2-year follow-up) were coded as “persistently secure,” 1 or 2 timepoints as “intermittently insecure,” and no timepoints as “persistently insecure.
Primary outcomes were changes in Healthy Eating Index-2020 (HEI-20) scores, BMI, and A1c over 2 years, calculated as the value at year 2 minus the value at baseline for each outcome. HEI-20, a valid and reliable measure of dietary quality (10) that aligns with the Dietary Guidelines for Americans 2015–2020 (11), was calculated from 2 Automated Self-Administered 24-Hour dietary recalls collected at baseline and annual follow-up using the National Cancer Institute’s simple scoring algorithm (12). Scores range from 0 (least healthy) to 100 (most healthy). BMI (kg/m2) was calculated using self-reported weight and height from annual surveys because EHR data was less complete due to pandemic-related virtual visits. Hemoglobin A1c was ascertained using EHR-recorded laboratory data. We used the average A1c within 24 months preceding enrollment to represent baseline A1c, and average A1c within 12 months preceding the 2-year follow-up survey to represent year 2 A1c. We chose this strategy to capture average exposure to hyperglycemia leading up to enrollment and 2-year follow-up dates.
Baseline differences in participant characteristics by FS category (persistently secure, intermittently insecure, persistently insecure) were compared using non-parametric tests. Separate multivariate generalized linear models adjusted for age, gender, and ethnicity were used to test the association of FS category with change in each outcome. Those missing FS data at any of the 3 time points were excluded (N = 37). Participants who were missing HEI-20 (N = 3), BMI (N = 73), and A1c (N = 56) at baseline or year 2 were also excluded from the relevant models. To address missing data on BMI, a sensitivity analysis was conducted using available EHR weight data to impute missing self-reported weights. All analyses were run using SAS 9.4 (Cary, NC).