Classification Differences in Food Insecurity Measures between the United States and Canada: Practical Implications for Trend Monitoring and Health Research

ABSTRACT

Background

Food insecurity, inadequate access to food due to financial constraints, is a major public health issue in the United States and Canada, where the same 18-item questionnaire is used to monitor food insecurity. Researchers often assume that findings on food insecurity from the 2 countries are comparable with each other, but there are between-country differences in how food insecurity status is determined.

Objectives

We aimed to compare the distribution of household food insecurity in the Canadian population applying the US and Canadian classification schemes. We also examined the extent to which associations between food insecurity and adults’ health differ under the 2 schemes.

Methods

We used the population-representative Canadian Community Health Survey 2005–2017 linked to administrative health records. Food insecurity was measured by the Household Food Security Survey Module. Adults 18 y and older with valid food insecurity status were included from all jurisdictions except Quebec (n = 403,200). We cross-tabulated food insecurity status classified by the US and Canadian schemes. We also fitted logistic regressions on self-reported and objective health measures adjusting for confounders.

Results

Applying the Canadian classification scheme, 7.7% of households were food insecure; the number fell to 6.0% with the US scheme. Associations between food insecurity status and health measures were mostly similar across classification schemes, although the associations between food insecurity and self-reported health were slightly larger if the US scheme was applied. Marginal food security/insecurity was associated with worse health measures irrespective of the classification scheme. United States–Canada discordance in classification of marginal food security/insecurity had a limited effect on health prediction.

Conclusions

United States–Canada differences in classification affected the apparent distribution of household food insecurity but not the associations between food insecurity and measures of adult health. Marginal food security/insecurity should be set apart from the food-secure group for trend monitoring and health research.

Introduction

Food insecurity, defined as inadequate access to food due to financial constraints, affects millions of people in the United States and Canada (1, 2). In both countries, food insecurity has been associated with poorer dietary intakes (3–7) and multiple negative health outcomes, including chronic illnesses and premature death (8–17). Greater health care utilization and cost as well as cost-related medication nonadherence have also been observed among food-insecure adults than among their food-secure counterparts in both countries (14, 18–22). Research findings from the 2 countries are often treated as interchangeable (8, 23–25), perhaps in part because of similarities in the measurement and monitoring of food insecurity.

Both the United States and Canada use the Household Food Security Survey Module (HFSSM) to monitor household food insecurity. This 18-item questionnaire was developed by the USDA to assess a household's experience of food insecurity in the prior 12 mo (2, 26–28). Implemented in the United States in 1995, this standardized questionnaire was adopted for food insecurity monitoring in Canada in 2004. A Rasch model, described in detail elsewhere (29), is the latent trait model underlying the food insecurity scale derived from the questionnaire. Ten of the 18 questions ask about the experiences of adults and the household in general, whereas 8 questions ask about experiences among children age <18 y if there are any in the household. The questions address different levels of food insecurity experiences, which range from worrying about running out of food to not eating for a whole day. Some questions are asked for adults and children separately, such as ever skipping a meal; other questions are exclusive to adults, such as losing weight. Answers to questions on children are indicative of adults’ food security experiences as well because adults usually sacrifice their own diet to shield children from food insecurity (30). Each affirmative answer is assigned a raw score of 1. A household could have a raw score of 0–10 on the adult scale; households with children could further score 0–8 on the child scale.

Both the United States and Canada base their determination of household food insecurity on the raw scores, but they apply different classification schemes and labels to describe the presence and severity of food insecurity (Table 1). When first developed, neither classification scheme defined a “marginal” category, but it was subsequently added in both countries (31, 32). Increasingly in food insecurity research, the “marginal” category is treated as a discrete category or a part of the “food insecure” group because of an abundance of evidence that individuals in marginally food-secure/insecure households have poorer health than those with 0 affirmatives on the food insecurity scales (1, 9, 10, 20, 22, 33–40).

TABLE 1.

Coding schemes of FI severity in Canada and the United States based on an 18-item standardized questionnaire¹

	No children	With children
Canada
Food secure	0–1A	0–1A or 0–1C
No FI	0A	0(A + C)
Marginal FI	1A	1A or 1C
Food insecure	2–10A	2–10A or 2–8C
Moderate FI	2–5A	2–5A or 2–4C
Severe FI	6–10A	6–10A or 5–8C
United States
Food secure	0–2A	0–2(A + C)
High FS	0A	0(A + C)
Marginal FS	1–2A	1–2(A + C)
Food insecure	3–10A	3–18(A + C)
Low FS	3–5A	3–7(A + C)
Very low FS	6–10A	8–18(A + C)

¹Based on Coleman-Jensen et al. (2), Government of Canada (31), and USDA (32). A, affirmative answers to questions on adults aged 18 y or older; C, affirmative answers to questions on children 0–17 y old; FI, food insecurity; FS, food security.

Direct comparisons of food insecurity statistics between the United States and Canada are complicated by differences in dimensionality and thresholds applied to classify households. The United States uses the total number of affirmative answers for food insecurity classification without distinguishing answers to adult-referenced questions from those to child-referenced questions (i.e., unidimensionality). In contrast, Canada considers answers to the adult- and child-referenced questions separately and uses both for food insecurity classification (i.e., bidimensionality). The discrepancy reflects 2 different approaches to addressing concerns about statistical bias in the determination of household food security status for households with and without children with the HFSSM (27, 41–43). Practically, this difference results in certain households with children being classified into different food insecurity categories depending on the classification scheme applied, a phenomenon we refer to as “discordance” in this article. Comparisons of unidimensional with bidimensional coding on the determination of food insecurity among US households with children revealed that neither scheme was inherently superior to the other (41, 42). The USDA's unidimensional scheme yielded a somewhat higher rate of very low food security than the bidimensional scheme (42), but food insecurity was similarly associated with dietary quality across the 2 schemes (41). These comparisons were specific to households with children, however, and conducted using the US thresholds to define food security status. How dimensionality affects food security statistics based on Canadian thresholds remains unknown.

Discordance can also emerge from differences in classification thresholds. Having 2 affirmative answers to the 10 questions for adults would place a household without children in the “marginal food security” category in the United States but the “moderate food insecurity” category in Canada. Among households with children, the threshold difference compounds the dimensionality difference, further complicating the pattern of discordance. For instance, a household with children with a raw score of 2 on the adult scale and 0 on the child scale is “marginally food secure” by the US scheme but “moderately food insecure” by the Canadian scheme. Whereas, another household scoring 4 on both the adult scale and child scale has experienced “very low food security” by the US scheme but still “moderate food insecurity” by the Canadian scheme. Differences in thresholds and dimensionality mean that Canadian and American estimates of food insecurity prevalence are not directly comparable.

Food insecurity has been established as a social determinant of health in the United States and Canada alike (8, 11, 13–18), but the extent to which differences in classification schemes affect the correlation between food insecurity and health has not been investigated. Understanding the implications of such differences is necessary to inform the interpretation of research findings from both countries. We aim to answer 3 questions in this study. First, we compare the distributions of household food insecurity in the Canadian adult population applying the American and Canadian classification schemes. Then, we probe the extent to which associations between household food insecurity and adults’ health measures differ under the US compared with the Canadian food insecurity classification scheme. Lastly, we conduct a targeted examination of the 1 large group of adults who are differentially classified by these 2 schemes to determine how their odds of various health measures compare with adults in the adjacent categories of food insecurity severity. We use all the available population monitoring data from Canada that have been linked to national health administrative data to address these 3 questions. We use adult health as an instrument to compare the implications of differences in food security classification schemes. Children's health is related to household food insecurity, yet beyond the scope of this study, given our lack of access to measures of health for children <12 y of age.

Methods

Data

We designed a cross-sectional study by merging a survey with administrative data. The study was approved by the Health Sciences Research Ethics Board in the University of Toronto.

The Canadian Community Health Survey (CCHS) is a nationally representative survey administered annually to ∼65,000 Canadians aged 12 y and older; it is the survey platform for food insecurity monitoring in Canada. Questions on household demographics, socioeconomic status, and self-assessed personal health are answered by a randomly selected member of surveyed households. Included since 2005, the HFSSM is answered by an adult representative in the household.

The Discharge Abstract Database (DAD) captures acute care hospitalization in Canada, containing administrative records on the date and causes of acute care admissions in all jurisdictions except Quebec from 1996 to 2017. The Canadian Vital Statistics Death Database (CVSD) captures the date and cause of all deaths nationwide; records from 2000 to 2017 are linkable to the CCHS. We merged DAD 2003–2017 and CVSD 2005–2017 with CCHS 2005–2017 to build objectively measured health variables for survey respondents.

Sample

CCHS 2005–2017 comprises 667,300 respondents (Supplemental Figure 1). The food security module of CCHS is not mandatory; certain jurisdictions have opted out of the food insecurity measurement in years when the choice was given (Supplemental Table 1). We excluded 264,100 respondents including everyone aged <18 y, from jurisdiction-years that opted out of food insecurity measurement, with missing food insecurity status, or in Quebec—the province without hospitalization records in DAD. The analytical sample consisted of 403,200 adults aged 18 y and older with nonmissing food insecurity status, representing ∼209 million people in 2005–2017 (16 million/y). We further differentiated adults in households with children (n = 93,200) and those without children (n = 310,000) given their differential food insecurity classifications in Canada compared with the United States. We also built a “marginal-moderate subsample” (n = 34,400) with households simultaneously in “marginal” or “moderate” food insecurity by the Canadian scheme and “marginal” or “low” food security by the US scheme because they were the subsample most affected by the classification discordance.

Measurements

Household food insecurity in the past 12 mo was measured by the 18-item HFSSM in the CCHS (Supplemental Questionnaire 1). Households were placed in “no,” “marginal,” “moderate,” and “severe” food insecurity categories based on the Canadian scheme and in “high,” “marginal,” “low,” and “very low” food security categories based on the US scheme (Table 1). We further dichotomized food insecurity status by collapsing the first 2 and last 2 categories under both schemes, respectively, consistent with government practices in the 2 countries (31, 32). In total, we constructed 2 four-level variables and 2 binary variables to measure food insecurity.

We built 8 binary health measures to examine their respective correlation with food insecurity across classification schemes, selecting measures that have been associated with food insecurity severity in Canada and the United States alike (3–6, 8, 9, 11, 12, 14, 20, 22). Four of the measures were self-reported variables from the CCHS (44), namely low fruit and vegetable intake (<5 compared with ≥5 servings of fruits and vegetables per day), fair or poor overall health (compared with excellent, very good, or good health), fair or poor mental health (compared with excellent, very good, or good mental health), and any chronic condition among heart diseases, stroke, hypertension, diabetes, cancers, mood disorders, and anxiety disorders (compared with none of them). We also built 4 objectively measured outcomes from the DAD and CVSD. They were any acute care admission in the past 12 mo (DAD), any acute care admission due to mental disorders in the past 12 mo (DAD), staying in acute care for ≥30 d in the past 12 mo (DAD), and death within 2 y since interview (CVSD).

We selected sociodemographic covariates that may confound the food insecurity–health relation based on theories and empirical evidence (8–12, 20, 45, 46). These covariates have independently predicted both food insecurity and health (11, 14, 20, 46); controlling for them should reduce type I error. We adjusted for respondents’ sex (female, male), age (y), smoking status (current smoker, former smoker, never smoked), and alcohol consumption (once or more per week, once every 2 wk, none). We further included highest education attained in the household (high school incomplete, high school diploma, some college, bachelor's degree), household income (in intervals of 10,000 Canadian dollars), homeownership (homeowner, renter), and household composition (couple with children, couple without children, single parents, other households) as proxies for household socioeconomic status. Survey cycle and province or territory of residence were controlled to account for unobserved geo-temporal influence. All of the aforementioned covariates came from the CCHS. We also adjusted for DAD-based acute care admission in the 13- to 24-mo period before interview to minimize the confounding effect from poor baseline health. Missing values in covariates were few and kept in the analyses as a separate category of the covariates.

Analysis

We started by cross-tabulating household food insecurity status defined by Canada with that by the United States to identify the extent of discordance. Then we presented the health profile by food insecurity status as defined by the 2 classification schemes and used Student's t-test to compare the percentages across schemes. Next, to examine differences in the health implications of food insecurity caused by classification differences, we fitted logistic regressions of the 8 dichotomous health measures on food insecurity status by different classification schemes adjusting for confounders. We also compared the ORs across schemes through Wald-type nonlinear tests.

The classification difference originates from thresholds and dimensionality for households with children but from thresholds only when it comes to households without children. Thus, we analyzed the 2 subgroups separately where appropriate to understand the difference and similarity in the impact of classification on food insecurity monitoring and its health implications.

Finally, to examine the impact of US–Canada discordance in classification of the “marginal” group, we used the “marginal-moderate subsample” to compare the prevalence of health measures between the “US-marginal Canada-moderate” discordant group and its 2 neighboring concordant groups (i.e., “US-marginal Canada-marginal” and “US-low Canada-moderate”).

Survey weights were applied to descriptive analyses to account for sampling bias. Significance level was set to P < 0.05; actual P values for comparisons are provided (Supplemental Tables 2, 3, and 4). We used Stata SE 17 (StataCorp LLC) for analyses.

Results

Of the 403,600 households sampled, 7.70% of households were food insecure by the Canadian classification scheme; the number fell to 6.04% with the US scheme applied (Table 2). Thus, only 78.4% of households classified as food insecure by the Canadian scheme would be considered food insecure by the US scheme. A total of 1.83% (1.66% + 0.12% + 0.05%) presented discordance across the 4-level food insecurity categories. With food insecurity dichotomies, 1.66% of the sample were food secure by the US scheme but food insecure by the Canadian scheme. The same discordance occurred in 1.61% of households with children and 1.68% of households without children, respectively (chi-square test P > 0.05). A trivial 0.17% of the overall sample—all with children—had discordance in the moderate and severe food insecurity categories, which did not affect the dichotomization of food insecurity.

TABLE 2.

Weighted FI status distribution as defined in Canada (rows) and the United States (columns) among adults 18 y old and older in the Canadian Community Health Survey linked file 2005–2017¹

	US classification
	Food-secure		Food-insecure
Canadian classification	High FS	Marginal FS	Low FS	Very low FS	Total
Overall (n = 403,200)
No FI	88.56				88.56
Marginal FI		3.74			3.74
Moderate FI		1.66	3.54	0.05	5.25
Severe FI			0.12	2.33	2.45
Total	88.56	5.40	3.66	2.38	100.00
Overall (n = 403,200)
Food-secure	92.30				92.30
Food-insecure	1.66		6.04		7.70
Total	93.96		6.04		100.00
With children (n = 93,200)
Food-secure	90.26				90.26
Food-insecure	1.61		8.13		9.74
Total	91.87		8.13		100.00
No children (n = 310,000)
Food-secure	93.30				93.30
Food-insecure	1.68		5.02		6.70
Total	94.98		5.02		100.00

¹Values are weighted percentages. The overall sample is representative of 209 million people in Canada (Quebec excluded) in the 13 y from 2005 to 2017. Cells do not necessarily sum to “total” owing to mandatory rounding during weighting. All cells’ percentages are significantly different between households with children and those without children except those for the “US-marginal Canada-moderate” adults (1.61% compared with 1.68%, P > 0.05). FI, food insecurity; FS, food security.

With few exceptions, greater shares of adults from food-insecure households had negative health indications than their food-secure peers (Table 3, Supplemental Table 2). The prevalence was slightly higher for self-reported health status (excluding fruit and vegetable intake) when the US instead of the Canadian scheme was used but similar across the schemes for fruit and vegetable intake and objective outcomes. The direction, magnitude, and significance of the US–Canada differences in the prevalence of self-reported health measures were similar for adults from households with and without children, although the differences were more evident among households without children. Among food-insecure households without children, the shares for all health measures except acute care and mortality were greater by the US than by the Canadian scheme; whereas, only the shares with fair or poor overall health and mental health were higher by the US than by the Canadian scheme among food-insecure households with children.

TABLE 3.

Descriptive health profile by FI defined by Canadian and US coding schemes among adults 18 y old and older in Canada (Quebec excluded)¹

		Self-reported measures in CCHS				Objective measures in DAD/CVSD
	n	Low fruit and vegetable intake	Fair/poor health	Fair/poor mental health	Any chronic condition	Acute care	Extensive acute care	Mental acute care	Death in 2 y
Overall sample	403,700	62.8	12.2	6.3	27.0	4.3	0.7	0.2	1.2
Canada: no FI (ref)	358,200	61.7	10.7	5.0	26.7	4.2	0.7	0.1	1.2
Marginal FI	14,300	68.3*	16.2*	10.2*	25.3	4.3	0.8	0.3*	1.0
Moderate FI	20,400	71.3*	23.1*	15.5*	28.8*	4.9*	0.9	0.5*	1.1
Severe FI	10,800	75.6*	39.5*	30.7*	40.2*	<7.0*	<2.0*	<2.0*	1.4
United States: high FS (ref)	358,200	61.7	10.7	5.0	26.7	4.2	0.7	0.1	1.2
Marginal FS	20,600	69.1*	18.2*,a	11.3*,a	26.4a	4.5	0.8	0.3*	1.1
Low FS	14,200	71.6*	23.5*	16.5*,a	28.6*	4.9*	0.9	0.6*	1.0
Very low FS	10,500	75.6*	39.7*	30.9*	41.0*	<7.0*	<2.0*	<2.0*	1.4
Canada: food secure (ref)	372,500	62.0	10.9	5.2	26.6	4.2	0.7	0.2	1.2
Food insecure	31,200	72.7*	28.3*	20.4*	32.1*	5.4*	1.1*	0.8*	1.2
United States: food secure (ref)	378,800	62.2	11.1a	5.3a	26.7	4.2	0.7	0.1	1.2
Food insecure	24,700	73.2*	29.9*,a	22.1*,a	33.1*,a	5.5*	1.1*	0.9*	1.2
With-children subsample	93,600	62.0	7.8	5.3	13.8	2.3	0.3	0.1	0.3
Canada: food secure (ref)	83,200	61.1	6.4	4.1	13.0	2.2	<0.5	<0.5	<0.5
Food insecure	10,400	70.5*	20.6*	16.0*	22.2*	3.8*	<1.0*	<1.0*	<0.5
United States: food secure (ref)	84,800	61.2	6.5	4.2	13.1	2.2	<0.5	<0.5	<0.5
Food insecure	8800	70.7*	21.8*,a	17.2*,a	22.8*	3.7*	<1.0*	<1.0*	<0.5
No-children subsample	310,100	63.2	14.4	6.9	33.4	5.3	0.9	0.2	1.6
Canada: food secure (ref)	289,300	62.4	13.0	5.7	33.0	5.2	0.9	0.2	1.6
Food insecure	20,800	74.3*	33.8*	23.5*	39.4*	6.6*	1.3*	1.0*	1.9
United States: food secure (ref)	294,100	62.6	13.2a	5.9a	33.0	5.2	0.9	0.2	1.6
Food insecure	15,900	75.2*,a	36.3*,a	26.1*,a	41.5*,a	7.0*	1.4*,a	1.2*,a	1.9

¹Values are percentages. Certain values are reported as “< threshold” owing to confidentiality requirements. Numbers of observations (n) are rounded to protect respondents’ identity. ns are lower for self-reported measures than the ones shown in the table owing to missing values. CCHS, Canadian Community Health Survey; CVSD, Canadian Vital Statistics Death Database; DAD, Discharge Abstract Database; FI, food insecurity; FS, food security; ref, reference category for t-test across food insecurity categories.

*Different from reference category (i.e., “food secure,” “no FI,” or “high FS”), P < 0.05.

^aDifferent from Canadian scheme, P < 0.05.

After adjusting for confounding factors, food insecurity was associated in a graded fashion with all health measures irrespective of classification schemes (Table 4, Supplemental Table 3). Marginal food security/insecurity was correlated with all health measures except with mortality in the overall sample. Dichotomous food insecurity was a significant health correlate across the board except for mortality among households with children. The differences in odds of self-reported health status (i.e., fair/poor health, fair/poor mental health, and any major chronic condition) between food-insecure and food-secure adults living without children were slightly smaller in magnitude when the Canadian rather than the US classification scheme was used. No difference in ORs of food insecurity was observed between the US and Canadian classification schemes in relation to fruit and vegetable intake or objectively measured health outcomes.

TABLE 4.

Adjusted ORs for CCHS self-reported health measures, DAD hospitalization, and CVSD death among adults 18 y old and older in CCHS 2005–2017 by FI status, in the overall sample and subsamples with and without children¹

	Self-reported measures in CCHS				Objective measures in DAD/CVSD
	Low fruit and vegetable intake	Fair/poor health	Fair/poor mental health	Any chronic condition	Acute care	Extensive acute care	Mental acute care	Death in 2 y
Overall sample
n	376,800	403,200	395,300	371,000	403,600	403,600	403,600	403,600
Canada: no FI (ref)	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00
Marginal FI	1.22*	1.72*	1.85*	1.46*	1.26*	1.39*	1.65*	1.30
Moderate FI	1.28*	2.22*	2.48*	1.78*	1.41*	1.50*	1.94*	1.16
Severe FI	1.42*	3.84*	4.48*	2.57*	1.69*	1.94*	2.94*	1.66*
US: high FS (ref)	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00
Marginal FS	1.23*	1.83*,a	1.98*,a	1.51*,a	1.29*	1.45*	1.72*	1.20
Low FS	1.29*	2.29*,a	2.61*,a	1.84*,a	1.43*	1.52*	1.99*	1.23
Very low FS	1.45*,a	3.86*	4.50*	2.61*,a	1.69*	1.88*	2.93*	1.69*
Canada: food secure (ref)	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00
Food insecure	1.30*	2.56*	2.90*	1.95*	1.48*	1.60*	2.17*	1.31*
United States: food secure (ref)	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00
Food insecure	1.31*	2.63*,a	2.98*,a	2.03*,a	1.49*	1.59*	2.17*	1.41*
With-children subsample
n	89,400	93,500	92,800	84,200	93,600	92,900	92,300	93,000
Canada: food secure (ref)	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00
Food insecure	1.33*	2.22*	2.68*	1.76*	1.45*	1.79*	1.77*	1.79
United States: food secure (ref)	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00
Food insecure	1.34*	2.25*	2.78*,a	1.78*	1.40*	1.78*	1.87*	2.07
No-children subsample
n	287,500	309,600	302,600	286,800	310,100	310,100	310,100	310,100
Canada: food secure (ref)	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00
Food insecure	1.31*	2.75*	3.03*	2.03*	1.48*	1.59*	2.31*	1.34*
United States: food secure (ref)	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00
Food insecure	1.33*	2.91*,a	3.16*,a	2.16*,a	1.52*	1.59*	2.33*	1.45*

¹Adjusted models adjusted for respondent's sex, age, ethnicity-indigeneity, immigrant status, smoker status, alcohol consumption history, acute care admission in the year before, household income, highest education in household, homeownership, household composition, jurisdictions, and CCHS cycles. Linear and quadratic forms of “days of follow-up” were also adjusted in the models predicting “death in 2 y” to account for differential risk exposure across respondents. Numbers of observations (n) are rounded to protect respondents’ identity. CCHS, Canadian Community Health Survey; CVSD, Canadian Vital Statistics Death Database; DAD, Discharge Abstract Database; FI, food insecurity; FS, food security; ref, reference category for t-test across food insecurity categories.

*Different from reference category (i.e., “food secure,” “no FI,” or “high FS”), P < 0.05.

^aDifferent from Canadian scheme, P < 0.05.

We also looked at the marginal-moderate subsample separately (Table 5, Supplemental Table 4). Compared with the (discordant) “US-marginal Canada-moderate” adults, smaller shares of (concordant) “US-marginal Canada-marginal” adults presented negative self-reported health status (excluding fruit and vegetable intake), whereas greater shares of (concordant) “US-low Canada-moderate” adults reported fair or poor mental health. Once adjusted for confounding factors, “US-marginal Canada-moderate” adults showed significantly higher odds of fair/poor health, fair/poor mental health, and reporting any major chronic condition than the “US-marginal Canada-marginal” adults and lower odds of these states than the “US-low Canada-moderate” adults. No difference was found in fruit and vegetable intake or objectively measured health outcomes.

TABLE 5.

Descriptive health profile and adjusted ORs for health measures by FI status defined by Canadian and US coding schemes among adults aged 18 y and older in Canada (Quebec excluded) in the marginal-moderate subsample¹

	Self-reported measures in CCHS				Objective measures in DAD/CVSD
Marginal-moderate subsample	Low fruit and vegetable intake	Fair/poor health	Fair/poor mental health	Any chronic condition	Acute care	Extensive acute care	Mental acute care	Death in 2 y
Mean percentages
n	32,000	34,400	33,800	28,900	34,400	34,400	34,200	34,400
Total	70.1	20.3	13.3	27.4	4.7	0.8	0.4	1.1
US-marginal Canada-marginal	68.3	16.2*	10.2*	25.3*	4.3	0.8	0.3	1.0
US-marginal  Canada-moderate (ref)	70.9	22.6	13.8	28.8	5.0	1.0	0.3	1.3
US-low Canada-moderate	71.7	23.6	16.4*	28.9	4.9	0.9	0.6	1.1
Adjusted ORs
n	32,000	34,400	33,800	28,900	34,400	34,400	34,200	34,400
US-marginal Canada-marginal	0.96	0.84*	0.82*	0.91*	0.92	0.89	0.92	1.31
US-marginal Canada-moderate (ref)	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00
US-low Canada-moderate	1.06	1.14*	1.20*	1.13*	1.02	0.96	1.10	1.27

¹“US-marginal” and “US-low” stand for “marginal food security” and “low food security” in the US classification scheme, respectively. “Canada-marginal” and “Canada-moderate” stand for “marginal food insecurity” and “moderate food insecurity” in the Canadian classification scheme, respectively. All logistic models adjusted for respondent's sex, age, ethnicity-indigeneity, immigrant status, smoker status, alcohol consumption history, acute care admission in the penultimate year before interview, household income, highest education in household, homeownership, household composition, jurisdictions, and CCHS cycles. Linear and quadratic forms of “days of follow-up” were also adjusted in the logistic models predicting mortality to account for differential risk exposure across respondents. Individuals categorized as either no or severe FI in Canada, or high or very low food security in the United States, are excluded. Numbers of observations (n) are rounded to protect respondents’ identity. CCHS, Canadian Community Health Survey; CVSD, Canadian Vital Statistics Death Database; DAD, Discharge Abstract Database; FI, food insecurity.

*Different from reference category (i.e., “US-marginal Canada-moderate”), P < 0.05.

Discussion

Using population survey data from 2005–2017, we found 1.66% of the Canadian adults (excluding Quebec) were in food-insecure households by the Canadian classification scheme but in food-secure households by the US scheme. The discordance affected >20% of the Canadian-defined food-insecure households. The presence of children did not affect the direction or magnitude of the discordance. Choice of scheme barely affected the pattern of associations between food insecurity and adult health, although the magnitude of association tended to be higher under the US than under the Canadian scheme for self-reported health measures except fruit and vegetable intake. The health profile of the “US-marginal Canada-moderate” discordant group was in between their 2 concordant neighboring groups, with minor differences in self-reported health, mental health, and chronic disease status and no difference in fruit and vegetable intake, acute care, or mortality.

Our findings illustrate the nontrivial impact of classification schemes on the population distribution of household food insecurity. The difference calls for caution when comparing the food insecurity prevalence between the United States and Canada and when interpreting research results on food insecurity. The sizable discordance among households without children relative to those with children and the small discordance found in very low food security and severe food insecurity among households with children both suggest that the discrepancy in prevalence estimates may be driven more by differences in the thresholds than by dimensionality. The share with “no food insecurity” in Canada is comparable with that with “high food security” in the United States because both are based on a raw score of 0 on the food security scales. “Severe food insecurity” in Canada is also largely comparable with “very low food security” in the United States given their negligible discordance.

Survey design could also contribute to the US–Canada disparity in food insecurity prevalence estimation. In 2005–2014, the CCHS used food insufficiency (a simplified food hardship indicator capturing a more severe form of food insecurity) as an additional screener for food insecurity to override question-skipping patterns within the HFSSM (43). Households who indicated food insufficiency (i.e., sometimes or often not having enough to eat in the last year) but did not respond affirmatively to any of the first 5 questions on the HFSSM (and would not be asked any further questions based on following the question-skipping patterns embedded in the module) were nonetheless asked the next questions on the module. Using food insufficiency to screen people into the respondent universe of the food insecurity questionnaire, a practice discontinued in 2015, must have increased the number of households classified as food-insecure, although the magnitude of the effect is likely small. Whereas, food sufficiency was used as a criterion of exclusion (as opposed to inclusion) from the respondent universe of the food security questionnaire in the Current Population Survey (CPS), the official data source for food security statistics in the United States. Specifically, the CPS excluded from the food security module households with income ≥185% of the federal poverty line that were food-sufficient and without food stress (i.e., never ran short of money for food in the past year) (47). The application of the screeners in the CPS may have increased the number of households categorized as food secure, although the extent to which screening affects the comparability of food insecurity statistics between Canada and the United States warrants further study.

The similar associations of food insecurity (binary and categorical alike) with health measures across classification schemes suggest that Canadian and US studies on food insecurity and health may yield results that are largely comparable to the extent that socioeconomic (e.g., family structure) and contextual (e.g., unemployment rate) confounders are controlled. The unadjusted prevalence of self-rated health and adjusted ORs of food insecurity on self-rated health were slightly lower under the Canadian than under the American classification scheme, suggesting that the health disparity by food insecurity status found in Canadian studies may be somewhat conservative compared with the American ones. However, our results suggest that the difference should be small, with no impact on the observed pattern of associations. The fact that statistically significant differences in ORs across classification schemes were only observed for subjective health measures may be partly due to the much lower frequency and associated statistical power of objectively measured events (e.g., past-year hospitalization) than of subjective ones (e.g., having a chronic condition). Substantially different findings on the relation between food security status and health across the 2 countries are rare in the literature and most likely stem from US–Canada policy and contextual disparities rather than discrepancies in food security classification schemes (20, 22). For instance, universal health care in Canada and nontrivial out-of-pocket expenses on health care utilization in the United States may partly explain the significant association between food insecurity severity and acute care admission in Canada and the absence thereof in the United States (20, 22).

The predicted health indicators of the “US-marginal Canada-moderate” discordant group lie in between its 2 neighboring concordant groups as one would expect based on the food security continuum. This suggests that the discordant group could be reasonably collapsed into either of its 2 neighboring concordant groups for health studies, further buttressing our argument that health prediction is insensitive to the food insecurity classification scheme applied.

Our findings lend support to the call to recognize marginal food security/insecurity as “food insecure” rather than “food secure” (36, 37). Whether using the US or Canadian scheme, adults in marginal food security/insecurity consistently had worse health than their counterparts with no indication of food insecurity. This resonates with other empirical findings on the “marginal” group in both countries (11, 19, 20, 22, 33–35). Separating the “marginal” from the “high” food security group would further facilitate United States–Canada comparison of food insecurity prevalence, providing an important benchmark for both countries’ efforts toward accurate portrayal of the issue.

The HFSSM is used around the globe (48–59), but, unlike in Canada, significant changes to the instrument have often taken place during its adaptation to other country contexts. For instance, Brazil uses a 14- to 16-item scale to measure household food insecurity in the past 3 mo (54, 58); the United Kingdom uses the 10-item adult-referenced scale regardless of the presence of children (56, 60). The dimensionality and threshold differences discussed here may not apply to adapted versions of the HFSSM elsewhere; separate studies are required to examine their implications. However, our finding of a sizable discrepancy in prevalence estimates from the application of 2 relatively similar classification schemes to the same instrument highlights the need for caution when contrasting “food insecurity” findings across different studies.

Strengths and limitations

This study's strengths included the use of a large, population-representative sample and multiple well-established self-reported and objective health measures. We also recognized several limitations. We did not replicate the findings in a US data set; therefore, it is unclear how the between-country differences in classification would affect relations between food insecurity and health in the United States. The limited dietary information contained in the CCHS precluded a more robust examination of nutrition measures beyond intake of fruits and vegetables. Our data did not include Quebec, which represented over one-fifth of Canada's population. Thus, our estimates were not nationally representative. But we do not anticipate significant changes even if Quebec is included because the food insecurity rate there is similar to the national average (1). We limited our analysis to adults; therefore, our findings may not necessarily apply to children <18 y old. Finally, our analysis focused on health measures, yet classification differences may potentially affect results of policy evaluation and program interventions with respect to food insecurity as well. Research is needed to determine the role of classification schemes in shaping policy studies.

Conclusion

There is substantial discordance in food insecurity classification between the United States and Canada, preventing direct cross-national comparison of food insecurity rates. However, the associations between food insecurity and various subjective and objective health measures are mostly insensitive to choice of classification schemes. Monitoring of food insecurity and related health research will benefit from separating marginally food-secure/insecure households from those with no indication of food insecurity.

Notes

Supported by Canadian Institutes of Health Research grant PJT-153260 (to VT).

Author disclosures: The authors report no conflicts of interest.

The funder, Canadian Institutes of Health Research, had no role in the design and conduct of the study; collection, analysis, and interpretation of data; writing of the article; or the decision to submit the article for publication.

Supplemental Figure 1, Supplemental Tables 1–4, and Supplemental Questionnaire 1 are available from the “Supplementary data” link in the online posting of the article and from the same link in the online table of contents at https://academic.oup.com/jn/.

Abbreviations used: CCHS, Canadian Community Health Survey; CVSD, Canadian Vital Statistics Death Database; DAD, Discharge Abstract Database; HFSSM, Household Food Security Survey Module.

Data Availability

Data described in the article, code book, and analytic code will not be made available because they are accessible by authorized researchers only at Statistics Canada's Research Data Centres.