The SNAP Cycle and Diabetes Management During a One-Time Change in Disbursement Schedule

For persons with diabetes, avoidance of hyperglycemic (high blood glucose) and hypoglycemic (low blood glucose) episodes is an indicator of optimal diabetes management, which is achieved through a balanced diet (1) and medication to support healthy blood glucose levels (2)—a balance that is particularly challenging for individuals below the poverty line. Suboptimal diabetes management is linked with food insecurity—that is, inability to afford enough food or lack of confidence in the ability to do so (3). In one study, 43% of surveyed households with food insecurity attributed hypoglycemic episodes to difficulty affording food (4). Further, individuals living in food insecure households had lower self-efficacy, lower blood glucose monitoring adherence, and more emergency department visits for hypoglycemia (5). Examinations of these blood glucose outcomes can inform policymakers about potential effects of an unexpected change in the availability of food resources.

Every year >40 million low-income households rely on the U.S. Department of Agriculture’s Supplemental Nutrition Assistance Program (SNAP), the country’s largest food assistance program, to pay for food. Strong evidence supports the program’s effectiveness in decreasing food insecurity (6–9) and self-assessed health (10).

Previous studies have examined the extent to which the once-a-month SNAP benefit disbursement (the “SNAP month”) results in a monthly “SNAP cycle” of food consumption. Most SNAP benefits are spent close to the date of receipt (11), and spending on food and caloric intake both decrease across the SNAP month (12). In addition to food consumption, the SNAP cycle has been linked with health and social behaviors, including drunk driving fatalities (decrease on first day of the month) (13) and grocery store theft (higher at the end of the benefit cycle) (14).

SNAP participants have higher diabetes mortality than both SNAP-eligible and SNAP-ineligible nonparticipants (15). However, spillover income benefits have been shown for SNAP participants. Specifically, older SNAP participants with diabetes are less likely to engage in cost-related medication nonadherence than similar nonparticipants, with the effect being greater for those who are food insecure or threatened by hunger (16,17). Therefore, the population used in this study, SNAP- and Medicaid-enrolled individuals, provides unique insight for policymakers as it is a population that is vulnerable to hyper- and hypoglycemic events and experiences a change in the benefit distribution schedule.

Management of chronic disease by low-income and SNAP households is particularly challenging toward the end of the month (18). Seligman et al. (19) found higher rates of hypoglycemia admissions in California at the end of the month for low-income populations (27% higher in the last week compared with the first week). However, Heflin et al. (20) did not identify cyclical hypoglycemia associated with the SNAP cycle in Missouri.

Events in winter 2018–2019 disrupted the normal cycle of SNAP benefit disbursements, which in turn affected the two health outcomes under study. Typically, SNAP benefits are disbursed by electronic benefit transfer once per month. However, as a result of the federal government shutdown that occurred from 22 December 2018 to 25 January 2019, SNAP benefits for February 2019 were disbursed to nearly all participating households on 20 January. (21,22) This one-time change to the disbursement schedule resulted in a shorter time since the previous disbursement (in early January) and a longer gap before the next benefit transfer on 1 March.

This study provides evidence of how the timing of food assistance disbursement affects two blood glucose outcomes that inform us about food sufficiency, consumption, and management of diabetes—hyperglycemia and hypoglycemia. To understand how this change in SNAP disbursement timings may affect blood glucose outcomes, we exploited the one-time change to disbursement timing in 2019 resulting from the federal government shutdown. We compared hyperglycemia and hypoglycemia encounters across the month during the 2 months of “treatment”— January and February 2019—to the trends in a typical month by using a fixed-effects model controlling for week, month, year, and county effects as well as individual characteristics.

We used administrative data from the West Virginia Department of Health and Human Resources, including both Medicaid claims and SNAP disbursement data. Supplementary Appendix A includes an in-depth description of these data. We limited our sample of West Virginia residents to those individuals aged 18–64 enrolled in both SNAP and Medicaid who had a previous diagnosis of diabetes as identified by a diabetes-associated claim in their records within the past 12 months. Because we were unable to observe Medicare claims, we limited our analysis to those not dual eligible and enrolled in both Medicaid and Medicare. Diabetes was defined using ICD-10 diagnosis codes from the Agency for Healthcare Research and Quality’s (AHRQ’s) Clinical Classification Software (see Supplementary Appendix B for a list of the codes used). In order to account for endogenous changes in participation due to the federal government shutdown in 2018–2019, we restricted the sample to individuals in households that were continuously enrolled in both programs for either of the two 6-month study periods (November 2017 to April 2018 or November 2018 to April 2019). Our analysis sample contained 819,913 person-weeks for 24,050 unique individuals. More than 60% of the observations and 42% of the unique individuals appeared in both periods. Most Medicaid recipients also receive SNAP benefits, since both programs reflect the need of an income-eligible population, and, further, in West Virginia, the application for public benefits includes both programs. Record-level data were aggregated to an analytic file with person-week observations.

We estimated whether the one-time change in benefit disbursement affected the rate of hyperglycemia or hypoglycemia encounters during that time by using a fixed-effects model with week-of-the-month dummy variables. The fixed-effects design allowed us to examine the difference in these diabetes-related encounters across the month during the treatment period, with relation to common monthly trends in November to April of that year and the previous year. We included in the model interactions of each week of the month within the 2 months of interest (i.e., January and February 2019). We controlled for seasonality (i.e., month of the year effects), year effects, and county effects in order to validly link between changes to blood glucose outcomes during the treatment period and the early disbursement itself. We also controlled for individual characteristics, including age, sex, race/ethnicity, and income, as well as clustering SEs by individual. Event study figures (presented in Supplementary Appendix C) supported the common trends assumption. However, there were statistically different coefficients for hyperglycemia. Although these differences were present primarily in the early part of the 10-week preperiod (prior to the start of the shutdown), our results should be interpreted with caution. For a detailed description of our empirical strategy and methods, see Supplementary Appendix C.

Figure 1 illustrates the patterns of SNAP disbursements from November to April 2017–2018 and 2018–2019. The graph indicates that distribution typically occurs evenly across the first 9 days of the month and shows a disruption of this pattern in early 2019. A spike in distribution occurred on 20 January 2019 and minimal distribution in early February 2019 due to the government shutdown. Further, due to the large gap in disbursement from 20 January through February, the state of West Virginia disbursed all March SNAP benefits on the 1st day of the month. This pattern was not reflected in data for the previous year, showing that it was driven by the shutdown-related early benefit disbursement.

Figures 2 and 3 illustrate regression results for the rates of hyperglycemia and hypoglycemia encounters, respectively, across the two time periods. A cyclical pattern existed for both outcomes. Compared with the first week of the month, where the bulk of SNAP distributions occur, hyperglycemia encounters in a typical month increase in weeks 2 and 3 and eventually decrease in week 4. Hypoglycemia, similarly, decreases in week 4.

Hyperglycemia encounters increased in mid-January, especially during the week of the early SNAP disbursement—week 3—when it increased by 0.86 percentage points on a constant that was not significantly different from zero. Compared with week 3 in a typical month of 3%, this 0.86 percentage point increase equates to 29%. Higher hyperglycemia continued into the first week of February, albeit at a lower rate, and finally decreased in the third week by 0.41 percentage points, representing a 13% change in hyperglycemia encounters from the mean of a typical month.

Hypoglycemia did not change across the month in January, but we found a small decrease of 0.06 percentage points in the third week of February. Compared with week 3 in a typical month of 0.17%, this represents a 35% change in hypoglycemia encounters.

Supplementary Appendix D includes additional results. We ran the analysis for all diabetes care encounters and found that the pattern of diabetes care encounters—typically higher midmonth—changed as a result of the shutdown’s changes to benefit disbursement. We also ran the analysis for visits related to appendicitis, kidney stones, and pharyngitis to test for placebo effects or negative control analysis, and we found no significant effect.

The main results presented in this study are for the entire sample of 24,049 individuals observed in our data. In Supplementary Appendix D we also restricted the analysis sample to the 10,119 individuals we observed in both time periods. The results for the full and restricted sample were equivalent, both in statistical significance and magnitude. As a result of churn induced through frequent recertifications for SNAP and Medicaid enrollment, the full sample is more generalizable because those individuals who maintain continuous enrollment on both programs year on year are a unique subsample of beneficiaries.

We found evidence of a SNAP cycle effect on hyperglycemia in a typical month during our study period. Compared with the first week of the month, hyperglycemia increases midmonth and decreases at the end of the month. This suggests that food consumption increases in weeks 2 and 3 and decreases in week 4. To our knowledge, this is the first study to consider the SNAP cycle effect on hyperglycemia.

Contrary to previous research by Heflin et al. (20), we also found evidence of a SNAP cycle effect on hypoglycemia in a typical month during our time period. And contrary to research by Seligman et al. (19), which found increased hypoglycemia at month’s end, hypoglycemia encounters decreased in the last week of the month. Since Heflin et al. (20) used only emergency department visits, we may be picking up on additional, less severe instances of hypoglycemia. This is supported by the fact that Seligman et al. (19) examined only admissions and therefore the most severe cases. We did not have sufficient emergency department visits for hypoglycemia to examine the emergency setting separately or to examine admissions only. Another explanation for the drop in hypoglycemia at the end of the month could be that individuals adjust their medication in expectation of reduced food resources and therefore successfully protect themselves from negative effects.

Our results suggest that early disbursement of SNAP funds impacted hyperglycemia, but not hypoglycemia, occurrences. When disbursement of SNAP benefits occurred for a second time in January, hyperglycemia events increased, suggesting that individuals consumed higher levels of food. In fact, given that hyperglycemia is defined as food consumption with insufficient insulin, it is likely that in many cases this is due to excess consumption (i.e., a “binge effect”). This increased intake persisted through early February. We did not find decreased hypoglycemia near the time of early disbursement, which is consistent with previous research finding no SNAP cycle effect on hypoglycemia (20).

By week 3 of February, we saw a larger-than-usual decrease in hyperglycemia cases, suggesting that any “binge effect” was eliminated by this time. The concurrent reduction in hypoglycemia cases is further evidence of changes in diabetes management related to this anomalous disbursement schedule. Managing diabetes is complex and distinct for each patient. Any change in blood glucose management reflects responses to changes in food and care resources. This is consistent with our proposed explanation of the typical monthly cycle—that expectations for decreased food resources allow individuals to adjust diabetes medication to prepare their bodies for decreased sugar intake. During the federal shutdown, Medicaid coverage was not disturbed, meaning that enrollees had consistent access to providers and drug coverage in both periods of our study.

Research on federal government shutdowns suggests that even short disruptions to financial and professional circumstances can impact individuals. Research on a 17-day shutdown in October 2013 included evidence about Washington DC-area crime (23), federal employee spending patterns (24,25), National Park gateway communities (26), a Weekly Economic Index, (27) and the role of social media(28). Existing evidence on the 2018–2019 shutdown suggests that firms become less compliant with environmental policies during shutdowns due to decreased regulatory capacity (29). A qualitative study of 26 SNAP participants in California found that the 2018–2019 shutdown increased existing uncertainty and stress about food for participants, who commented that SNAP benefits were inadequate for food security given other household expenses. Responses included diversion of money from other expenses toward food (30).

This study’s results on health outcomes and SNAP benefits help inform understanding of the potential effects of a major public program on households’ ability to purchase foods and manage a chronic disease. Further, they inform discussions regarding the frequency of benefits, which is related to timing. These findings may also support local agencies and organizations in targeting other types of support during times of the month in which there is more risk of poor nutrition outcomes, thus protecting the health of at-risk populations and reducing health care costs. Finally, understanding potential impacts of a government shutdown and its effects is an important policy question with broad implications for health and health equity.

Limitations to this research are primarily related to the generalizability of the sample. We included only individuals diagnosed with diabetes, and therefore, results could not be extrapolated to the whole population. However, we argue that blood glucose regulation among people with diabetes is a severe response to the more prevalent issue of food insecurity and that examining hyperglycemia and hypoglycemia encounters provides the opportunity to measure food insecurity in a uniquely biological way. Food insecurity measures not only inability to acquire and consume food but also stress over the financial uncertainty related to acquiring food. Further, low-income households have higher rates of diabetes (31), and therefore, these households are more representative than in the general population. Experiencing a hypoglycemic or hyperglycemic event severe enough to seek medical care occurs relatively infrequently. Hyperglycemic events are relatively more common than hypoglycemic events. In our analysis sample, we observed 24,130 hyperglycemic events and 1,302 hypoglycemic events. We are the first to study both in this setting.

The generalizability of these results is also limited due to its focus on West Virginia Medicaid and SNAP participants. This population is fairly homogenous and is made up almost exclusively of non-Hispanic White households. Further, the study included only diabetes management outcomes and did not include relationships between SNAP timing changes and management of other chronic diseases. Future research should be completed on more representative populations in order to generalize to the national level.

We also limited the study only to households with continuous enrollment within a study period. While this was necessary to avoid endogenous changes to participation, it also prevented some generalizability.

Further, since the changes to SNAP benefit disbursement affected nearly all SNAP participants, who make up the majority of Medicaid enrollees, we were unable to compare changes in outcomes with an unaffected subgroup in the same period. However, we did compare with a previous period, allowing us to control for expected cyclicality in outcomes.

Another set of limitations is inherent to the nature of retrospective cohort studies. Since we did not identify individuals with diabetes prospectively and instead based diabetes diagnosis on previous Medicaid records, there will be a degree of misclassification bias. Misclassification bias due to the administrative data is also possible. Since we did not expect misclassification of diabetes or diabetes management outcomes to be associated with the treatment—changes to SNAP disbursement timing—and since we did not compare the sample of patients with diabetes to patients without diabetes, we did not expect this to affect our results. Further, while we attempted to mitigate confounding with our fixed-effects method, there may be additional confounding not accounted for.

This study is subject to several limitations inherent to the administrative data sources used. For example, Medicaid claims data only include data on health care utilization that is directly reimbursed by Medicaid. There may have been cases during the study period where individuals experienced hypoglycemia or hyperglycemia but did not seek any medical treatment or sought treatment that was not billed to West Virginia Medicaid. These outcomes would be not be observed in our claims data.

Finally, prescription claims were not included in the data approved for use in this study, limiting our ability to capture what exactly drove the changes in diabetes management outcomes—diet, medication adherence, or a combination of the two. However, many of our findings were more suggestive of changes in food consumption rather than changes to medication adherence. For example, our findings of an increase in hyperglycemia after 20 January, interpreted through medication adherence, would suggest that the early release of SNAP funds decreased medication use. This is less likely than an increase in consumption since consumption of goods is likely to rise with more access to funds. Future work that incorporates prescription drug claims data with medical claims data may be able to address whether or not individuals adjust their use of diabetes medication in response to changes in SNAP availability and will aid in understanding further the mechanisms for changes in the monthly cycle that we document in this article.

This study contributes to the literature on cyclical consumption related to SNAP benefit disbursement. We found a statistically significant increase in hyperglycemia encounters in the 2nd and 3rd week of January 2019, and this was highest during the latter week—the week of early SNAP disbursement. This increase in hyperglycemia, in conjunction with no change to hypoglycemia during this time, offers evidence of an above-normal increase in consumption (i.e., a “binge”) around the time of disbursement. Given no previous literature relating SNAP disbursement to hyperglycemia, to our knowledge, this is a contribution to our understanding of the monthly cycle of caloric consumption across the month, especially as it relates to SNAP participants with diabetes. Both hyperglycemia and hypoglycemia encounters decreased in late February (i.e., the month with extended time between disbursements), suggesting that as households awaited more benefits, individuals in those households were no longer bingeing but continued to struggle to manage their diabetes. These findings provide an increased understanding of diabetes experiences beyond the most severe.

This study contributes to understanding how income-eligible households manage monthly food budgets. SNAP is particularly well-suited for responding to fluctuations in the economy and household circumstances (32), and it is useful to understand its role in supporting household food insecurity and nutrition during unexpected circumstances. We found that a shorter duration between receipt of benefits increased hyperglycemia—an indicator of excess consumption—among individuals with diabetes. We also found that a longer duration between benefit receipts increased hypoglycemia—a biological indicator of food insecurity. This can inform the ongoing discussions surrounding the number of disbursements scheduled per month as well, although with consideration that the benefit in this case was not split into smaller amounts.

The effects of any changes to SNAP disbursement are affected by both the timing and adequacy of benefits. Benefits that must be stretched to last are more likely to cause distress when disrupted than generous benefits. In the current study, the dollar amount of SNAP benefits received remained the same and the timing of disbursement changed. Responses to a change in timings contribute to an understanding of how humans respond to the “full wallet” effect. Our findings suggest an increased “binge” effect occurred when the February benefits were transferred earlier than normal due to the shutdown. This supports the “full wallet” concept, in that when money arrives it is spent (33). The government only changed the pattern of disbursement, not the benefit amount received; but given the potential interplay with benefit adequacy, we encourage other studies to explore changes in benefit amount to provide insight on benefit adequacy—especially in light of the fiscal year 2022 increase to SNAP benefits after an update to the Thrifty Food Plan.

We also add to previous literature on the unintended effect on consumption smoothing of federal government shutdowns for affected federal employees by providing evidence that recipients of a large federal safety net program, SNAP, are impacted similarly. SNAP is a safety net program, and its participants are unlikely to have savings, other assets, or access to credit to cover their needs when faced with an income shock (34). The federal shutdown and subsequent change in the distribution schedule of SNAP benefits was an income shock to all SNAP households, and unintended consequences could result due to this change in timing. Since SNAP benefits are in-kind transfers for the purchase of food, we examined the subgroup of SNAP participants with diabetes for specific health outcomes that could directly be influenced by food consumption. Future work using prescription claim and/or food intake data could further inform the mechanisms for the blood glucose changes we found.

While we examined the relationship between changes to SNAP benefit distribution timing and patterns of hyper- and hypoglycemic events, there are many other social safety net programs whose benefit distribution timing may be impacted by government shutdowns. It is likely other health outcomes, such as mental health and stress, could be impacted for the larger SNAP population. Additionally, the federal government shutdown meant not only SNAP payments were altered unexpectedly but other social safety net program benefit transfers were impacted as well. Understanding the spillover effects of federal government shutdowns facilitates policymaker decision making on the tradeoffs of future shutdowns with legislative priorities. Further work can delve deeper in exploring these concepts.

Acknowledgments. The authors would like to acknowledge the support of the West Virginia Department of Health and Human Services in providing the data for this study.

Funding. This work was funded in part by National Institutes of Health National Cancer Institute grant numbers T32CA057699 and F31CA232324.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. S.K.Y. and A.A. conceptualized the study design, conducted and interpreted the analyses, and drafted the manuscript, with input from N.P. L.A. and N.P. provided access to the data and manuscript review. N.P. additionally curated and prepared the data for analysis. All authors approved the final version of the article. S.K.Y. and A.A. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.