Food Insecurity in Israeli Elderly is Associated with Sociodemographic Characteristics, Disability, and Depression: Policy Implications

Abstract

Background

Food insecurity (FI) and poor health can turn into a vicious cycle with detrimental effects, especially in the elderly, however, few studies have examined the relationship between FI and health in this age group.

Objectives

We investigated associations of FI with physical and mental health and health behaviors among community-dwelling elderly.

Methods

We used nationally representative, cross-sectional data from the 2014–2015 Israel National Health and Nutrition Survey of the Elderly (Rav Mabat Zahav) on FI, sociodemographic characteristics, noncommunicable diseases (NCDs), disability, self-assessed physical, oral, and mental health for 1006 individuals aged ≥65 y.

Results

FI affected 12.3% of all households with elderly and was significantly higher among late immigrants and Arabs. Bivariate associations of FI with the number of NCDs, depression, disability in all 6 domains (vision, hearing, mobility, self-care, remembering, communication), poor self-assessed physical and oral health, chewing and swallowing problems, feelings of loneliness, insufficient physical activity, and smoking were significant (P < 0.05). In a multivariable logistic regression controlling for population group, household size, age and sex, FI was significantly associated (P < 0.05) with lack of formal education (OR: 6.26; 95% CI: 1.66, 23.65), being in the lowest (OR: 23.56; 95% CI: 3.71, 149.76) or second-lowest (OR: 16.75; 95% CI: 2.68, 104.52) per capita household income quartile, having one (OR: 2.11; 95% CI: 1.05, 4.23) or several disabilities (OR: 4.04; 95% CI: 1.72, 9.45), and having ever been diagnosed with depression (OR: 3.34; 95% CI: 1.35, 8.28).

Conclusions

FI is associated with physical and mental health problems, multiple disabilities, and loneliness among Israeli elderly. Providing income support could reduce FI, and subsidized congregate and home-delivered meal services could be expanded to meet the needs of elderly with disabilities and counter social isolation. Because low education, disability, and depression are particularly prevalent among the food insecure and vulnerable groups face language barriers, assistance with applications for these services should be increased.

Introduction

In industrialized countries, food insecurity (FI) is generally defined as “the limited or uncertain availability of nutritionally adequate and safe foods or limited or uncertain ability to acquire acceptable foods in socially acceptable ways” ([1], p. 6). To capture the complex and multidimensional phenomenon of FI resulting from financial resource constraints, the household food security scale was developed in the United States [1].

FI can have non-negligible proportions even in high-income countries [[2], [3], [4], [5], [6], [7], [8], [9], [10], [11], [12]]. As of 2019, 10.5% of US households were food insecure [2], and 6.8% of Canadian households were food insecure in 2011–2012 [3]. Using the same 18-item household food security scale and classification, Israel has consistently experienced higher rates of FI than the USA and Canada, ranging from 22% in 2003 [4] to 17.8% in 2016 [5]. In all 3 countries, elderly citizens have lower rates of FI than the general population [[2], [3], [4], [5]]. However, the elderly are at increased risk of social, functional, and medical problems that affect nutritional and health status [13]. Poor health and FI can easily turn into a vicious cycle that has detrimental effects [14].

The conceptual framework in Figure 1 illustrates the determinants and consequences of FI and highlights possible interactions between poor health and FI. Since elderly people are prone to functional impairments and ill health, they often have difficulty to physically access, prepare, and eat food, which may lead to inadequate diets [16, 17]. FI tends to aggravate this problem because it usually involves harmful compensatory behaviors: To maintain dietary energy intakes, a common strategy of the food insecure is to reduce the consumption of more expensive foods that are of high nutritional value, such as fruits, vegetables, meat, and dairy products, and to rely on low-cost, energy-dense, nutritionally poor, carbohydrate-rich foods, such as rice, pasta, bread, and sweets [4, [17], [18], [19]].

FIGURE 1.

Determinants and consequences of food insecurity. The light grey arrows denote pathways through which income, education, and health affect food insecurity directly or indirectly. The dark grey arrows show reverse causal effects through which food insecurity affects health directly or indirectly. The black arrow refers to the direct impact of poor health on dietary intakes. Some effects are long-term and occur over the life cycle, whereas others are short-term and can manifest within weeks or months. The direct influence of low income on health is not shown for the sake of clarity and because the main focus here is on food insecurity. Low incomes can pose health risks, even though they may not always cause food insecurity: Many elderly Israelis who were facing financial hardship reported difficulty with paying for heating, electricity, communication, transportation, medication, medical and dental care, visual and hearing aids, and they were also more likely to live in an apartment with maintenance problems, such as dampness in the winter, an inadequate heating system or insufficient hot water supply [15]. The impact of poor health on educational achievement is also not shown. Region of residence and the neighborhood food environment are not included as modifying factors because Israel is a small, highly developed and highly urbanized country with a modern infrastructure and a diversified food retail system that provides virtually universal access to healthy foods. Source: Authors, based on pathways and associations discussed in the literature [[1], [2], [3], [4], [5], [11], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25], [26], [27], [28], [29], [30], [31], [32], [33], [34], [35], [36], [37], [38], [39], [40], [41], [42], [43], [44], [45], [46], [47], [48], [49], [50], [51], [52], [53], [54], [55], [56], [57], [58], [59], [60], [61], [62], [63]].

A balanced, nutrient-dense diet is important to help the elderly in maintaining optimal levels of health and preventing or delaying the onset of disease [13]. Health promotion and disease prevention for the elderly will have to be expanded, considering that the number and proportion of elderly people will rise dramatically in the coming decades [64]. A solid understanding of the factors that are associated with FI among the elderly is essential to better tailor nutrition and health services to the needs of this growing population and to assist vulnerable older adults in achieving a satisfactory quality of life [14, 16].

Only a few studies outside the USA have systematically examined the risk factors of FI among the elderly. The present study contributes to the existing body of knowledge and extends the geographical focus to the Middle East. Other studies of FI among the elderly have relied on Activities of Daily Living (ADL) and Instrumental Activities of Daily Living (IADL) scores to measure functional impairments [16, [20], [21], [22], [23]]. This study provides new insights by using the measures of disability that were developed by the Washington Group on Disability Statistics for international comparisons [65]. In addition, a unique feature of the Israeli population—namely that the large majority of the elderly were born in other countries—allows to consider the long-term effects of immigration on food security.

Previous nationally representative studies among the general population in Israel reported the prevalence of FI among various subgroups of the population, including the elderly, but did not examine FI among the elderly in relation to sociodemographic and health indicators [4, 5]. A national study about the broader topic of the effects of financial hardship on the living conditions of the elderly in 2003 presented cross-tabulations of FI against sociodemographic characteristics and selected health indicators, but considered only a small set of health-related measures and did not examine associations with FI in a multivariable regression framework [15].

The aim of the present study is to comprehensively explore bivariate associations of FI with sociodemographic factors, physical and mental health, and health behaviors among Israeli elderly and to identify independent risk factors of FI for this age group by means of multivariable regressions, using the validated, 6-item short form of the USDA household food security scale. It is hoped that the findings will inform programs and policies that seek to prevent and mitigate FI among the elderly in Israel and beyond.

Subjects and Methods

Survey population and sampling

The second National Health and Nutrition Survey of the Elderly (Rav Mabat Zahav) was conducted between March 2014 and February 2015 by the Israel Ministry of Health in collaboration with the Central Bureau of Statistics (CBS). The sample of elderly citizens aged ≥65 y was drawn from the Population Registry of September 2013 and included elderly living in their homes and in institutions that are not for therapeutic purposes, such as assisted living facilities. Emigrants not living in Israel and new immigrants who immigrated to Israel <6 mo before the start of the survey were excluded.

The survey was cross-sectional and nationally representative and served multiple purposes. The target sample size (n = 1270) was set to achieve adequate precision for estimating the medians of calcium, iron, and saturated fat intakes and the percentages of subjects with recommended daily intakes of these nutrients, and power calculations were based on the distributions of intakes observed in the first National Health and Nutrition Survey of the Elderly in 2005–2006 (Mabat Zahav). We note that the aim for setting the target sample size was unrelated to the purpose of this study, since we refrained from analyzing nutrient intakes.

The sampling was carried out in 2 stages: In the first stage, a locality (a city or village) was selected. Large localities with >20,000 residents were selected with certainty. About one-tenth of the smaller localities from each district were randomly selected, with each district constituting a stratum. Bedouin tribes living outside of localities, nonrecognized localities, and very small localities with only a few dozen residents were excluded; further details on exclusions of specific localities have been provided elsewhere [66].

In the second stage, 1705 elderly were sampled from the selected localities, stratified by age group, sex, and sector (Jewish/Arab). Following the definition by the CBS, the Jewish sector includes Jews, Christians who are not Arabs, and people with no religious affiliation, and the Arab sector includes Muslims, Christians who are Arabs, and Druze. The sample was stratified by sector since the Jewish majority and the Arab minority differ in culture, religion, language, health outcomes, and socioeconomic status [67].

The number of completed interviews was 1039, resulting in a response rate of 60.9%: 59.6% for the Jewish sector elderly (JSE) and 77.8% for the Arab sector elderly (ASE). The actual response rate was lower than the response rate of at least 70% that was initially assumed, and consequently, the target sample size (n = 1270) was not achieved. This does not jeopardize our analysis, however, because the target sample size was set for a different objective that was not related to the specific purpose of this study, as explained above.

The survey was approved by the Helsinki Supreme Committee for Medical Experiments on Humans, which is a statutory, multidisciplinary committee that deals with the examination of proposals for genetic research on humans in Israel. This committee also functions as the Ethics Committee of the Ministry of Health, which is responsible for examining and approving research carried out by the employees of the main office of the Ministry of Health.

Each interviewee provided written informed consent: Before the beginning of the interview, participants were informed about the purpose and procedures of the survey and signed an informed consent form. With their signature, participants confirmed that they understood all the explanations about the survey and authorized the use of their data in anonymized form in published research. Respondents were not compensated for their participation; they agreed to be interviewed because they were convinced that the purpose of the survey was important.

The CBS implemented the survey on behalf of the Ministry of Health and was responsible for the security of identifiable data. The data were anonymized by the CBS at the point of data entry, and no personal details that might permit the identification of respondents were shared with the Ministry of Health.

Survey tools and procedure

Face-to-face interviews were conducted in the respondents’ homes by specially trained interviewers. The questionnaire, which has been described in detail elsewhere [66], was pretested and administered in Hebrew, Russian, or Arabic, depending on the preferred language of the respondent.

Food security was assessed using the short form of the USDA household food security scale [1]. The short scale has 6 items and is a validated, abridged version of the 18-item food security scale. Compared with the 18-item scale, the 6-item scale decreases respondent burden and has been shown to have high sensitivity and specificity and a minimal bias in prevalence estimates [1, 68]. Depending on the number of affirmative answers to the 6 food security questions (see Table 1 for details), households were classified as food secure (0–1 affirmative answers), moderately food insecure (2–4 affirmative answers), or severely food insecure (5–6 affirmative answers). This classification scheme was developed for the USDA 6-item household food security scale [1] and has been adopted by authors of other studies that used the short form of the scale in other countries [6, 8, 24]. We chose the present classification in keeping with this practice and because it corresponds to the categorization of the USDA 18-item scale [68] that was employed in earlier studies on food security in Israel [4, 5, 15]. Thus, our estimates of FI can be compared with results from previous research in Israel. For the category labels, we preferred the terms “moderately food insecure” and “severely food insecure” that were used in a study on FI in Israel in 2003 [4] to the revised labels in recent USDA food security reports [2], both for the sake of consistency with earlier work in Israel and easy interpretability.

TABLE 1.

Percent of affirmative answers to food security questions among Israeli elderly aged ≥65 y, 2014–2015¹

Food security question/statement		Affirmative answer2		95% CI
		(n)	(weighted %)	(weighted %)
1	The food that I bought/we bought did not last, and we did not have money to buy more.	109	11.5	9.3, 14.0
2	I/we did not have enough money to eat balanced meals.	123	12.9	10.6, 15.7
3	In the last 12 months, did you or other adults in your household ever reduce the size of your meals or skip meals because there was not enough money for food?	81	8.4	6.6, 10.8
4	How often does this happen?	69	7.1	5.5, 9.3
5	In the last 12 months, did you ever eat less than you felt you should because there was not enough money to buy food?	86	8.5	6.7, 10.9
6	In the last 12 months, were you ever hungry but did not eat because you did not have enough money to buy food?	34	3.6	2.4, 5.1

¹The nationally representative sample included 1006 elderly people.

²Affirmative answers are defined as follows: “Often true”/“Sometimes true” for questions 1 and 2; “Yes” for questions 3, 5, and 6; “Almost every month”/“Some months yes, some months not” for question 4. Question 4 was only asked if question 3 was answered affirmatively.

To our knowledge, neither the 18-item nor the 6-item household food security scale has been validated in the Israeli setting. However, an earlier study among the general population in Israel that adopted the USDA 18-item scale and classification scheme found strong associations of FI with income (defined as quintiles of household income per standard person) in both bivariate analyses and multivariable logistic regressions. The same study reported that respondents from severely food-insecure households consumed significantly less chicken, beef, fish, fruit, vegetables, and milk products than respondents from food-secure households, but considerably more starchy staples such as wheat products, potatoes, and rice. Consequently, the severely food insecure had lower intakes of essential nutrients such as protein, iron, calcium, and vitamin B₁₂. Respondents from moderately food-insecure households also consumed less meat products, fish, and milk products and had lower nutrient intakes than respondents from food-secure households, but the differences were less pronounced [4]. These findings provide strong evidence for the external validity of the 18-item scale and support its use in the Israeli context.

FI as measured by the USDA household food security scale arises from a lack of financial resources (Table 1), and indicators of economic well-being are an important element of other nationally representative food security studies [[2], [3], [4], [5], [6], [7], [8], [9], 15, 16, 25]. Since the food security scale was designed as a household-level measure [1] and household members can be expected to pool their resources at least to some extent, household income, and not individual income, should be used to construct an indicator of economic status for the analysis. Furthermore, household income has to be adjusted for household size to take into account that larger households require more resources to meet the needs of their members. This approach was also pursued in the aforementioned national studies, with the exception of the study in Portugal that used unadjusted total household income because data on household size were not available; the authors noted explicitly that this was a limitation of their study [6].

Instead of simply computing household income per capita, other national studies on FI have mostly applied more complex formulas to consider that a household’s resource needs increase less than proportionally with each additional member [4, 7, 9, 15] or have divided household income by national poverty lines or low-income thresholds that are specific to household size [2, 3, 8, 16, 25]. For example, a study on FI in France used total household net income per consumption unit, with the number of consumption units being defined according to a modified equivalence scale (1.0 for the first person, 0.5 for each additional household member aged ≥14 y, and 0.3 for each child under 14 y) [7]. However, we cannot compute household income per consumption unit or “adult equivalent” from our data because the Rav Mabat Zahav survey only gathered information on the age of the respondent, not on the age of other household members. Since a study on FI in Israel in 2016 that used household income per capita found that this indicator had a similar relationship with FI as household income per standard person [5], we opted for the simpler approach to calculate per capita household income.

Respondents were asked to indicate the net monthly income of the household from all sources by specifying one out of 10 income ranges, and per capita household income was computed by dividing the midpoint of the specified household income range by the number of household members. Households were then divided into per capita household income quartiles based on weighted cumulative percentages of per capita household income. Valid data on per capita household income were available for only 786 of 1006 households whose food security status could be determined, since a fairly large number of respondents said they did not know their household’s income or refused to answer (Figure 2). We did not impute income for respondents with missing values, these individuals were excluded from all analyses that included per capita household income. We performed additional analyses to ascertain the effects of the large loss in sample size, as described in the section about the analytical strategy below.

Figure 2.

Flow chart showing sample sizes and sources of missing data in the 2014–2015 National Health and Nutrition Survey of the Elderly in Israel (Rav Mabat Zahav). ASE, Arab sector elderly; JSE, Jewish sector elderly; NCDs, noncommunicable diseases, including the following nutrition-related diseases: cardiovascular disease, stroke, diabetes, hypertension, hypercholesterolemia, and osteoporosis. ¹The respondents were unable to remember or to focus on the questions and were excluded because cognitive impairment was suspected; details have been described elsewhere [66]. ²Values could be imputed for 65 elderly with incomplete data on food security, following established procedures [1]. ³In the bivariate analysis, missing values for some covariates led to lower sample sizes. ⁴Missing values were overlapping for some covariates. The number of elderly who were excluded because of missing data on covariates is therefore lower than the sum of the numbers of elderly with missing data across all covariates. ⁵It was not possible to determine whether this JSE was a long-term resident or a late immigrant because the year of immigration was missing.

Disability was measured using the Washington Group short set of questions on disability, which covers six basic activity domains (vision, hearing, mobility, self-care, remembering, and communicating) [65].

Drawing on US guidelines for older adults and the definitions used in a study on physical activity among Israeli elderly [[69], [70], [71], [72]], physical activity was regarded as sufficient if respondents performed at least 150 minutes of light-to-moderate-intensity leisure time activity per week (or 75 minutes of vigorous-intensity aerobic activity per week, or an equivalent mix of light-to-moderate-intensity leisure time activity and vigorous-intensity aerobic activity). Any lesser amount of physical activity was classified as insufficient physical activity. Respondents who did not engage in any light, moderate, or vigorous activity in their leisure time were considered physically inactive.

Analytical strategy

We undertook descriptive analyses to determine the percentage of affirmative responses to the six items in the short form of the household food security scale (Table 1) and the resulting prevalence of FI among households with elderly (Table 2). The 6 items are presented in the same order as in the USDA household food security scale [1] in Table 1, with the item for which the lowest severity is assumed at the top, and the item for which the highest severity is assumed at the bottom. We note that the frequency of affirmative responses is expected to decline as the severity of FI increases.

TABLE 2.

Prevalence of food insecurity among Israeli elderly aged ≥65 y, 2014–2015¹

Food security category	Prevalence		95% CI
	(n)	(weighted %)	(weighted %)
Food secure (0–1 affirmative answers)	889	87.7	85.1, 90.0
Food insecure, total (2–6 affirmative answers)	117	12.3	10.0, 14.9
Subcategories of food insecurity:
Moderately food insecure (2–4 affirmative answers)	65	7.0	5.4, 9.1
Severely food insecure (5–6 affirmative answers)	52	5.3	3.8, 7.2

¹The nationally representative sample included 1006 elderly people.

We also computed item-score correlations (the correlation of each item with the overall scale) to assess the suitability of each item for inclusion in the household food security scale and Cronbach’s alpha to evaluate the internal consistency of the scale. We thereby followed the example of several other studies that applied or validated the 6-item scale in countries outside the USA [8, 24, 73].

We then carried out bivariate analyses to explore the associations of FI with sociodemographic characteristics and indicators of health status and health behaviors (TABLE 3, TABLE 4, TABLE 5). Taking into account the cultural and socioeconomic differences between Jews and Arabs that have been documented elsewhere [67], we analyzed associations of FI with sociodemographic characteristics not only for all subjects, but also separately for each sector (Table 3). We produced cross-tabulations and tested for significant differences to examine how the prevalence of FI varied across demographic and socioeconomic groups (TABLE 3, TABLE 4), and how physical and mental health status and health behaviors differed by food security status (Table 5). Both types of information are important for developing well-targeted programs and interventions and for tailoring services to the needs of potential beneficiaries.

TABLE 3.

Prevalence of food insecurity by sociodemographic characteristics among Israeli elderly aged ≥65 y, 2014–2015

Characteristic	All subjects						Jewish sector						Arab sector
	Prevalence in study population1		Prevalence of food insecurity2			P value3	Prevalence in study population1		Prevalence of food insecurity2			P value3	Prevalence in study population1		Prevalence of food insecurity2			P value3
	(weighted %)	(n)	total	mod.	sev.		(weighted %)	(n)	total	mod.	sev.		(weighted %)	(n)	total	mod.	sev.
			(weighted %)						(weighted %)						(weighted %)
Sector
JSE	91.8	(905)	10.6	6.4	4.3	<0.001∗∗∗
ASE	8.2	(101)	30.3	14.1	16.2
Sex
Male	44.7	(472)	14.2	7.7	6.5	0.273	44.6	(426)	12.9	8.0	4.9	0.248	45.6	(46)	28.3	4.5	23.8	0.015∗
Female	55.3	(534)	10.7	6.5	4.3		55.4	(479)	8.8	5.1	3.8		54.4	(55)	31.9	22.0	9.8
Age group (y)
65–74	56.7	(588)	11.8	6.1	5.8	0.554	55.8	(516)	9.4	5.4	3.9	0.498	67.3	(72)	34.8	12.0	22.9	0.106
75–84	31.9	(313)	13.9	8.6	5.3		33.1	(296)	13.6	8.0	5.6		18.2	(17)	[19.9]	[19.9]	[0.0]
85 or more	11.4	(105)	9.6	7.2	2.5		11.1	(93)	8.2	6.1	2.1		14.5	(12)	[22.1]	[16.4]	[5.7]
Household size
1	22.7	(227)	12.2	7.6	4.7	0.103	23.5	(211)	11.6	7.3	4.3	0.720	13.6	(16)	[24.9]	[12.7]	[12.2]	0.007∗∗
2	54.6	(581)	10.0	5.7	4.3		56.3	(542)	9.9	6.0	3.9		35.0	(39)	11.7	0.0	11.7
3	14.4	(125)	18.6	11.7	6.9		12.9	(98)	10.5	7.2	3.3		31.0	(27)	56.3	32.8	23.5
4 or more	8.3	(70)	16.7	5.9	10.8		7.2	(51)	14.3	4.8	9.5		20.3	(19)	[26.1]	[10.5]	[15.6]
Education (y)
0	5.2	(58)	36.1	14.1	22.0	<0.001∗∗∗	2.6	(26)	31.5	10.1	21.4	0.015∗	35.3	(32)	40.1	17.5	22.6	0.341
1–8	18.7	(184)	17.0	9.1	7.9		15.6	(128)	14.8	8.4	6.5		54.9	(56)	23.9	11.4	12.5
9–12	29.7	(274)	12.5	8.0	4.5		31.8	(268)	11.7	7.4	4.3		6.2	(6)	—	—	—
13 or more	46.3	(464)	8.1	5.0	3.1		50.0	(460)	7.9	5.0	2.9		3.6	(4)	—	—	—
Per capita household income quartile
Bottom	22.0	(172)	28.8	13.1	15.7	<0.001∗∗∗	16.9	(110)	23.5	10.3	13.1	<0.001∗∗∗	73.3	(62)	41.2	19.5	21.7	0.444
2nd	29.9	(230)	18.4	12.2	6.2		31.3	(214)	18.3	12.1	6.2		16.2	(16)	[20.8]	[14.7]	[6.1]
3rd	22.6	(174)	5.3	4.0	1.3		23.9	(165)	5.1	4.2	0.9		9.4	(9)	—	—	—
Top	25.5	(210)	0.7	0.2	0.4		27.9	(209)	0.7	0.2	0.4		1.2	(1)	—	—	—

ASE, Arab sector elderly; JSE, Jewish sector elderly; mod., moderate; sev., severe.

Prevalence rates in square brackets are based on 10-24 unweighted cases. A dash denotes a prevalence rate based on <10 unweighted cases that is not shown.

¹For each characteristic, the weighted percentages add up to column totals of 100%, with possible minor deviations of +/−0.1 percentage points because of rounding. The unweighted number of people in the sample who belong to each category is denoted by n. Data on food insecurity are available for 1006 people, but the sum of n across categories can be lower because of missing values in the sociodemographic variables.

²The prevalence rates of total food insecurity, moderate food insecurity, and severe food insecurity were computed for each category in each row and are based on n unweighted cases. Because of rounding, the prevalence of total food insecurity may deviate from the sum of the prevalence rates of moderate and severe food insecurity by +/−0.1 percentage points.

³Pearson’s chi-squared test was corrected for the survey design and converted to an F-statistic to determine bivariate associations between food insecurity and each characteristic. ∗Significant at P value < 0.05; ∗∗significant at P value < 0.01; ∗∗∗significant at P value < 0.001.

TABLE 4.

Prevalence of food insecurity by immigration status and region of origin among Israeli elderly aged ≥65 y from the Jewish sector, 2014–2015¹

Characteristic	Prevalence in study population2		Prevalence of food insecurity3			P value4
	(weighted %)	(n)	total	moderate	severe
			(weighted %)
Immigration status5
Long-term residents	79.8	(748)	7.6	4.2	3.5	<0.001∗∗∗
Late immigrants	20.2	(156)	22.5	14.9	7.6
Region of origin
Israel	22.1	(231)	2.8	1.7	1.1	<0.001∗∗∗
Europe	19.8	(180)	4.4	4.0	0.4
Americas	3.6	(38)	9.7	3.9	5.8
Asia	13.3	(120)	13.1	5.7	7.4
Africa	16.1	(147)	13.1	6.5	6.6
FSU	25.1	(189)	19.7	12.9	6.8

FSU, Former Soviet Union.

¹The findings are shown for Jewish sector elderly only (n = 904 for immigration status, n = 905 for region of origin) because the proportion of immigrants is very high in this population group. All late immigrants in the study sample belonged to the Jewish sector.

²For each characteristic, the weighted percentages add up to column totals of 100%. The unweighted number of people in the sample who belong to each category is denoted by n.

³The prevalence rates of total food insecurity, moderate food insecurity, and severe food insecurity were computed for each category in each row and are based on n unweighted cases. Because of rounding, the prevalence of total food insecurity may deviate from the sum of the prevalence rates of moderate and severe food insecurity by +/−0.1 percentage points.

⁴Pearson’s chi-squared test was corrected for the survey design and converted to an F-statistic to determine bivariate associations between food insecurity and each characteristic. ∗Significant at P value < 0.05; ∗∗significant at P value < 0.01; ∗∗∗significant at P value < 0.001.

⁵Late immigrants immigrated to Israel in 1990 or later, mainly from the Former Soviet Union.

TABLE 5.

Health status and health behaviors by food security group among Israeli elderly aged ≥65 y, 2014–2015

Health indicator	Food secure1	Food insecure1		P value2
		moderately	severely
	(weighted %)	(weighted %)
General health
Self-assessed health status
Very good	16.2	4.7	8.2	<0.001∗∗∗
Good	35.6	23.2	10.6
Reasonable	36.0	35.9	35.3
Bad or very bad	12.3	36.3	45.8
Number of NCDs3
0–1	41.8	37.4	32.1	0.030∗
2–3	47.5	48.6	39.3
4–6	10.7	14.1	28.6
Disability
Vision
No difficulty	75.4	47.1	42.7	<0.001∗∗∗
Some difficulty	17.8	36.9	38.5
A lot of difficulty	5.9	10.7	11.8
Cannot at all	1.0	5.3	7.0
Hearing
No difficulty	62.5	48.5	43.7	0.010∗
Some difficulty	29.5	40.0	43.5
A lot of difficulty	7.9	9.5	12.8
Cannot at all	0.1	2.0	0.0
Mobility
No difficulty	48.8	26.0	15.3	<0.001∗∗∗
Some difficulty	26.6	23.0	24.7
A lot of difficulty	20.3	43.3	50.3
Cannot at all	4.4	7.8	9.7
Self-care
No difficulty	82.0	52.0	45.8	<0.001∗∗∗
Some difficulty	9.7	23.8	23.1
A lot of difficulty	4.7	17.6	28.5
Cannot at all	3.6	6.5	2.7
Remembering
No difficulty	66.5	53.5	39.6	<0.001∗∗∗
Some difficulty	28.1	37.6	36.4
A lot of difficulty	4.9	8.9	22.7
Cannot at all	0.6	0.0	1.4
Communication
No difficulty	91.4	72.8	72.4	<0.001∗∗∗
Some difficulty	6.3	25.2	18.2
A lot of difficulty	1.7	2.0	9.4
Cannot at all	0.6	0.0	0.0
Number of disabilities4
None	69.3	32.7	36.3	<0.001∗∗∗
One	17.6	32.9	19.5
More than one	13.1	34.4	44.2
Oral health and problems with eating
Self-assessed oral health
Very good	15.3	6.0	3.1	<0.001∗∗∗
Good	73.2	76.0	55.3
Not good	8.8	8.1	24.2
Bad	2.7	10.0	17.4
Chewing problems
No	83.4	70.0	44.1	<0.001∗∗∗
Yes	7.2	9.5	11.3
Yes, eats soft/minced food	9.4	20.5	44.6
Swallowing problems
None	87.3	72.1	53.0	<0.001∗∗∗
Liquids only	1.1	0.0	0.0
Food only	8.1	23.1	26.6
Food and liquids	3.5	4.8	20.4
Mental health and emotional well-being
Emotional problems (such as depression, anxiety) in past month	14.5	34.4	29.7	<0.001∗∗∗
Psychological stress or acute illness in past 3 mo	14.7	30.9	29.4	0.002∗∗
Depression, ever diagnosed by a doctor	10.2	22.1	30.7	<0.001∗∗∗
Feeling lonely
Often	7.9	16.8	23.4	<0.001∗∗∗
Sometimes	14.8	24.4	31.2
Rarely	6.1	5.0	2.7
Never	71.1	53.8	42.8
Health behaviors
Physical activity
None	50.2	75.8	68.4	<0.001∗∗∗
Insufficient	18.5	15.7	22.8
Sufficient	31.4	8.5	8.9
Smoker
Yes	9.0	19.4	24.2	0.002∗∗
No, used to smoke	28.0	17.0	29.4
Never smoked	63.1	63.6	46.4
Alcohol consumption
0 g/wk	70.5	82.4	76.1	0.374
1–49 g/wk	25.2	12.9	19.9
50 or more g/wk	4.3	4.7	4.0

NCDs, noncommunicable diseases

¹For each characteristic and food security group, the weighted percentages add up to column totals of 100% across categories, with possible minor deviations of +/− 0.1 percentage points because of rounding. Of the 1006 people in the total sample, 889 were food secure, 65 were moderately food insecure, and 52 severely food insecure. The number of people to which the statistics in this table refer ranges from 964 to 1005 people, depending on the number of missing values for each health indicator.

²Pearson’s chi-squared test was corrected for the survey design and converted to an F-statistic to determine bivariate associations between food insecurity and each health indicator. ∗Significant at P value < 0.05; ∗∗significant at P value < 0.01; ∗∗∗significant at P value < 0.001. All associations that were significant for the sample as a whole were also significant for the subgroup of long-term residents from the Jewish sector, except for the associations of food security status with the number of NCDs and smoking.

³The following nutrition-related NCDs were considered: cardiovascular disease, stroke, diabetes, hypertension, hypercholesterolemia, and osteoporosis.

⁴The number of disabilities counts the domains (comprising vision, hearing, mobility, self-care, remembering, communication) for which the respondent said he or she had a lot of difficulty or could not do this at all.

We performed multivariable logistic regressions to obtain a deeper understanding of the associations of FI with sociodemographic and health indicators, using either total FI or severe FI as the dependent variable (TABLE 6, TABLE 7). The selection of independent variables was based on the conceptual framework (Figure 1). We successively introduced 3 sets of variables into the regressions: basic demographic variables, which were included in Model 1 (population group, household size, age, and sex), socioeconomic characteristics, which were added in Model 2 (per capita household income and education), and health status indicators, which were added in Model 3 (the number of disabilities, the number of NCDs, and depression, ever diagnosed by a doctor). The population group variable combines information on sector and immigration status and distinguishes three subgroups (JSE who are long-term residents, JSE who are late immigrants, and ASE), and the health status indicators were chosen to represent disability and other aspects of physical and mental health.

TABLE 6.

Odds ratios from multivariable logistic regressions of total household food insecurity (moderate and severe), Israeli elderly aged ≥65 y, 2014–2015¹

Variable	Basis for Dummy Variable2	Model 1		Model 2		Model 3
		Odds Ratio	95% CI	Odds Ratio	95% CI	Odds Ratio	95% CI
Population group
JSE, late immigrant	JSE, long-term resident	3.27∗∗∗	1.71, 6.24	2.75∗	1.11, 6.78	3.62∗∗	1.41, 9.25
ASE	JSE, long-term resident	5.47∗∗∗	2.83, 10.58	1.58	0.65, 3.84	1.20	0.48, 3.03
Household size
Single	2 persons	1.59	0.74, 3.44	1.42	0.64, 3.19	1.38	0.58, 3.27
3 persons	2 persons	2.12∗	1.06, 4.26	1.78	0.82, 3.86	1.59	0.69, 3.68
4+ persons	2 persons	1.89	0.62, 5.73	0.95	0.32, 2.80	0.89	0.30, 2.68
Age and sex
Female	Male	0.70	0.40, 1.24	0.55	0.30, 1.00	0.44∗	0.23, 0.84
75–84 y	65–74 y	1.77	0.96, 3.26	1.29	0.69, 2.40	0.92	0.47, 1.78
85+ y	65–74 y	0.60	0.19, 1.88	0.43	0.13, 1.44	0.17∗	0.04, 0.69
Education (years of schooling)
None	13+ y			5.56∗	1.49, 20.76	6.26∗∗	1.66, 23.65
1–8 y	13+ y			2.07	0.72, 5.91	2.08	0.73, 5.90
9–12 y	13+ y			1.96	0.78, 4.95	2.12	0.83, 5.44
Per capita household income quartile
Bottom	Fourth			26.88∗∗∗	4.44, 162.90	23.56∗∗	3.71, 149.76
Second	Fourth			18.68∗∗	3.17, 110.03	16.75∗∗	2.68, 104.52
Third	Fourth			5.91	0.94, 37.20	4.74	0.69, 32.55
Number of NCDs3
2–3	0–1					0.68	0.35, 1.31
4–6	0–1					0.79	0.32, 1.94
Number of disabilities4
One	None					2.11∗	1.05, 4.23
More than one	None					4.04∗∗	1.72, 9.45
Depression, ever diagnosed by a doctor
Yes	No					3.34∗	1.35, 8.28

ASE, Arab sector elderly; JSE, Jewish sector elderly; NCDs, noncommunicable diseases.

¹The multivariable logistic regression models are defined as follows: Model 1: Food insecurity (total) = demographic characteristics; Model 2: Food insecurity (total) = demographic characteristics + socioeconomic characteristics; Model 3: Food insecurity (total) = demographic characteristics + socioeconomic characteristics + health status indicators. All models include only observations with nonmissing data for the variables used in Model 3 (n=739). Significance levels of odds ratios: ∗Significant at P value < 0.05; ∗∗significant at P value < 0.01; ∗∗∗significant at P value < 0.001.

²For variables with significant differences in food insecurity in the bivariate analysis for all subjects (population group, which combines sector and immigration status; education; per capita household income quartile; number of NCDs; number of disabilities; depression; TABLE 3, TABLE 4, TABLE 5), the category with the lowest prevalence of total food insecurity was used as the basis. For variables without significant differences in food insecurity in the bivariate analysis for all subjects and three or more categories (household size and age group; Table 3), the category with the largest weighted percentage was used as the basis. For sex, which has no significant difference in food insecurity in the bivariate analysis for all subjects and only two categories with roughly similar weighted percentages (Table 3), male sex was used as the basis.

³The following nutrition-related NCDs were considered: cardiovascular disease, stroke, diabetes, hypertension, hypercholesterolemia, and osteoporosis.

⁴The number of disabilities counts the domains (comprising vision, hearing, mobility, self-care, remembering, and communication) for which the respondent reported a lot of difficulty or said he or she could not do this at all.

TABLE 7.

Odds ratios from multivariable logistic regressions of severe household food insecurity, Israeli elderly aged ≥65 y, 2014–2015¹

Variable	Basis for Dummy Variable2	Model 1		Model 2		Model 3
		Odds Ratio	95% CI	Odds Ratio	95% CI	Odds Ratio	95% CI
Population group
JSE, late immigrant	JSE, long-term resident	1.71	0.59, 5.01	1.32	0.37, 4.66	1.91	0.51, 7.12
ASE	JSE, long-term resident	4.65∗∗∗	1.88, 11.47	1.17	0.36, 3.77	0.92	0.28, 3.07
Household size
Single	2 persons	1.21	0.40, 3.72	0.95	0.31, 2.95	0.84	0.25, 2.80
3 persons	2 persons	0.97	0.39, 2.40	0.60	0.23, 1.56	0.53	0.19, 1.49
4+ persons	2 persons	2.87	0.73, 11.19	1.08	0.31, 3.79	1.11	0.29, 4.24
Age and sex
Female	Male	0.70	0.31, 1.57	0.51	0.21, 1.23	0.43	0.17, 1.08
75–84 y	65–74 y	1.58	0.66, 3.74	1.20	0.50, 2.88	0.78	0.30, 2.03
85+ y	65–74 y	0.44	0.07, 2.87	0.31	0.04, 2.40	0.15	0.02, 1.21
Education (years of schooling)
None	13+ y			4.31	0.69, 27.03	5.24	0.84, 32.58
1–8 y	13+ y			1.74	0.42, 7.22	1.77	0.42, 7.45
9–12 y	13+ y			1.48	0.39, 5.60	1.52	0.40, 5.82
Per capita household income quartile
Bottom	Fourth			27.63∗∗	2.56, 298.38	20.11∗	1.89, 213.88
Second	Fourth			10.51	0.97, 114.02	8.11	0.74, 88.88
Third	Fourth			2.53	0.18, 34.90	2.02	0.16, 26.15
Number of NCDs3
2–3	0–1					1.05	0.42, 2.59
4–6	0–1					1.60	0.47, 5.39
Number of disabilities4
One	None					1.28	0.48, 3.42
More than one	None					2.98∗	1.12, 7.88
Depression, ever diagnosed by a doctor
Yes	No					3.33∗	1.21, 9.19

ASE, Arab sector elderly; JSE, Jewish sector elderly; NCDs, noncommunicable diseases.

¹The multivariable logistic regression models are defined as follows: Model 1: Food insecurity (severe) = demographic characteristics; Model 2: Food insecurity (severe) = demographic characteristics + socioeconomic characteristics; Model 3: Food insecurity (severe) = demographic characteristics + socioeconomic characteristics + health status indicators. All models include only observations with nonmissing data for the variables used in Model 3 (n = 739). Significance levels of odds ratios: ∗Significant at P value < 0.05; ∗∗significant at P value < 0.01; ∗∗∗significant at P value < 0.001.

²For variables with significant differences in food insecurity in the bivariate analysis for all subjects (population group, which combines sector and immigration status; education; per capita household income quartile; number of NCDs; number of disabilities; depression; TABLE 3, TABLE 4, TABLE 5), the category with the lowest prevalence of total food insecurity was used as the basis. For variables without significant differences in food insecurity in the bivariate analysis for all subjects and three or more categories (household size and age group; Table 3), the category with the largest weighted percentage was used as the basis. For sex, which has no significant difference in food insecurity in the bivariate analysis for all subjects and only two categories with roughly similar weighted percentages (Table 3), male sex was used as the basis.

³The following nutrition-related NCDs were considered: cardiovascular disease, stroke, diabetes, hypertension, hypercholesterolemia, and osteoporosis.

⁴The number of disabilities counts the domains (comprising vision, hearing, mobility, self-care, remembering, and communication) for which the respondent reported a lot of difficulty or said he or she could not do this at all.

We compared the results of Models 1 and 2 to determine if disparities in socioeconomic status are likely to explain associations of basic demographic characteristics with FI, and we assessed whether poor health status was independently associated with FI by controlling for all sociodemographic variables in Model 3. Although we could not infer causality because of the cross-sectional nature of the data, we performed this analysis in the expectation that it would shed some light on independent risk factors of FI and provide useful insights for designing policies to reduce FI.

The analytical sample included 1006 elderly for the descriptive analysis, but only 739 elderly for the 6 multivariable logistic regressions presented in TABLE 6, TABLE 7 because of missing values in covariates (Figure 2) and the need to keep the sample size constant across Models 1–3 for the sake of comparability. Compared with elderly with nonmissing values in the multivariable logistic regressions, elderly with missing values in covariates were more likely to belong to the Jewish sector, be older and suffer from a higher number of noncommunicable diseases (NCDs) and disabilities; no significant differences were found for food security status, household size, sex, education, and having ever been diagnosed with depression (Supplemental Table A1).

To examine potential bias, the descriptive and bivariate analyses were repeated for a restricted sample of 739 elderly who were included in the multivariable logistic regressions in TABLE 6, TABLE 7 (Supplemental Tables A2–A6). In addition, a sensitivity analysis was undertaken by omitting per capita household income from the multivariable logistic regressions (Supplemental Tables A7–A10). When per capita household income was dropped, the sample was at first limited to the 739 subjects represented in the multivariable logistic regressions in TABLE 6, TABLE 7 to provide benchmark estimates (Supplemental Tables A7 and A9), and then allowed to increase to the larger sample of 924 elderly that could be analyzed in the absence of the income variable (Supplemental Tables A8 and A10).

Statistical analysis

Statistical analysis was carried out using Stata software (version 15.0) [74] at an adopted significance level of 0.05. We are aware that testing multiple outcomes entails an increased risk of Type 1 error and denote 3 levels of significance (P value < 0.05, P value < 0.01, and P value < 0.001) in all tables that present test results for associations of variables (TABLE 3, TABLE 4, TABLE 5, TABLE 6, TABLE 7, Supplemental Table A1, and Supplemental Tables A4–A10).

The complex survey design with certainty units, clusters, and strata was taken into account and sampling weights were applied to estimate results at the national level. Sampling weights were developed by the CBS to adjust the distributions of the following demographic characteristics in the sample to their respective distributions in the population: household size, education (academic compared with nonacademic), sector/continent of birth [Jewish sector: Israel, Asia-Africa, Former Soviet Union (FSU), Europe-America; Arab sector], geographic area, sex, and age group. Individual case weights were expressed as the number of people in the population each respondent represented. We used Stata’s svyset command to set the survey design.

In the descriptive analysis, the prevalence of FI and percentages of affirmative answers to the 6 items in the household food security scale were determined together with 95% CIs by means of Stata’s command to estimate proportions, considering the survey design (svy: proportion) (TABLE 1, TABLE 2, and Supplemental Tables A2 and A3). We used the command alpha (with the options item and casewise for casewise deletion) to calculate Cronbach’s alpha and item-score correlations.

In the bivariate analysis, Pearson’s chi-squared was computed for the hypothesis that the rows and columns in a cross-tabulation of FI and a sociodemographic or health status variable were independent, and Pearson’s chi-squared was then corrected for the survey design with the second-order correction of Rao and Scott [75] and converted into an F-statistic to derive the significance level, using Stata’s command for two-way tables from survey data (svy: tabulate) with its default option (pearson) [76] (TABLE 3, TABLE 4, TABLE 5 and Supplemental Tables A1 and A4–A6). Multivariable logistic regressions (svy: logistic in Stata) were used to identify independent risk factors of FI, using either total FI (moderate and severe FI combined) or severe FI as the dependent variable (TABLE 6, TABLE 7, and Supplemental Tables A7–A10).

Results

The percent of affirmative answers to the 6 food security questions and the prevalence of FI are reported in TABLE 1, TABLE 2. Overall, about 12.3% of households with elderly members experienced FI, and 5.3% were affected by severe FI. The prevalence of FI was slightly higher if the restricted sample was used, but the difference amounted to <1 percentage point (Supplemental Table A3) and was not significant.

The frequency of affirmative responses tends to decline across items from the top to the bottom of Table 1, but the percentages do not fit the expected pattern perfectly: For the second item and fifth item, affirmative answers were given slightly more frequently than for the first item and fourth item, respectively, although the latter items are assumed to be less severe in the USDA 6-item scale. Yet, the differences between these percentages were relatively small and insignificant. The values for Cronbach’s alpha amounted to 0.91 for all subjects (0.89 for JSE and 0.94 for ASE), indicating high internal consistency. Item-scale correlations were also high, ranging from 0.67 to 0.90 for all subjects (0.66–0.88 for JSE and 0.67–0.95 for ASE), which suggests that all 6 items are suitable for inclusion in the household food security scale.

Table 3 shows the associations between sociodemographic characteristics and FI. FI differed significantly by sector, per capita household income quartile, and years of education. To ascertain whether the pronounced differences between JSE and ASE with regard to food security status, per capita household income, and education mainly drive the significant associations of FI with per capita household income and education for the sample as a whole, cross-tabulations are presented separately for the Jewish and Arab sectors (Table 3). Per capita household income and education were associated with FI among JSE, but not among ASE, whereas sex of the respondent and household size were associated with FI among ASE, but not among JSE. It has to be noted, however, that the results for ASE are based on a small subsample of 101 elderly persons, and very few of them belonged to the higher per capita household income and education categories.

Strong disparities in FI existed not only between JSE and ASE, but also within the Jewish sector. Table 4 shows that FI was significantly higher among JSE who are late immigrants (they immigrated to Israel in 1990 or later) than among JSE who are long-term residents (they were born in Israel or immigrated before 1990). Total FI was below 3% among JSE who were born in Israel, but 7 times as high among JSE from the FSU, who mostly immigrated in 1990 or later. The prevalence of severe FI amounted to about 7% among households with JSE who were born in Africa, Asia, or the FSU.

Table 5 presents indicators of physical and mental health, emotional well-being, and health behaviors by food security status. The food insecure were significantly more likely than the food secure to assess their general health as “bad” or “very bad,” to have been diagnosed with 4 or more nutrition-related NCDs, to report limited functioning in vision, hearing, mobility, self-care, remembering, or communicating, and to suffer from >1 disability in these 6 domains.

Poor oral health and problems with chewing and swallowing affect the ability of the elderly to use available food. Nearly 90% of the food secure described their oral health as “very good” or “good,” but <60% of the severely food insecure did so. Likewise, problems with chewing or swallowing were significantly more prevalent among the food insecure than among the food secure, and the severely food insecure were particularly affected.

In terms of mental health and emotional well-being, the food insecure were also disadvantaged. The rate of self-reported emotional problems (such as depression and anxiety) in the past month, the rate of psychological problems or acute illness in the past 3 mo, and the proportion of elderly who had been diagnosed with depression at some point in their lives were more than twice as high among the food insecure than among the food secure. Feelings of loneliness were significantly more common among the food insecure, although they were not more likely to live alone.

Regarding health behaviors, the moderately and severely food insecure were more likely to be physically inactive than the food secure, and the prevalence of smoking was significantly higher among the food insecure. Alcohol consumption was low on average and did not differ significantly by food security status; the association of FI with alcohol consumption was also insignificant if it was analyzed separately for subgroups with distinct alcohol consumption patterns, namely, men and women, and JSE and ASE. It should be noted in this context that the large majority of ASE was Muslim and did not consume any alcohol because of religious observance.

The associations described above proved robust for the restricted sample, with the exception of an insignificant association of a disability in hearing with FI (Supplemental Tables A4–A6).

Table 6 depicts the odds ratios from multivariable logistic regressions of total FI (including moderate and severe FI) on demographic, socioeconomic, and health indicators. Considering that multigenerational living arrangements are common among Arab Israelis and immigrants from the FSU [26], household size serves as a proxy variable for social support and is preferred to marital status, which lacks any significant bivariate association with food security among all subjects, JSE, ASE, men, and women.

In Model 1 which includes only basic demographic variables, being a late immigrant (this group consists exclusively of JSE) and belonging to the Arab sector significantly increased the odds of being food insecure as compared with being a JSE and long-term resident. After education and per capita household income are introduced as additional variables in Model 2, the odds ratio for ASE decreases considerably and becomes insignificant, whereas the odds ratio for late immigration declines only slightly and remains weakly significant. (Yet, if only per capita household income is introduced, without education—or if interaction terms for late immigration and primary schooling, and late immigration and secondary schooling are considered together with per capita household income and education—the odds ratio for late immigrants becomes insignificant; the results are available upon request.) The high odds ratios for the bottom 2 per capita household income quartiles stand out in Model 2. At the same time, elderly without schooling have higher odds of being food insecure than those with a completed high school education, even when controlling for per capita household income.

In Model 3, indicators of health status are included: the number of nutrition-related NCDs, the number of disabilities, and a diagnosis of depression at some point in the person’s life. The number of NCDs was insignificant, and this also holds if no indicators of disability and depression are introduced into the regression together with the number of NCDs (results not shown). However, controlling for demographic and socioeconomic variables and the other health status indicators in Model 3, elderly who reported disability in 1 out of 6 domains were twice as likely to be food insecure as elderly without disability, and elderly with >1 disability were 4 times as likely to be food insecure as those without disability. At the same time, those who were ever diagnosed with depression had 3 times higher odds to be food insecure than those who were never diagnosed with depression. Similar findings were obtained when the variable for the diagnosis of depression was replaced with a variable indicating that the respondent had emotional problems, such as depression or anxiety, in the past 3 mo (results not shown).

The results of multivariable logistic regressions of severe FI are presented in Table 7. Only being Arab is significant in the logistic regression with basic demographic variables in Model 1, and becomes insignificant when per capita household income and education are added in Model 2. In contrast to the regression of total FI, only the bottom quartile has a significantly elevated odds ratio in the regression of severe FI, not the second quartile, and education is insignificant when per capita household income is controlled for. Similarly, only elderly with >1 disability had higher odds of being severely food insecure than those without any disability. Elderly who were ever diagnosed with depression were about 3 times more likely to be severely food insecure. A variable indicating self-reported emotional problems in the past 3 mo turned out insignificant when it replaced the depression variable (results not shown).

Additional regressions using interaction terms showed that age and sex did not modify the associations of the health status variables with FI (results not shown).

Omitting per capita household income from the regressions increased the odds ratios of education—as would be expected—but had little effect on the size and significance of the odds ratios of the health variables (Supplemental Tables A7–A10).

Discussion

The prevalence of food insecurity

Our estimates of FI among the elderly (12.3%) are very similar to those published by the National Insurance Institute, which found that 13.0% of households with elderly were food insecure in 2016 (with 7.3% being moderately food insecure and 5.7% severely food insecure) [5], which lends credibility to our findings. We observed this similarity in estimated prevalence rates despite the differences in the timing and methodological approach of the surveys: The survey by the National Insurance Institute used the 18-item USDA household food security scale and not the abridged 6-item household food security scale, employed a different sampling frame and sampling strategy to construct a representative sample of the general population, and was conducted via telephone and not face to face [5]. An earlier study about the effects of financial hardship on the living conditions of the elderly in Israel also used the USDA 18-item scale and reported rates of FI that exceeded our more recent estimates for 2014–2015: 19% of households with elderly were food insecure in 2003, with 12% being moderately food insecure and 7% severely food insecure [15]. Evidently, FI among the elderly has been more common in Israel than in the USA, where it affected only 7.2% of households with elderly in 2019 [2].

Bivariate associations with sociodemographic characteristics

Bivariate associations of FI with per capita household income, education, and ethnicity were highly significant for the total sample, and so were associations of FI with late immigration and region of origin for JSE. The observed strong association of per capita household income with food security is not surprising, since the household food security scale was tailored to capture FI that is attributable to financial resource constraints [1]. Several other national studies found that low income (defined almost exclusively as total household income adjusted for household size, see the methods section for details) was a primary risk factor for FI, both among the general population [[2], [3], [4], [5], [6], [7], [8], [9]] and among the elderly in Israel [15] and elsewhere [16, 25].

Similarly, elevated levels of FI have been documented for ethnic minorities and the less educated in other studies, both for the general population [[2], [3], [4], [5], 10] and the elderly [16, 25]. A study about the effects of financial hardship on the living conditions of the elderly in Israel in 2003 found that 26% of households with ASE, 17% of households with long-term resident JSE, and 21% of households with late immigrant JSE were food insecure, but the differences were not statistically significant (P ≥ 0.05) [15]; it remains unclear whether no significant bivariate association could be determined because the disparities observed in this earlier study were less pronounced or because the smaller sample size (n = 644) [15] limited statistical power. Higher FI among ASE is related to their economic vulnerability: They are mostly dependent on small payments from the social security system because few of them receive a pension from their work. ASE faced difficulties in the job market and are lacking private pensions because they worked predominantly in workplaces that did not offer pensions [26].

The food security of immigrants has been studied among small samples of recently arrived refugees in the UK and the USA: The occurrence and severity of FI were associated with low income, low education, language barriers and difficulty in the shopping environment, and FI tended to decline over time in the course of acculturation [[27], [28], [29]]. Studies among the general population in Israel found that the rate of FI among late immigrants, who arrived in Israel in 1990 or later, still exceeded the national prevalence rate in 2003 [4], but that this was no longer true in 2011–2012 and 2016 [5].

Yet, we found that in 2014–2015, households with elderly late immigrants had higher FI than all households with elderly. Late immigrants predominantly originated from the FSU and most of them were highly educated, but for many, their degrees were not recognized, so those with a tertiary education had to learn Hebrew, take retraining courses and pass exams to work in their previous professions. Many Russian-speaking immigrants—especially those who were older at the time of immigration—resorted to working in low-paying jobs, such as factory worker, waiter, or security guard [[77], [78], [79]]. A sizable portion of late immigrants from the FSU who arrived at a later age also did not have enough time in Israel to qualify for a meaningful pension [26], and most of them did not receive pensions from their countries of origin. This largely explains their economic vulnerability and higher risk of FI.

Since >70% of all elderly from the FSU immigrated late, FI is elevated among those who were born in this region. The higher rates of FI among immigrants from Africa and Asia have different reasons: They mostly came from the Middle East and North Africa, departing from their countries of origin during the War of Independence and in the years that followed. They had lower levels of formal education and a different cultural background than the European-born founders of the state and many were settled in Israel’s periphery, factors that contributed to their economic disadvantage [80].

Bivariate associations with health status and health behaviors

A systematic review of the literature on FI and health outcomes that was published in 2015 found that surprisingly little work had been done on the relationship between FI and health among the elderly [30]. To date, there are only few related publications, including studies among the elderly in the USA [16, [20], [21], [22], [31], [32], [33], 81], Australia [34], Ghana [35], Korea [36], Malaysia [24], Mexico [23], Portugal [37], and South Africa [82]. These studies documented associations of FI with poor self-rated health [20, 21, 34], chronic diseases [21, 37], the use of health services for adverse health events [33], functional limitations measured with the ADL or IADL [16, [21], [22], [23]], mild cognitive decline [82], psychological distress [35], depression or suicide ideation [21, 23, 24, 31, 36, 81], social isolation [16], loneliness [32], and smoking [81].

Consistent with evidence from other parts of the world, the food-insecure elderly in Israel in our study had worse general physical and mental health than their food-secure peers, and they were more likely to feel lonely and smoke. In addition, this study finds associations of FI among the elderly with the full spectrum of disabilities (Washington scale), poor self-rated oral health, problems with chewing and swallowing, and physical inactivity that, to the best of our knowledge, have not been reported elsewhere for this age group. We note that our findings on disability and oral health are broadly in line with the results of an earlier study among Israeli elderly in 2003 that examined a set of 4 different health indicators—limited mobility (defined as inability to walk outside the house without the help of a person or device) and untreated vision, hearing, and dental problems—and found significant associations of FI with all of these measures [15].

For working-age adults, the relationship between FI and health has been researched more extensively: Numerous earlier studies confirmed associations of FI with health outcomes such as poor self-rated health, chronic conditions and disability [11, 30, [38], [39], [40], [41]], poor oral health [42], mental health problems and psychological distress [11, 30, 41, [43], [44], [45]], and some studies found associations of FI with adverse health behaviors such as physical inactivity and smoking [83, 84]. Similar findings for health outcomes and behaviors were obtained for combined samples of working-age and older adults using population-representative data [6, 7, 10, 30, [85], [86], [87], [88], [89]].

The relationship between FI and poor health is complex and reciprocal. Chronic illness may cause higher health-related expenses and strain an already tight budget [41]. Health problems can also affect people’s ability to cope with limited resources: Individuals with mental health issues may fail to budget resources rationally, and physical limitations can prevent people from employing time-intensive strategies to obtain food at reduced cost and seek assistance from charities [41, 46]. Health shocks can cause income losses for elderly who continue to work beyond retirement age1 and thereby increase the risk of FI [48]. Conversely, FI can contribute to poor health: FI was associated with cost-related medication underuse among older adults [[49], [50], [51]], feelings of worry and anxiety among elderly who cannot afford the right quality and types of foods for their specific health problems or overall health and well-being [17], and increased stress that may contribute to physical and mental health problems [18, 52]. In addition, the food insecure may be unable to afford a balanced, nutritious diet, with negative consequences for their health [17, 18].

However, the observed association between FI and poor health could also be driven by common causes that influence both, such as low income and low education. Financial hardship forced some elderly Israelis to forego purchasing medication or medical treatment in favor of buying food, or vice versa [15]. Low education can affect income [47, 48, 53] and the ability to cope with scarce resources [46], and may also constrain knowledge about healthy behaviors and the ability to follow medical instructions [54, 55]. The multivariable logistic regression analysis sheds more light on the nature of the association between FI and poor health by controlling for socioeconomic factors.

Multivariable logistic regression analysis

Belonging to the Arab minority and being a late immigrant were significantly associated with total FI in a multivariable logistic regression on basic demographic variables. The lower socioeconomic status of ASE accounted for their higher rate of overall FI: The odds ratio for being Arab declined and became insignificant if both per capita household income and education were introduced into the regression. Low education not only limits professional opportunities and income [48, 53], it can also make it more difficult to cope with scarce resources in order to avert or mitigate FI. Formal education and on-the-job training are thought to improve people’s ability to manage a tight budget and enhance the communication skills that are necessary to negotiate with state welfare offices and private charities [46].

For Arab Israelis, education reduces the language barrier, facilitating access to government services. Arab Israelis grow up with Arabic as their native tongue [90], and proficiency in the Hebrew language is associated with age and level of education [91]. According to data from 2011, 62% of Arabs aged 60–69 y, and 92% of those aged ≥70 y had difficulty filling in forms and writing official letters in Hebrew, and one-quarter of all Arab Israeli adults aged ≥20 y who did not speak Hebrew well refrained from applying for government services [91].2 This will not only affect the monetary assistance Arab elderly receive (which should be reflected in the per capita household income variable), but also the in-kind services they receive: Senior citizens are eligible for noncash services, such as community-based services, and other benefits provided by various government ministries [92].

Low incomes were sufficient to explain higher overall FI among late immigrants, since the odds ratio for this group became insignificant if only per capita household income was introduced into the model (results not shown). However, the odds ratio for late immigrants became significant again if education was added. The mismatch between education and per capita household income among late immigrants that has been discussed earlier explains this observation, as the regression with interaction terms for late immigration and education confirmed.

FI was associated with the number of disabilities (one or several compared with none) and a history of depression (or, alternatively, recent emotional problems), even when controlling for demographic and socioeconomic factors. This finding suggests that the association of FI with disability and depression was not entirely driven by potential common causes such as low education and low per capita household income; analyses for the general population indeed showed that the receipt of disability benefits was independently associated with FI when controlling for a range of sociodemographic variables [5]. Yet, the number of NCDs was not associated with FI when other factors were considered. Late immigrants and ASE had higher rates of NCDs, which coincided with higher FI among these groups, explaining the weakly significant bivariate association for the sample as a whole; among long-term residents from the Jewish sector, no significant association was found.

In the logistic regression of severe FI on basic demographic variables, being Arab was a significant risk factor, but not being a late immigrant. This finding is plausible, since the rate of severe FI was more than twice as high among ASE than among elderly late immigrants. Only being in the bottom per capita household income quartile and having more than one disability were significantly associated with severe FI, but not being in the second per capita household income quartile and having only one disability, as in the regressions of total FI. This is consistent with the notion that severe FI reflects a more serious form of deprivation than total FI.

Strengths and limitations of the study

Among the strengths of this study are the comprehensive conceptual framework that was specifically developed for this study, and the use of data from the second National Health and Nutrition Survey of the Elderly in Israel. This nationally representative survey is particularly suitable for our analysis because it collected detailed data on socioeconomic and demographic characteristics, food security, health status, and health behaviors. The rich data set and the unique composition of the elderly population in Israel allowed derivation of new, policy-relevant empirical insights about the association between FI and health among the elderly on the one hand, and about the link between immigration and FI on the other hand.

Our study adds to the existing body of research on FI in Israel in several ways. In a report on FI among the general population in Israel in 2016, the National Insurance Institute presented the following types of statistics for different population subgroups, including households with elderly: the prevalence of FI, the permanence of, entry into and exit from FI between 2011–2012 and 2016, and the percentage of households who received assistance from family members or aid organizations by food security status [5]. This study provided a point of comparison for some of our findings, such as the prevalence of FI among households with elderly, but it did not examine associations between FI and sociodemographic characteristics or health indicators among the elderly as a population subgroup, as our study did. Another, much earlier study on the effects of financial hardship on the living conditions of the elderly in 2003 explored bivariate associations of FI with sociodemographic characteristics and a limited set of health indicators [15]. In contrast to our study, this study did not perform multivariable regression analysis to identify independent risk factors of FI, and the only health indicators that were examined were limited mobility (defined as inability to walk outside the house without the help of a person or device) and untreated vision, hearing, and dental problems [15]; mental health problems, vision, hearing, and dental problems that persisted despite treatment, and other functional limitations were not considered. We investigated associations of FI with a more comprehensive set of health indicators, including indicators of mental health and measures of disability that follow standard definitions (Washington scale) and also cover the domains of self-care, remembering, and communication, as well as health behaviors.

A major limitation of this study is the cross-sectional nature of the data, which implies that the direction of the associations cannot be determined and that they cannot be interpreted as causal. We also acknowledge the possibility that the significance of some associations reported in this study may have resulted from Type 1 error, as we tested multiple outcomes using a significance level of 0.05. However, we frequently observed associations at higher significance levels, which are less likely to have occurred by chance; for example, testing for bivariate associations of FI with 19 health indicators produced an insignificant result (P = 0.374) for 1 indicator, P < 0.05 for 2 indicators, P < 0.01 for another 2 indicators, and P < 0.001 for the remaining 14 indicators.

We also note that the confidence intervals for most of the odds ratios in the multivariable logistic regressions are wide and that they are extremely wide for some categories, especially for a lack of formal education (no schooling) and for the per capita household income quartiles. In addition, the small numbers of Arabs and late immigrants limited statistical power when analyzing subsamples. The number of ultra-orthodox elderly was too small for a separate analysis, and other populations at risk, such as the homeless and the Bedouin in unrecognized localities, were not included in the survey.

Considering these limitations, further research is warranted to corroborate the findings of this study and to fill remaining knowledge gaps. We recommend that future surveys will be prospective, oversample vulnerable groups, and include questions about the knowledge of and use of local food and nutrition services, the reasons why existing programs were used or not used, and the receipt of income supplements or disability benefits from the National Insurance Institute.

Conclusions

In conclusion, this study found associations of FI with depression and disability that are independent of sociodemographic characteristics. We conclude that providing adequate income support—for example, by raising government allowances—is vital to reduce FI among the elderly, but is unlikely to ensure good nutrition for those who experience difficulty to physically access, prepare, and eat food. We therefore recommend to expand the outreach of existing services that provide subsidized congregate and home-delivered meals to meet the needs of elderly with disabilities and counter social isolation [93, 94]. We advise that the menu should include soft and minced foods in addition to regular foods to accommodate the needs of recipients who have difficulty chewing or swallowing, as we found that these problems were especially common among the food-insecure elderly. Since this study revealed that potential beneficiaries were also disproportionately affected by multiple NCDs, we recommend that meal planners take the special dietary requirements of these individuals into account. Considering that smoking and insufficient physical activity were found to be prevalent among the food-insecure elderly, we propose exploring options to integrate community-based food and nutrition services with health promotion efforts [14].

We would like to emphasize that additional measures may be necessary to address the common problem of underutilization of services and assistance programs by elderly citizens [32, 92, 95]. Lack of awareness of existing programs and services, confusion about eligibility criteria, difficulty with the application process, and stigma related to receiving assistance have been identified as key barriers to participation in food assistance programs in the USA [32, 95]. Other US studies have suggested that health care centers could serve as a focal point for screening for FI and for directing food-insecure elderly to available resources and services [32, 56, 96], and we consider it worthwhile to test this approach in the Israeli setting as well. We found that disadvantaged groups—elderly citizens with low education, difficulty in Hebrew, and functional impairments—are overrepresented among the food insecure in Israel, and we recommend to offer more assistance with the application for government assistance or community-based services to these groups to help overcome barriers to access.

Since this study found associations of FI with low education, ethnic minority status and related difficulties in the job market, and late immigration, we conclude that long-term strategies to prevent FI in old age should target youth and working-age adults and promote education, access to gainful employment and the swift integration of new immigrants.

Author disclosures

The authors report no conflicts of interest.

Funding

This work was supported by the Israel Center for Disease Control, Ministry of Health, Israel. This research received no specific grant from any funding agency, commercial or not-for-profit sectors.

Author responsibilities

The authors’ responsibilities were as follows – DW, T Shohat, and LKB designed the study; LKB and T Shohat coordinated and supervised the data collection; RG and T Shimony collected data; DW, LN, and T Shimony analyzed data; DW, RG, T Shimony, and LN wrote the paper; DW had primary responsibility for final content; and all authors have read and approved the manuscript.

Data Availability

Data described in the manuscript, code book, and analytic code will be made available upon request pending the release of an anonymized data set by the Central Bureau of Statistics of Israel for public use, in accordance with CBS regulations.

Appendix A. Supplementary data

The following is the Supplementary data to this article: