Household-level food insecurity is a major public health concern in the USA. Household food insecurity is defined as the absence of sufficient, reliable access to food due to a lack of money and/or resources(Reference Gundersen, Kreider and Pepper1). An increasingly higher number of Americans are food-insecure. In the USA in 2019, 13·7 million people, or 10·5 % of all households, lived in food-insecure households, with 4·1 % of those being very low food-secure. Nearly 7 % of households with an older adult aged 65 years or older were food-insecure, and 7·2 % of households with an older adult living alone were food-insecure(Reference Coleman-Jensen, Gregory and Singh2). The COVID-19 pandemic disrupted income, employment and overall stability to a magnitude not seen in recent history, and food insecurity has risen substantially as a result. Estimates from October 2020 projected that rates were 4·1 percentage points higher than they were in 2018 for adults and nearly 5 % points higher for children resulting in 50·4 million food-insecure individuals(3). As for older adults in particular, a survey from July 2020 indicated that the Meals on Wheels ‘programs …(were) serving an average of 77 % more meals and 47 % more seniors than they were March 1, (2020)’(4).
Food insecurity is a critical public health concern, as it is known to have detrimental short-term and long-term health consequences. Food insecurity is associated with poor physical and mental health outcomes, and food-insecure people face significant unmet needs for chronic disease prevention(Reference Pruitt, Leonard and Xuan5). Food insecurity is also associated with several chronic diseases including diabetes, depression, high blood pressure, CHD and chronic kidney disease and is associated with substantially higher healthcare costs(Reference Berkowitz, Basu and Gundersen6).
One relatively understudied group in the published literature(Reference Gundersen, Kreider and Pepper1,Reference Bruening, Dinour and Chavez7–Reference Weaver and Fasel12) facing food insecurity are older adults or the senior population. This is problematic given the well-known and severe consequences associated with food insecurity among seniors. Seniors are especially vulnerable given the increased risk for acute and chronic health conditions. For example, food-insecure seniors are 91 % more likely to have asthma, 64 % more likely to be diabetic and 57 % more likely to have congestive heart failure(Reference Gundersen and Ziliak13). Additionally, a large percentage of seniors live on a fixed income and are often forced to make spending trade-offs(3). In other words, they are forced to choose between paying for food and paying for other necessities such as housing and/or transportation. The population of seniors is expected to grow as people continue to live longer. For example, the 85 and older population is expected to see a 123 % increase by 2040(14). Of the current senior population in the USA, 7·3 %, or 5·3 million, were estimated to be food-insecure in 2018(Reference Ziliak and Gundersen15).
A better understanding of the factors associated with senior food insecurity is key to understanding senior-specific needs to develop targeted interventions and ultimately lower the prevalence and the incidence of food insecurity. To our knowledge, no study has yet to systematically examine the published literature to identify associated factors of senior food insecurity in the USA. The purpose of this study is to systematically review the literature and summarise the factors associated with senior food insecurity in the USA.
Methods
Search Strategy
A search was conducted in five electronic databases to identify articles that examined food insecurity and its correlates among older adults in the USA. The databases included PubMed, Scopus, Web of science, EconLit and JSTOR. In order to conduct the search, the following MeSH terms were used: Senior OR old* adults OR elderly OR ageing adults OR aged AND ‘food insecurity’ OR ‘food security’ AND ‘United States’.
Inclusion and exclusion criteria
Studies included in this review were those that assessed food security, and its correlates specifically among people aged 60 years and older were peer-reviewed and published in English, conducted in the USA, and published between January 2005 and September 2019. Other inclusion criteria included studies that assessed food insecurity as the dependent or independent variable of interest. Studies excluded from this review were those that were published before 2005, did not examine food security rates for people aged 60 years and older explicitly, or were conducted outside of the USA. Qualitative studies were also excluded from this review.
Data screening and extraction
All articles resulting from the five-database search were exported into the reference management software RefWorks, and duplicates were identified and removed. Data screening was conducted in two steps. In step 1, articles’ titles and abstracts were screened for eligibility. Titles and abstracts that met the inclusion criteria were moved to step 2. The full texts of eligible articles from step 1 were screened in step 2 to assess their adherence to the inclusion criteria. The screening process was conducted by three researchers to ensure quality and accuracy. First, two researchers independently reviewed titles and abstracts. Any discordances were then resolved by a third researcher. Second, two researchers independently examined the full texts of articles that were ‘screened in’ in step 1. Any discordances were again resolved by a third researcher. A Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) diagram summarising the results from the screening process is found in Fig. 1. Data of interest was extracted from all eligible articles by one researcher and reviewed for accuracy by a second researcher. Data extracted included the study authors, location, design, population, results, type of dataset and reported limitations. See full results in Table 1.
USDA, United States Department of Agriculture; FI, food-insecure; SNAP, Supplementary Nutrition Assistance Program; ED, emergency department.
Quality assessment
The quality of the included articles was assessed using the quality assessment tool developed by the Effective Public Health Practice Project (EPHPP). This tool is designed to evaluate the reliability, validity and biases of quantitative studies(16). The quality of the included articles was separately assessed by two researchers to ensure accuracy. These two researchers then compared their results and came to agreement on any discrepancies. Studies were rated as strong, moderate or weak based on criteria established by EPHPP for components including Selection Bias, Study Design, Confounders, Blinding, Data Collection Method, and Withdrawal and Dropouts. Table 2 presents a summary of the quality assessment of included studies.
1, Strong; 2, Moderate; 3, Weak; NA, Not Applicable.
Results
The search of the five electronic databases yielded 3579 potential articles. After removing exact duplicates, 3435 articles remained to be evaluated in step 1. A total of 315 articles met the inclusion criteria after their titles and abstracts were screened. The full text of these articles were screened in step 2. Of the 315 full-text articles that were screened, 295 articles were eliminated and 20 articles were retained for this review (see Fig. 1).
Quality assessment
Of the twenty studies included in this review, twelve were rated as Moderate, four as Strong and four as Weak. Referring specifically to the twenty studies included here, the four studies classified as Strong obtained such a rating in at least one of the quality dimensions with no score lower than 2, with a score of 1 denoting Strong, a score of 2 denoting Moderate and a score of 3 denoting Weak. The twelve studies indicated as Moderate had a quality rating of 3 in at least, but not more than, one of the quality dimensions. Lastly, the remaining four studies designated as Weak had multiple quality ratings of one along multiple dimensions of quality. See Table 2 for more detailed results.
Description of included studies (Table 1)
Fourteen of the studies included in this review were cross-sectional(Reference Afulani, Herman and Coleman-Jensen17–Reference Redmond, Dong and Goetz30); and three were longitudinal studies(Reference Lee, Johnson and Brown31–Reference Schroeder, Zeng and Sterrett33). There was one retrospective study(Reference Steiner, Stenmark and Sterrett34), one exploratory study(Reference Duerr35), and a study that was both cross-sectional and longitudinal(Reference Bhargava and Lee36). Most of the included studies were conducted using data from one state (n 12). Six studies were conducted in Georgia(Reference Bengle, Sinnett and Johnson18,Reference Brewer, Catlett and Porter19,Reference Myles, Porter Starr and Johnson29,Reference Lee, Johnson and Brown31,Reference Sattler and Lee32,Reference Bhargava and Lee36) , two in Colorado(Reference Schroeder, Zeng and Sterrett33,Reference Steiner, Stenmark and Sterrett34) , one in Florida(Reference Kihlström, Burris and Dobbins28), one in Indiana(Reference Duerr35), one in North Carolina(Reference Johnson, Sharkey and Dean27) and one in Ohio(Reference Holben, Barnett and Holcomb25). The remaining eight studies analysed national datasets(Reference Afulani, Herman and Coleman-Jensen17,Reference Brooks, Petersen and Titus20–Reference Hernandez, Reesor and Murillo24,Reference Jackson, Branscum and Tang26,Reference Redmond, Dong and Goetz30) .
Overall, 40 % of the included studies used a national dataset(Reference Afulani, Herman and Coleman-Jensen17,Reference Brooks, Petersen and Titus20–Reference Hernandez, Reesor and Murillo24,Reference Jackson, Branscum and Tang26,Reference Redmond, Dong and Goetz30) , 25 % used a state dataset(Reference Bengle, Sinnett and Johnson18,Reference Johnson, Sharkey and Dean27,Reference Lee, Johnson and Brown31,Reference Sattler and Lee32,Reference Bhargava and Lee36) and 35 % used a local dataset(Reference Brewer, Catlett and Porter19,Reference Holben, Barnett and Holcomb25,Reference Kihlström, Burris and Dobbins28,Reference Myles, Porter Starr and Johnson29,Reference Schroeder, Zeng and Sterrett33–Reference Duerr35) . Of the state datasets, four were data from the Georgia Advanced Performance Outcomes Measures Project –6 (GA Advanced POMP6)(Reference Bengle, Sinnett and Johnson18,Reference Lee, Johnson and Brown31,Reference Sattler and Lee32,Reference Bhargava and Lee36) , one was from the Nutrition and Function Study (NAFS)(Reference Johnson, Sharkey and Dean27), and one was from state data linked to the Centers for Medicare and Medicaid Services (CMS) data(Reference Bhargava and Lee36). National datasets included data from the National Health and Nutrition Examination Survey (NHANES)(Reference Brooks, Petersen and Titus20,Reference Frith and Loprinzi22,Reference Goldberg and Mawn23,Reference Redmond, Dong and Goetz30) , the National Health Interview Survey (NHIS)(Reference Afulani, Herman and Coleman-Jensen17,Reference Hernandez, Reesor and Murillo24) , and the Health and Retirement Study (HRS)(Reference Brostow, Gunzburger and Abbate21). Studies included adults aged 60+ years (n 14) or adults aged 65+ years (n 6); the mean age ranged from 69·8 to 78·2 years. To measure food insecurity status and/or severity, the majority of studies used one of the United States Department of Agriculture (USDA) Food Security Survey Modules (18 item = 2(Reference Frith and Loprinzi22), 10 item = 7(Reference Afulani, Herman and Coleman-Jensen17) and 6 item = 8(Reference Bengle, Sinnett and Johnson18,Reference Brewer, Catlett and Porter19,Reference Brostow, Gunzburger and Abbate21,Reference Kihlström, Burris and Dobbins28,Reference Myles, Porter Starr and Johnson29,Reference Lee, Johnson and Brown31,Reference Sattler and Lee32,Reference Bhargava and Lee36) ), two studies used a single screening question from the Nutrition Screening Initiative(Reference Schroeder, Zeng and Sterrett33,Reference Steiner, Stenmark and Sterrett34) , and one used a four-question survey(Reference Johnson, Sharkey and Dean27). Note, the four studies mentioned above using the GA Advanced POMP6 data all analyse the same sample of respondents(Reference Bengle, Sinnett and Johnson18,Reference Lee, Johnson and Brown31,Reference Sattler and Lee32,Reference Bhargava and Lee36) . The exception is that in one of the studies(Reference Bhargava and Lee36) the authors further match the sample of respondents with the CMS data resulting in a smaller analytic sample relative to the other three studies using the GA Advanced POMP6 data (n 957 v. n 1594).
The sample size in the studies greatly varied, with below 500 older adults in five studies(Reference Holben, Barnett and Holcomb25,Reference Johnson, Sharkey and Dean27–Reference Myles, Porter Starr and Johnson29,Reference Duerr35) , 500 to 1000 older adults in five studies(Reference Bengle, Sinnett and Johnson18,Reference Brewer, Catlett and Porter19,Reference Lee, Johnson and Brown31,Reference Sattler and Lee32,Reference Bhargava and Lee36) , 1500 to 2500 older adults in three studies(Reference Frith and Loprinzi22,Reference Goldberg and Mawn23,Reference Redmond, Dong and Goetz30) , 2501 to 5000 in two studies(Reference Brostow, Gunzburger and Abbate21,Reference Schroeder, Zeng and Sterrett33) , 5001 to 10 000 older adults in three studies(Reference Brooks, Petersen and Titus20,Reference Hernandez, Reesor and Murillo24,Reference Jackson, Branscum and Tang26) and greater than 10 000 older adults in two studies(Reference Afulani, Herman and Coleman-Jensen17,Reference Steiner, Stenmark and Sterrett34) . In addition, a cross-sectional study design was the main limitation reported by 55 % of the studies(Reference Afulani, Herman and Coleman-Jensen17,Reference Bengle, Sinnett and Johnson18,Reference Brooks, Petersen and Titus20,Reference Frith and Loprinzi22,Reference Goldberg and Mawn23,Reference Jackson, Branscum and Tang26–Reference Redmond, Dong and Goetz30,Reference Bhargava and Lee36) . Other reported limitations were self-reported data in 30 % of studies(Reference Afulani, Herman and Coleman-Jensen17,Reference Bengle, Sinnett and Johnson18,Reference Brooks, Petersen and Titus20,Reference Brostow, Gunzburger and Abbate21,Reference Hernandez, Reesor and Murillo24,Reference Redmond, Dong and Goetz30) , selection bias reported by 20 % of studies(Reference Bengle, Sinnett and Johnson18,Reference Lee, Johnson and Brown31,Reference Sattler and Lee32,Reference Steiner, Stenmark and Sterrett34) and non-generalisable results reported by 15 % of the studies(Reference Brostow, Gunzburger and Abbate21,Reference Myles, Porter Starr and Johnson29,Reference Steiner, Stenmark and Sterrett34) .
Outcomes
Table 1 contains a summary of findings from all twenty studies included in this review. Food-insecure individuals were more likely to be younger(Reference Bengle, Sinnett and Johnson18,Reference Brewer, Catlett and Porter19,Reference Jackson, Branscum and Tang26,Reference Lee, Johnson and Brown31,Reference Duerr35,Reference Bhargava and Lee36) , less educated(Reference Bengle, Sinnett and Johnson18,Reference Jackson, Branscum and Tang26,Reference Lee, Johnson and Brown31,Reference Schroeder, Zeng and Sterrett33,Reference Duerr35,Reference Bhargava and Lee36) , Black or African American(Reference Bengle, Sinnett and Johnson18,Reference Brewer, Catlett and Porter19,Reference Lee, Johnson and Brown31,Reference Steiner, Stenmark and Sterrett34) , female(Reference Jackson, Branscum and Tang26,Reference Schroeder, Zeng and Sterrett33,Reference Duerr35) , a current smoker(Reference Jackson, Branscum and Tang26,Reference Schroeder, Zeng and Sterrett33) and low income(Reference Bengle, Sinnett and Johnson18,Reference Jackson, Branscum and Tang26) . Food-insecure individuals were also more likely to self-report fair to poor health and have chronic conditions(Reference Bengle, Sinnett and Johnson18) and to report three or more chronic diseases(Reference Jackson, Branscum and Tang26). Moreover, individuals having co-morbidities, higher A1c, lower perceived quality of life, geriatric conditions and those taking diabetes medication were more likely to be food-insecure(Reference Schroeder, Zeng and Sterrett33). In addition, ‘… non-White race, history of a heart condition, preventative health behaviours, and especially depression and self-report of a psychiatric diagnosis were all associated with increased odds of being food-insecure’(Reference Brostow, Gunzburger and Abbate21). Being non-married, non-White, having lower educational attainment, being depressed, not having financial help and lacking insurance coverage were negatively associated with being food-secure(Reference Goldberg and Mawn23), and ‘marginal, low, or very low food security (was) associated with increased odds of having peripheral arterial disease …’(Reference Redmond, Dong and Goetz30).
Food insecurity and government assistance programmes
Ever receiving the Supplemental Nutrition Assistance Program (SNAP), formerly known as food stamp benefits, was associated with food insecurity in two studies(Reference Goldberg and Mawn23,Reference Lee, Johnson and Brown31) . Moreover, individuals on the waitlist for the Older Americans Act Nutrition Program (OAANP) were more likely to be persistently food-insecure than current participants, and participating in either meal delivery or congregate meals contributed to achieving food security(Reference Lee, Johnson and Brown31). However, results related to the impact of SNAP on food insecurity need to be viewed cautiously given the endogenous and misreported nature of SNAP participation(Reference Kreider, Pepper and Gundersen37). In addition, individuals that were eligible for both Medicaid and Medicare were more likely to be food-insecure(Reference Schroeder, Zeng and Sterrett33), and individuals who had Medicaid insurance were more likely to be food-insecure(Reference Steiner, Stenmark and Sterrett34).
Food insecurity and weight status
Food-insecure individuals were more likely to be obese(Reference Jackson, Branscum and Tang26), have a higher BMI(Reference Brewer, Catlett and Porter19,Reference Schroeder, Zeng and Sterrett33) , and waist circumference, and have arthritis, joint pain, and weight-related disability(Reference Brewer, Catlett and Porter19). However, Brostow et al. (2019) found that being overweight or obese was not associated with increased odds of food insecurity(Reference Brostow, Gunzburger and Abbate21). Furthermore, Hernandez et al. (2017) found that food insecurity was not associated with weight status in women, and ‘food-insecure men had 42 % lower odds of being overweight compared with normal weight and 41 % lower odds of being overweight or obese compared with normal weight…’(Reference Hernandez, Reesor and Murillo24).
Food insecurity and cost-related medication use and healthcare utilisation
One study found a ‘…dose-response relationship between (food insecurity) and cost-related medication underuse (CRMU) …behaviors’ of foregoing or taking less medication and delaying refills to save money, inability to afford medication and asking a prescriber for a lower cost medication(Reference Afulani, Herman and Coleman-Jensen17). Another study concluded that individuals who practiced cost-related medication non-adherence were more likely to respond affirmative to questions indicating food insecurity, and ‘… food-insecure individuals were approximately 2·95 times …more likely to report (practicing cost related medication non-adherence)’(Reference Bengle, Sinnett and Johnson18). In one study, researchers found that individuals who were persistently food-insecure and those who became food-insecure were more likely to practice medication non-adherence(Reference Sattler and Lee32). In addition, Bhargava and Lee reported that there was no significant difference in healthcare utilisation by food security status(Reference Bhargava and Lee36).
Food insecurity and mental health
Food insecurity was associated with depression in four studies(Reference Brooks, Petersen and Titus20,Reference Brostow, Gunzburger and Abbate21,Reference Goldberg and Mawn23,Reference Johnson, Sharkey and Dean27) . Johnson et al. (2011) found that individuals who were food-insecure ‘…were almost five times as likely to report depressive symptoms compared to those who were food secure’(Reference Johnson, Sharkey and Dean27). Food insecurity was also associated with a self-reported psychiatric diagnosis(Reference Brostow, Gunzburger and Abbate21). One study concluded that ‘individuals who were marginally food secure, food insecure without hunger and food insecure with hunger had significantly lower cognitive function ….’(Reference Frith and Loprinzi22), and another study found that food insecurity was associated with cognitive restraint after controlling for confounding variables(Reference Myles, Porter Starr and Johnson29).
Food insecurity and physical health
Physical functioning limitations increased as food insecurity increased(Reference Holben, Barnett and Holcomb25,Reference Jackson, Branscum and Tang26) . Moreover, all eight domain scores from a frequently used health survey measuring quality of life (SF-36) including physical functioning, physical role limitations, bodily pain, general health perceptions, energy/vitality, social functioning, emotional role limitations and mental health were associated with severity of food insecurity(Reference Holben, Barnett and Holcomb25). Jackson et al. (2019) found that the odds of food insecurity were greater for those with physical functioning limitations and more than three chronic diseases(Reference Jackson, Branscum and Tang26). Another study found that food-insecure individuals were more likely to report ≥ 14 physically unhealthy days and ≥ 14 d with activity limitations(Reference Kihlström, Burris and Dobbins28).
With respect to the relationships noted above, no discernable patterns emerge whereby studies classified as strong find one relationship relative to those classified as moderate or weak finding another. Most of the associations, in terms of the direction, are consistent across the studies, which individually vary in quality. The one exception is the relationship between food security and obesity/weight status where there are divergent findings across studies. However, there appears to be no pattern as it relates to the quality of study and the direction of the documented relationship. Specifically, three studies respectively classified as strong, moderate and weak all find a positive association between food insecurity and weight, whereas two studies both classified as moderate find either no relationship or an inverse relationship between food insecurity and weight. See Table 2 for a further breakdown of study quality.
Discussion
Despite the high prevalence and the detrimental health and well-being effects of food insecurity among older adults, a limited number of studies over the past 15 years have assessed the associated factors of food insecurity in this population. Overall, this review uncovered that social determinants of health including education(Reference Bengle, Sinnett and Johnson18,Reference Jackson, Branscum and Tang26,Reference Lee, Johnson and Brown31,Reference Schroeder, Zeng and Sterrett33,Reference Duerr35,Reference Bhargava and Lee36) , race and ethnicity(Reference Bengle, Sinnett and Johnson18,Reference Brewer, Catlett and Porter19,Reference Lee, Johnson and Brown31,Reference Steiner, Stenmark and Sterrett34) , gender(Reference Jackson, Branscum and Tang26,Reference Schroeder, Zeng and Sterrett33,Reference Duerr35) , and income(Reference Bengle, Sinnett and Johnson18,Reference Jackson, Branscum and Tang26) were consistently associated with food insecurity. These results are consistent with previous findings of higher rates of food insecurity among lower-income older adults and those from racial or ethnic minorities(Reference Ziliak and Gundersen15). Many of the factors associated with food security in older adults are similar in the estimated direction of the relationship to those found in other adult age groups. Lower educational attainment, lower household income, female gender, having a disability and being non-White race/ethnicity are associated factors that have been consistently documented to have a negative association with food security for decades by the USDA through the annual Current Population Survey Food Security Supplement(38). Similarly, being a smoker(Reference Bergmans, Coughlin and Wilson39–Reference Kim-Mozeleski, Seligman and Yen42) and having poorer self-reported health(Reference Brown, Noonan and Nord43–Reference Weiser, Bangsberg and Kegeles46), chronic disease(Reference Weaver and Fasel12), poor mental health outcomes(Reference Bruening, Dinour and Chavez7,Reference Pourmotabbed, Moradi and Babaei11) , and medication non-adherence(Reference Silverman, Krieger and Kiefer45,Reference Kalichman, Washington and Grebler47–Reference Walker, Campbell and Egede49) are documented in the peer-reviewed literature to have a negative association with food security. Though limits in physical functioning are less documented in non-older adult populations, one well-established likely related factor is disability status(38,Reference Coleman-Jensen, Gregory and Singh50) . These results suggest that upstream systemic-level interventions, though difficult to implement, may be better suited to deal with food insecurity among the senior population.
In addition, ‘younger’ older adults were found to have(Reference Bengle, Sinnett and Johnson18,Reference Brewer, Catlett and Porter19,Reference Jackson, Branscum and Tang26,Reference Lee, Johnson and Brown31,Reference Duerr35,Reference Bhargava and Lee36) higher rates of food insecurity (age ranges from 60–64, 60–69, 60–74, 65–74 and 60–84 years). This is also consistent with other findings(Reference Ziliak and Gundersen15,Reference Madden, Shetty and Zhang51,Reference Ziliak and Gundersen52) . For example, in their report for Feeding America, Zilak & Gundersen (2020) uncovered that food insecurity rates among seniors aged 60–64 years were twice as high as seniors aged 80 years and older(Reference Ziliak and Gundersen15). This increased likelihood of being food-insecure may be explained, in part at least, by eligibility for Medicare and other safety net programmes that help to buffer resource limitations(Reference Madden, Shetty and Zhang51).
This review revealed an inconsistent relationship between food insecurity and weight status among older adults with some studies finding a link between food insecurity and obesity(Reference Jackson, Branscum and Tang26), higher BMI(Reference Brewer, Catlett and Porter19,Reference Schroeder, Zeng and Sterrett33) , and waist circumference(Reference Brewer, Catlett and Porter19), and other studies finding no association with being overweight or obese(Reference Brostow, Gunzburger and Abbate21). This is not surprising given our understanding that human behaviour is complex, and that there exist inherent statistical issues around measurement error in both food security and obesity(Reference Courtemanche, Pinkston and Stewart53,Reference Millimet and Roy54) . Specifically, the socioecological model postulates that health behaviour is influenced by factors at several levels including intrapersonal, community, organisation, government, industry and societal(Reference Bronfenbrenner55). This is a model that is frequently cited in obesity research(Reference Ohri-Vachaspati, DeLia and DeWeese56). The relationship between food insecurity and weight status is possibly bidirectional; it is possible that food insecurity preceded obesity for some and for others it followed. Additionally, obesity is likely to occur over the long term, and people are likely to ebb and flow in and out of food insecurity(Reference McDonough, Roy and Roychowdhury57). Further, measurement error related to assessing food security and/or obesity/weight status can introduce bias in widely used parametric estimators given the non-classical nature of such misclassification. This is a reasonable concern given the vague and somewhat arbitrary nature that food security is defined and measured by the USDA, the misreporting of food security status due to perceived stigma, and/or the inexact methods to measure BMI. Given the non-classical nature of such measurement error, the estimated relationship between food insecurity and weight/obesity can be wrong in terms of magnitude as well as in the sign of the relationship. Directly confronting such measurement error becomes extremely difficult, though progress has been made in the economics literature(Reference Nguimkeu, Denteh and Tchernis58).
This review found that food-insecure older adults are likely to make spending trade-offs including cost-related medication non-adherence(Reference Afulani, Herman and Coleman-Jensen17,Reference Bengle, Sinnett and Johnson18,Reference Sattler and Lee32) . These results may be linked to the rise in healthcare costs. On average, Medicare enrollees spend over $5000 out of pocket annually, including over $650 on prescription drugs(Reference Cubanski, Koma and Damico59). The price of prescription drugs is thought to be the driving force in the increasing cost(60). ‘Since 2001, prices on prescription drugs have grown at an average annual rate of about six percent as measured by the producer price index for pharmaceuticals – a much higher rate than general inflation’(60). In addition to increasing medication cost, most Medicare prescription drug plans have a coverage gap, also called the ‘donut hole’, which is a temporary limit on what the insurance plan can cover in terms of prescription drugs(61). While recent reforms have shifted the structure of this gap, it still leaves many seniors potentially paying higher out-of-pocket costs, dependent on the cost of their medications and the new cost share. Given the importance of medication adherence to maintaining health, policy-level interventions aimed at drug costs to help mitigate spending trade-offs are warranted.
Food insecurity was associated with depression(Reference Brooks, Petersen and Titus20,Reference Brostow, Gunzburger and Abbate21,Reference Goldberg and Mawn23,Reference Johnson, Sharkey and Dean27) , a self-reported psychiatric diagnosis(Reference Brostow, Gunzburger and Abbate21) and significantly lower cognitive function(Reference Frith and Loprinzi22). These results are consistent with recent findings. For example, Madden et al. (2020) reported that food-insecure seniors younger than 65 years of age were 2·65 times more likely to report depression, and seniors aged 65 years and older were 1·6 times more likely to report depression relative to food-secure seniors(Reference Madden, Shetty and Zhang51). The relationship between food insecurity and mental health can be bidirectional, where poor health increases financial strains and food insecurity, and financial strain and food insecurity may increase the risk of poor health. Additionally, mental wellness can affect one’s ability to attain and maintain employment/steady income. This relationship is likely to be bidirectional as well, where the hardships imposed by food insecurity may result in poor mental health outcomes(Reference Noonan, Corman and Reichman62). In a systematic review, Bruening et al. ‘suggest a bidirectional association whereby food insecurity increases the risk of poor emotional health, and poor emotional health increases the risk of food insecurity’(Reference Bruening, Dinour and Chavez7).
Most of the studies included in this review were cross-sectional in nature making it difficult to infer causality. In addition to the quality measures highlighted in Table 2 and the measurement error issues commented on earlier, readers should interpret results of the included studies cautiously given the bidirectional nature of how food insecurity and other measures of interest are determined. The consequence of estimating the effect of some independent variable (e.g. mental health) on a particular dependent variable (e.g. food insecurity) when such bidirectionality exists is the estimated effect being contaminated with simultaneity bias(Reference Wooldridge63). The reason for such bias stems from the failure of the assumption that the error term in regression-based models is uncorrelated with included model covariates. In addition to instrumental variables (IV) and partial identification methods using cross-sectional data, incorporating the dimension of time can potentially help in dealing with such endogeneity. With that said, if one is to incorporate lagged values as a means to avoid simultaneity, it should be done so in the context of using the lagged endogenous variable in an IV estimation strategy and only if the lagged regressor meets the criteria of being a valid exclusion restriction(Reference Reed64). Even so, few studies have examined the relationship between food insecurity and the associated factors that were found to be significant in this review over time. This highlights the need for more longitudinal studies that would allow researchers to employ panel data methods, including causal inference methods such as difference-in-differences, to, under a specific set of assumptions, tease out the causal relationship between food insecurity and its associated factors among older adults. Additionally, 40 % of the studies included in this review used a national dataset with representative samples of older adults in the USA, thus increasing the generalisability of the results. Similarly, a majority of the studies used state or multistate datasets consisting of a representative sample of the states’ older adult population. However, 80 % of the state datasets were from Georgia. Given the wide distribution of food insecurity rates by state, studies are warranted for other states and regions that have distinct characteristics.
The sample sizes in the studies included in this review were relatively large with half of the studies including 1500 participants or more. After conducting a quality assessment, 60 % of the studies were rated as moderate quality with many studies reporting several limitations including survey tool validity and reliability, self-reported data, and selection bias. However, some studies did not include a self-evaluation of the research or a clear list of limitations. Future studies must ameliorate quality-related factors in their studies and clearly discuss limitations so that others can properly interpret and potentially replicate findings.
This review systematically assessing the associated factors of food insecurity in the USA is subject to several limitations. The inclusion criteria limited this review to studies conducted in the USA and published in English, possibly excluding relevant studies conducted elsewhere and/or published in other languages. This limits the generalisability of this review to other countries and parts of the world. Future studies may consider expanding criteria to include more countries and articles published in other languages. While this study restricted the sample to adults aged 60+ years, there are still compositional differences among the study samples; thus, attention should be paid when making comparisons. Additionally, because we limited our research to peer-reviewed articles that were published between 2005 and 2019, we may have missed relevant findings that were published in non-peer-reviewed sources or those that were published outside of our inclusion dates. And while our selection process was well defined, it is possible that others doing the screening may have resulted in the inclusion of different articles. Further, each included study is subject to its own limitations and biases. Lastly, there is no discernable pattern related to the consistency of findings by the assessed quality of the included studies.
Overall, the correlates of food insecurity among older adults identified during this review are younger age(Reference Bengle, Sinnett and Johnson18,Reference Brewer, Catlett and Porter19,Reference Jackson, Branscum and Tang26,Reference Lee, Johnson and Brown31,Reference Bhargava and Lee36) , lower educational level(Reference Bengle, Sinnett and Johnson18,Reference Jackson, Branscum and Tang26,Reference Lee, Johnson and Brown31,Reference Schroeder, Zeng and Sterrett33,Reference Duerr35,Reference Bhargava and Lee36) , Black or African American race(Reference Bengle, Sinnett and Johnson18,Reference Brewer, Catlett and Porter19,Reference Lee, Johnson and Brown31,Reference Steiner, Stenmark and Sterrett34) , female gender(Reference Jackson, Branscum and Tang26,Reference Schroeder, Zeng and Sterrett33,Reference Duerr35) , being a current smoker(Reference Jackson, Branscum and Tang26,Reference Schroeder, Zeng and Sterrett33) , low-income(Reference Bengle, Sinnett and Johnson18,Reference Jackson, Branscum and Tang26) , fair to poor health status (self-reported), and having chronic conditions and other co-morbidities(Reference Bengle, Sinnett and Johnson18,Reference Jackson, Branscum and Tang26,Reference Schroeder, Zeng and Sterrett33) . In addition, depression(Reference Brooks, Petersen and Titus20,Reference Brostow, Gunzburger and Abbate21,Reference Goldberg and Mawn23,Reference Johnson, Sharkey and Dean27) , non-married status, lack of insurance coverage(Reference Goldberg and Mawn23), cost-related medication underuse(Reference Afulani, Herman and Coleman-Jensen17,Reference Bengle, Sinnett and Johnson18,Reference Sattler and Lee32) , lower cognitive functioning(Reference Frith and Loprinzi22) and physical functioning limitations(Reference Holben, Barnett and Holcomb25,Reference Jackson, Branscum and Tang26) were other significant correlates of food insecurity among older adults. Safety net programmes generally help to buffer some effects of food insecurity; however, individuals sometimes employ coping mechanisms that have the potential to exacerbate the issue, such as skipping or cutting medications and consuming lower nutrient foods. Future studies may want to employ a meta-analysis of such findings to provide a more precise estimate of the effects of food insecurity on the health and well-being of seniors. Public health interventions should be upstream and systemic to address the underlying determinants of food insecurity.
Acknowledgements
Acknowledgements: The authors would like to acknowledge and thank Dr Xan Goodman for her help with scoping and finalising the search terms and strategies. Financial support: The authors have no sources of support to disclose. Authorship: BA, C.C., and IM designed the research question and protocols, BA and A.G. conducted the research, BA, A.G., and C.C. screened and reviewed the articles, and BA, C.C., and IM wrote and edited the manuscript. All authors have read and approve the final manuscript. Ethics of human subject participation: NA.
Conflict of interest:
The authors have no conflicts of interest to disclose.