Eligible participants at baseline were 40 years of age and older, English or Spanish speaking, able to provide informed consent, non-institutionalized at the time of data collection, had reliable contact/residential information, were semi-permanent or permanent residents of the island for 10 or more years, and had no plans to permanently relocate in the next 5 years.

The sampling methodology for the baseline ECHORN cohort (n = 2,961) has been described in detail elsewhere (35). Briefly, in Trinidad, Puerto Rico, and Barbados, stratified multistage probability sampling was used to empanel the baseline cohort between 2013 and 2018. In the US Virgin Islands simple random sampling was used across the islands of St. Thomas St. Croix and Saint John. Participants visited a community assessment center, centrally located on each island site, for their baseline assessment. After informed consent was obtained, participants were asked to complete a health survey, a clinical assessment, and provide a blood sample for immediate testing to identify markers of disease. The health survey consisted of questions pertaining to health status and chronic disease history, health behaviors, diet, household food insecurity, access to health care, migration history, social support, health networks, neighborhood factors, and demographic information. The cross-sectional sample used in this analysis included all participants with household food insecurity data at baseline and non-missing values for the examined risk factors (n = 1,939).

The primary outcome was household food insecurity as measured by the Latin American and Caribbean Food Security Scale (or ELCSA by its Spanish acronym) (34). The 9-item ELCSA scale for adults (Table 1) is a household-level experiential food security scale and is scored by assigning 1-point to each affirmatively answered yes/no question. Next, responses are divided into the following categories: food secure (score of 0), mild food insecurity (score of 1–3), moderate (4–6), and severe food insecurity (score of 7–9). Respondents with moderate and severe food insecurity scores (4–9) were grouped into a single category.

Sixteen risk factors were chosen and examined based on existing literature and potential risk factors specific to this population, based on experience working in the region. Demographic factors included age at baseline interview (continuous), sex, level of education, perceived economic status, marital status, island site (Puerto Rico, USVI, Trinidad, or Barbados), home ownership status (Yes/No), and whether the participant had moved in the past year (Yes/No). Sex was measured on the baseline survey using the following question, “What sex were you at birth?” Educational attainment was measured using the question, “What is the highest year of school that you completed?” Responses were categorized into less than high school (or secondary school), high school graduate, some college, and college and higher. Perceived economic status was measured using an adapted version of the World Gallup Poll^® question: “Please look at this figure, with steps numbered from 1 at the bottom to 10 at the top. Suppose the top of the ladder represents the richest people of this island and the bottom represents the poorest people of this island. Taking into consideration your current personal situation, what is the number of the step on which you would place yourself?” Responses ranged from 1 poor to 10 high and were categorized into bottom, middle, and top quantiles. Marital status was measured by asking “What is your current relationship status” and responses categorized into married, single, separated/divorced, or widowed.

Psychosocial factors included were emotional support, and depression. Emotional support was measured using the PROMIS Emotional Support short form (36). Responses were dichotomized (Yes/No) as to whether each participant had a low emotional support score, meaning less than 12. Depression (Yes/No) was measured by the Patient Health Questionnaire (PHQ-2) (37, 38).

Behavioral factors included current smoking and self-reported physical health scores. Current smoking status (Yes/No) was measured using two variables: “Have you EVER smoked any tobacco product, such as cigarettes, cigars, or tobacco pipe? Yes/No. Those that answered Yes were asked “Do you still smoke cigarettes, cigars, or tobacco pipe regularly? By regularly we mean at least 20 cigarettes or 1 cigar or half an ounce sachet of loose tobacco per month.” The PROMIS Global Physical Health score was used to assess participant reported physical health (39, 40). The score ranges from 4 to 20, with 4 being poor health and 20 excellent health. The score was created using 4 items (Table 2).

Environmental factors included fruit and vegetable availability and quality, mode of transportation to the grocery store, and water source as a proxy for water security. Fruit and vegetable availability and quality were measured as follows: “Thinking about food resources in your neighborhood, how often are a large selection of fresh fruits and vegetables, excluding provisions, available in my neighborhood?” and “Thinking about food resources in your neighborhood, how often are the fresh fruits and vegetables in your neighborhood of high quality?” Responses were dichotomized into never/rarely/sometimes or usually/always. Mode of transportation to the grocery store was measured with a single item: “What is the most typical way you travel to the store for your groceries?” and responses were dichotomized: drive own car/ride with friend/family or take the bus/taxi/bike/walk. Water insecurity was measured by a single item asking about source of potable water: “What is the main source of water supply for members of your household? This item was used as a proxy for water insecurity. Responses were dichotomized as water secure (water piped into dwelling) and water insecure (water not piped into dwelling).

Rasch modeling was used to assess the ELCSA scale’s psychometric properties since this was the first time the scale was being used in the ECHORN Cohort. The Rasch model is a 1-parameter item response model, a modeling technique that is consistently applied in studies using the ELCSA and other food insecurity scales (26, 41–43). RASCH modeling was completed using the full data set (n = 2,961). The model was run in the following ways: (1) On the full sample using all 9 ELCSA items; (2) on the full sample using 8 of 9 ELCSA items (removing number 9—begging); (3) on the sample from each island (using all 9 items and 8 items as above); (4) removing individual participants identified as outliers—both on the full sample using all 9 items and for each island site using the 9-item scale. Unidimensionality of the scale by island was further assessed using Differential Item Functioning (DIF). DIF analysis was performed to compare scale performance for each island to the full sample. Measure, Infit values, and differences in item performance were assessed by island site. A detailed description of the RASCH results can be found in Supplementary material.

Next, univariate and bivariate analyses were conducted to determine the prevalence of household food insecurity, describe the overall sample by each risk factor, and to examine the association between household food insecurity and each risk factor. Study variables were summarized using means and standard deviations or frequency distributions for the total sample and by level of household food insecurity. The analysis of variance and the chi-square test were used to examine the association between potential risk factor variables and household food insecurity. Tetrachoric and polychoric correlation coefficients were also examined to determine whether collinearity existed between specific study variables: education, perceived economic status, water supply (which may be a marker for economic status in the Caribbean region), and home ownership. Finally, multivariable ordinal logistic regression was then used to determine the association between household food insecurity (mild to moderate/severe) and each risk factor holding all other variables constant. The analysis was first conducted for the full sample, then stratified by sex as women are more likely to experience HFI than men (44).

Rasch modeling of the ELCSA scale in the ECHORN sample indicated that the full 9-item scale for adults used with the full cohort (rather than by island site) was the best fit. The Cronbach’s Alpha for the scale was 0.90. Figure 1 shows the ELCSA scale item infit values. Each item is shown along the X-axis and the item infit value on the Y-axis. Infit is a fit statistic that is less sensitive to outliers and more sensitive to observations near the respondent’s ability level (45). Acceptable infit values range from 0.7 to 1.3 (46). All 9-items of the ELCSA scale had acceptable infit values in the ECHORN sample. This means that the infit values demonstrate that the items measure the same construct and are independent of one another. Further information on the psychometric validity of the scale—as demonstrated through item prevalence, item severity, and differential item functioning by island site—is presented in Supplementary material.

The final sample size consisted of 1,939 individuals with a baseline household food insecurity score and non-missing data for the examined risk factors. More than one-quarter of the sample (27.3%) experienced some level of household food insecurity (17.0% mild, 10.3% moderate or severe). Respondents were on average 57 years of age (S.D. 10.5) and nearly two-thirds were female. Nearly one-third had no high school education, 41% were married, nearly two-thirds owned their own home and 50% rated themselves in the middle economic quartile of their respective island. Nearly 15% of respondents did not have water piped directly into their homes (Table 3, total column). Correlation coefficients ranged from −0.04 to 0.23, indicating that collinearity was not present between education, perceived economic status, water supply, and home ownership.

In bivariate analyses, all risk factors examined, except for sex were significantly associated with household food security status (Table 3). Those who were food secure were 4.8–6.3 years older than those who had mild or moderate/severe food insecurity. Those who reported that they had not completed high school or college, were single, did not own a home, moved in the past year, or had a self-reported economic status in the bottom quantile were more likely to report food insecurity than their counterparts. Participants in Trinidad were more likely to report food insecurity compared to the other three island sites. The prevalence of water insecurity was 14.5% in this sample. A dose response relationship with water insecurity was found such that food insecurity worsened as the proportion of respondents who did not have potable water piped directly into their dwelling increased.

The multivariable analysis modeled the odds of worsening household food insecurity (mild to moderate/severe; Table 4). Increasing age was protective against experiencing worsening HFI in this sample. All variables except sex, education, marital status, smoking status, and residing in Puerto Rico (compared to Barbados) were significant predictors of HFI in the adjusted model. Sex was not associated with HFI in either the unadjusted or the adjusted model. In the adjusted model, those who were depressed had 71% increased odds (OR: 1.71; 95% CI: 1.28–2.28) of worsening HFI compared to those who were not depressed. Those who did not have water piped directly into their dwelling had 59% increased odds of worsening HFI compared to those with water piped directly into their home (OR 1.59; 95% CI: 1.17–2.17).

In the sex stratified analysis, age was a significant protective factor against worsening HFI; however, there was a greater protective effect for women compared to men (Table 5). For every 1-year increase in age, women were 6% less likely to experience worsening HFI, while men were 3% less likely. Self-reported economic status in the middle or bottom quantiles, not owning a home, having moved in the past year, low emotional support, lack of car or ride to get to the grocery store, and lack of water piped directly into the home were significantly associated with worsening HFI among both men and women. Education, marital status, current smoking status, and access to high quality foods were not associated with worsening HFI in men or women.

Depression, food availability, self-rated physical health, and island site were significantly associated with increased odds of worsening HFI for women, but not for men. Women who screened positive for depression had 72% increased odds of worsening HFI (OR: 1.72; CI: 1.22–2.44) compared to those without depressive symptoms. This is compared to 59% increased odds among men; however, this effect estimate was not statistically significant (OR: 1.59; CI: 0.92–2.74).

We used a validated and well-tested measure of household food insecurity for this research and confirmed its robust psychometric properties in the ECHORN cohort. We also examined known and potential risk factors for HFI (based on our knowledge of the region), which allowed us to identify important risk factors specific to the populations under study. This research fills a gap in the literature with respect to identifying and understanding risk factors for household food insecurity in the Caribbean region, and strongly calls for applying the lessons learned in these settings to the design of similar policy relevant studies in other regions of the world. Importantly, we present evidence of a link between source of potable water, a proxy for water security, and household food insecurity in the ECHORN cohort. These findings have important implications for understanding how to improve the governance of food and water security systems and the coordination needed between them.

With respect to study limitations, the cross-sectional nature of this analysis only allows us to draw conclusions about the association between the studied risk factors and household food insecurity in the region, without comment on causality. In addition, these findings pertain to the ECHORN cohort study sites only and cannot be extrapolated to other nations or territories in region.