This study used a microlongitudinal design with 6 observations across 3 months.

Eligibility criteria are shown in Textbox 1.

As a US state, Connecticut is predominantly white, but its capital city of Hartford is largely Latino. Consistent with national data, Latinos in Connecticut have a significantly higher rate of diabetes than their white counterparts [22]. The study was led and coordinated by UConn Health and implemented in close partnership with the Hispanic Health Council (HHC) that used the study’s 2 bilingual and bicultural community health workers (CHWs). HHC is a community-based research, service, training, and advocacy organization headquartered in Hartford. This institutional partnership was built from a trusting relationship built over years of successful collaboration between UConn Health, Yale School of Public Health and HHC co-designing and implementing research studies using CHWs [23].

A comprehensive staff training program was developed and delivered for HHC recruiters and data collectors, including a core manual to support fidelity to the assessment procedures and study protocol. CITI-trained, bilingual and bicultural recruiters distributed a recruitment flyer across HHC programs serving the study’s target population. The recruitment flyer briefly described the study and informed intersted individuals to call the recruiters. Programs included the Santa Marquez Special Supplemental Nutrition Program for Women, Infant, and Children Program (WIC) Center, which serves to safeguard the health of low-income women, infants, and children up to age 5 who are at nutritional risk. The recruitment flyer was also posted in the HHC social media platforms (ie, Hispanic Health Council Facebook, Instagram, and Twitter pages). HHC staff also distributed flyers at health fairs, Hartford mobile markets, community health agencies, local convenience stores, and local daycare centers and it was mentioned on Spanish-language radio. Recruitment also took place at local food pantries that serve the target community. Foodshare is the largest antihunger organization in the Greater Hartford area, with a network of food pantries, meal programs, and mobile foodshare sites.

Eligibility screening was conducted by the CHW recruiters. Due to local rates of COVID-19 infection during the recruitment, most screening took place by phone, and some in a private room at the HHC. These sessions were scheduled through a virtual scheduling calendar accessible only to team personnel. The recruiters asked questions ascertaining the inclusion and exclusion criteria.

Demographic and clinical inclusion and exclusion criteria were self-reported. The 2-item Hunger Vital Sign FI screener was used to detect likely FI [17]. The 2 questions were “In the past 3 months, have you or other adults in your household worried whether your food would run out before you got money to buy more?” and “In the past 3 months, has the food you or other adults in your household bought just didn’t last and you didn’t have money to get more?” Response options were “yes,” or “no.” A response of “yes” to either question met study inclusion criteria for food insecurity [17].

The American Diabetes Association risk test was used to determine elevated risk for developing T2D. It includes seven self-report questions (total score of 0‐11) on age, gender, gestational diabetes mellitus, family history of diabetes, high blood pressure, physical activity, and obesity (based on BMI through a weight-height chart). The American Diabetes Association uses a cutoff of 5 or higher to indicate increased for diabetes. In this study, since all participants endorsed FI, which is itself a novel risk factor for diabetes, a score of ≥4 was required for inclusion [19,20].

The 4-item CAGE-AID (CAGE Adapted to Include Drugs) screener [24] was used to detect likely alcohol and substance abuse. A score of 2 is recommended as a positive screen for alcohol or substance abuse and indicates a potential problem that warrants further assessment by a health care professional; a score of 2 or higher excluded a recruit from participation.

After obtaining consent to participate and HIPAA (Health Insurance Portability and Accountability Act) authorization, physical assessments were conducted in the participant’s preferred language (English or Spanish), at the HHC. Participants had their weight, height, waist, and hip circumferences measured and were asked to provide fasting blood samples. Assessments were performed early morning (ie, before 9 am) to obtain fasting blood samples (30 ml). Study phlebotomist from HHC took blood samples from an antecubital vein, using sterile technique. Samples were then promptly transported by trained personnel, compliant with HIPAA and biohazard regulations, in an ice chest to UConn Health Clinical Research Center for processing and analysis. There, blood samples were aliquoted. One aliquot went to UConn Health Hospital Clinical Laboratory for immediate assessment of glucose. The remaining aliquots went to the UConn Health Clinical Research Center where they were centrifuged and stored at −80°C until analyzed for all other assays described below.

Following the baseline assessment, participants provided data at 2 time points per month for 3 months, in order to capture the disbursement cycle of SNAP benefits. Participants received cash compensation for participation, US $15 at each timepoint for each blood draw and US $10 for surveys, totaling US $150.

Because the temporal cycling of food insecurity over time is central to our research question, retention of this difficult-to-reach population was of utmost importance. Several community-engaged efforts were made to retain participants including the following: (1) Socializing the participant to the research endeavor for the community and describing the importance of each data collection timepoint; (2) Maintaining updated contact information for participants and a designated secondary contact person; (3) Speaking in the participant’s preferred language and fostering good rapport between CHWs and participants; (4) responding fully, genuinely, and in a timely manner to any participant concerns; (5) phone and text reminders for study appointments; and, (6) offering financial compensation for time spent on study activities.

Next, participants were scheduled for a baseline assessment. Due to COVID-19 social distancing mandates, the majority of the survey instruments were collected over the phone by trained bilingual interviewers who entered participant verbal responses into REDCap (Research Electronic Data Capture) [25] on large-screen electronic tablets. For all self-report surveys, response options referenced the preceding 7-days. Refer to Table 1 for data collection timeline.

Considering insulin resistance as our primary outcome, a previous study by our team provided data for a power analysis based on insulin resistance among persons at risk for T2D [50]. In that study, the SD for logHOMA-IR was approximately 0.3 and we expected to see a difference in logHOMA-IR of 0.4 between FI and food security, and a difference of 0.5. Our targeted sample size of n=100 would have 99% power to detect a main effect of FI, 96% power for mental distress and 87% power for the (FI × distress) interaction in a regression model using a 2-sided α of 0.05.

Summary statistics will be used initially to present the results from the categorical and continuous variables. Skewness, kurtosis, and potential outliers will be examined and if needed transformations or categorical variables will be created. The internal consistency of self-report measures will be assessed by Cronbach α.

To assess the relationship between monthly cycling of FI with diabetes risk markers, we will use repeated measures general linear mixed models. With 3 months of data, collected at weeks 2 and 4, depending on the distribution of for each outcome variable, the most appropriate distribution and link function will be chosen. In separate analyses it will be possible to divide the sample by frequency of benefits (ie, once vs twice monthly benefits) and include a dummy variable to evaluate the effects of frequency of benefits in the statistical models. It will also be possible to divide the sample into food secure, intermittent and persistent FI based on longitudinal studies using this approach [51].

To assess the hypothesis that psychological distress, diet, and eating disturbance mediate the relationship between FI and diabetes risk, we will perform a causal pathway analysis. We are interested in examining whether there is a change in our mediators over time, and if so, whether this change significantly leads to a change in diabetes risk markers. For this we will conduct dynamic mediation using latent change (difference) score mediation analysis [52,53].

For food frequency measures, linear regression models will compare food security status on household food availability. Spearman rank-order correlations of total inventoried availability of fruit and vegetable servings with the food frequency questionnaire will be conducted. We plan to use “GENMOD” or “MIXED” procedures in SAS (SAS Institute), or similar, for regression models and MPLUS software (Muthén & Muthén), or similar, for the mediation causal pathway analysis.

If a recruit was not excluded, then recruiters obtained verbal consent, collected contact information, and scheduled the first visit at HHC. During this first visit, the recruiters obtained written informed consent. Participants were given a signed, dated copy of the informed consent form for their records and the recruiter kept the original, signed, dated copy. All verbal and written study materials, including recruitment materials and informed consent forms, were provided to the participants in their preferred language (English or Spanish). Self-report measures and their response options were pilot tested for readability, understanding, logic, flow, cultural relevance, and face validity by 5 Spanish speakers and 5 English speakers at the Hispanic Health Council. The final assessment battery was deemed appropriate by the research team. This protocol was approved by the UConn Health Institutional Review Board Study 21-045OS-1 and HHC agreed for UConn’s Health Institutional Review Board to be the institutional review board of record. All procedures were conducted according to the Declaration of Helsinki. Incentives were calculated to compensate participants for the time spent in study-related activities but were not coercive in a sample that screens positive for food insecurity (refer to Assessment Procedures and Incentives).

The main findings reported here are that the study successfully recruited and retained a sample of the target population. There was excellent participation over time with 71% of the sample completing all 6 assessment timepoints. Participants were able to complete assessments using only their smart phones by cell phone service. Nonetheless, higher completion of assessments was associated with having internet and a tablet at home. This may reflect ease of connectivity (rather than access to connectivity) or a familiarity with technology that was advantageous. Other indicators of socioeconomic status (income and educational attainment) were not associated with number of assessments completed. As in other parts of the United States, digital inequity in Connecticut persists across race and ethnicity, income and education [54]. The surprisingly high rate of internet observed in this sample of Latinas who screened positive for food insecurity was likely due to assistance during COVID-19 pandemic for internet connection in the home. Future studies may achieve even higher rates of participation by providing tablets or internet service to participants, but the expense and ethical dilemmas of terminating internet service at the end of the study makes this approach potentially problematic.

This study was conducted during the COVID-19 pandemic. Social distancing requirements presented numerous challenges especially for recruitment, retention, and data collection. The study team remained flexible regarding format for assessments (telephone, Cisco WebEx, or in person when it was safe to do so) and strived to balance standardization with participant preference and evolving public health guidelines.

The innovative study protocol presented here will provide a comprehensive description of how FI is related to SNAP distribution, as well as, importantly, whether and to what degree FI cycling itself is related to risk for T2D. Furthermore, the study will provide early evidence for some putative mechanisms of this association, including contextual mechanisms such as household food availability, psychosocial mechanisms such as symptoms of depression and anxiety, and behavioral mechanisms such as distress-related binge eating. Each of these findings may suggest potential intervention targets.

Food insecurity is a known risk factor for onset of new T2D [55]. However, we postulate that food insecurity is not static; rather, it varies according to access to food which is largely driven by the reductions in the amount of money available from social assistance programs (including SNAP) over the course of the month in some instances with families finding themselves with hardly any economic resources left for food or other basic needs by the last week of the benefits monthly cycle. This cycling has been hypothesized in the literature [56] yet microlongitudinal data regarding FI cycling is not available, as far as we know (eg, [57]). Furthermore, we propose that this cycling is related not only to household food inventory, but also to distress and distress-related eating behaviors. Our group has documented that FI is associated cross-sectionally with higher symptoms of depression, anxiety, and stress hormones as well as lower sleep quality [41]. Yet it remains unknown whether FI and emotional distress covary over time.

Another novel contribution of this study is the procedure for the household food inventory. We taught participants to complete the inventory checklist themselves using REDCap [25]. Participants were instructed how to access and enter data through a link sent to them through email. Lessons learned regarding the feasibility of the method, and its fidelity, will be important to inform future studies using household food inventories with hard-to-reach populations, or studies when home visits by researchers are not possible due to pandemic, natural disaster, or other situation that disrupts home visits. We pilot tested items and omitted the items in the checklist that were not understood. However, were unable to conduct true fidelity checks to compare the participant responses with CHW responses, because the university protocols were not allowing home visits for research purposes. If found to be feasible and reliable, this method may also be more efficient for research teams and perhaps preferable for participants.

There are several limitations to this study. There is no experimental manipulation, so, whereas the longitudinal design can establish temporal precedence, no conclusions about causation can be drawn. This study was conducted during the COVID-19 pandemic. SNAP benefits since early 2020 have been affected by temporary pandemic-related benefit increases; an adjustment to the Thrifty Food Plan, upon which SNAP benefits are based, and higher-than-normal cost-of-living adjustments to reflect high food price inflation. Thus, findings may be influenced by changes to SNAP benefits during COVID-19 pandemic. COVID-19 pandemic also delayed and then paused the study, thus hindering recruitment. Data collection therefore took longer than anticipated and we fell short of our enrollment target by 13 participants. Future research with larger sample sizes should explore interactions between the exposure (FI) and mediators (psychological distress, diet, and eating disturbance) on outcomes (biomarkers of diabetes risk). Finally, the generalizability of our findings will need to be tested in other populations (eg, men, non-Latinos, or other parts of the country).

Limitations are balanced by the strengths of this study. The microlongitudinal design allows investigation of the temporal unfolding of covariance between FI and outcomes of interest over time. Home-based methods, particularly household food inventories, greatly improve contextual validity of assessments. The study affords the opportunity to confirm or assess the validity of certain measures in a Latina population, such as the Block Fat, Sugar, Fruit, and Vegetable Screener food frequency questionnaire. The population is high-risk and difficult-to-reach. Our procedures leverage the rigor and resources of a university medical center with the experience, knowledge, and reach of a community-based organization. Use of CHWs for recruitment, retention, and data collection improves trust and legitimacy in the target population. Multilevel bio-behavioral-psychosocial-contextual assessments are a particular strength. If our hypotheses are not refuted and this line of research goes on to establish causal pathways, interventions to disrupt the cycling of FI may prove beneficial for Latina health.

Results of this study may have potential policy implications. In 2022, 375,400 Connecticut residents, or 10% of the state population, received SNAP benefits [58]. More than 50% of recipients were in households with children and more than 37% were working families. Findings of this study may have implications for SNAP policy, particularly amount and timing of SNAP benefit distribution. Qualitative findings show that SNAP benefits are often inadequate for participants to purchase nutrient-rich foods that are sufficient for 1 month of household intake, particularly for participants at the higher end of the SNAP income distribution. Two studies have found that participant responses to food security surveys depend upon when in the SNAP cycle they are queried, with higher reports of FI when SNAP benefits are most salient [56,59,60]. This suggests considerable variability in food security within the month.

Inadequate benefit levels may contribute to cyclical eating patterns, where participants increase food consumption when benefits are received and reduce food consumption when benefits are depleted [14]. It may also lead to the purchase of quite unhealthy ultraprocessed foods and sugar sweetened beverages that have a long shelf-life, rather than fresh fruits and vegetables that are more likely to spoil over the month. In another study, individuals were not receiving benefits at baseline and then received 3 months of benefits. At baseline, food expenditures were not cyclical and were not related to impulsivity. However, total food expenditures became cyclical after households started receiving benefits, with more impulsive shoppers demonstrating a more accentuated benefit cycle [15]. The current monthly distribution of benefits could be replaced with a biweekly distribution to address cyclical food security, food expenditures, and eating patterns. Such a disbursement would provide SNAP recipients with greater flexibility.