Food insecurity and the risk of depression in people living with HIV/AIDS: a systematic review and meta-analysis

We conduct this systematic review and meta-analysis in accordance with the preferred reporting items for systematic review and meta-analysis guidelines (PRISMA) [15]. We utilized a pre-defined protocol for search strategy, data extraction, study selection, as well as analysis.

Three authoritative electronic databases (EMBASE, PubMed, and Scopus) were systematically searched for pertinent studies. We searched without restriction on the date of publication. The systematic literature search was conducted in May 2019. The search terms and keywords included (Food insecurity OR food insufficiency) AND (HIV OR Human immune deficiency Virus OR AIDS OR Acquired immune deficiency syndrome) AND (depression OR depressive symptom OR depressive disorder). We also manually searched to identify additional relevant studies.

Articles satisfying the following criteria were included in this study: First, all observational study (case–control, cross-sectional or cohort stud). Second, the exposure of interest was food insecurity. Third, the outcome was depression. Fourth, the study population was PLWHA. Fifth, studies that reported relative risks (RR) or odds ratio, estimates with the respective 95% confidence intervals (CIs), or studies that reported data to calculate these. In this review, we excluded editorials, comments, case reports, reviews, letters, abstracts presentations, as well as studies published in a non-English language.

Two reviewers (MS and LT), in an independent manner, used a predesigned standard form to extract the data, which included the first author’s name, publication year, the study design, country, confounders adjusted for, risk estimate (OR/RR) and their 95% CI as suggested by PRISMA guidelines [16]. We resolved disagreements were by consensus.

The Newcastle–Ottawa quality evaluation scale (NOS) [17] was used to assess the quality of the included studies. The NOS scale evaluates the quality of the included studies in three areas such as comparability between the groups, recruitment of the participants, and assessment of exposure and outcome. For the cross-sectional studies, we used a modified version of NOS [18]. In fact, studies were not excluded based on the quality assessment score alone.

Comprehensive Meta-Analysis (CMA) software version3 was employed to conduct the meta-analysis [19]. In those studies that reported multiple estimates, we used the estimate with the most extensive adjustment and with the highest degree of food insecurity. To account for the heterogeneity across the studies, the random effect model was used to combine the effect estimate from included studies [20]. Q and the I² statistics were used to assess heterogeneity [20]. The I² statistics values such as 25, 50, and 75% represented low, moderate, and high heterogeneity respectively [21]. All the reported probabilities were two-sided. To assess the key studies that exerted a considerable impact on the heterogeneity we carried out a leave-one-out sensitivity analysis [22]. We also performed subgroup and sensitivity analysis to compare the risk between the groups as well as determining the potential source of heterogeneity between the studies. We evaluated a publication by funnel plot and Egger’s regression test [23].

The search strategy resulted in 278 studies. Six additional relevant references were identified through our manual search. Our review of these studies by title, duplicate, and abstract resulted in the exclusion of 264 studies, as they did not meet the predefined inclusion criteria. Our further screening of a full text of the remaining 20 articles resulted in the exclusion of further 13 studies. Therefore, seven studies were found eligible for the final analysis (Fig. 1).

Table 1 shows the characteristics of the included studies. The included studies were published between December 2011 [24] and June 2018 [11]. Most of the included studies (5 studies) were conducted in the USA [11, 14, 25‐27], one study conducted in Uganda [24], and one in Ethiopia [28]. Five were cohort studies [11, 14, 25‐27] and two were cross-sectional studies [24, 28]. Two studies adjusted for the possible confounding effects of drug use [11, 14], two adjusted for ART drug use [11, 24], and two adjusted for social support [14, 24].

Regarding the tools used to assess depression, from the total, four studies used the Center for Epidemiologic Studies Depression Scale (CESD) [11, 25‐28], one study used Mini-International Neuropsychiatric Interview (MINI) [24], one study used Beck depression inventory (BDI) [14] and one study used Burnam depression screen (a short form of CESD) [14]. Of these four instruments used to assess depression, three of them are screening [29‐31] and one is a diagnostic instrument [32]. Regarding the instruments used to assess food insecurity, two studies used the Household Food Insecurity Access Scale (HFIAS) [27, 28], two studies used the Household Food Insecurity Survey Module (HFSSM) [11, 26], one study used Radimer/corner questionnaire of hunger and food insecurity [14], one study used Core Food Insecurity Module [25], and one study used self-reported food insecurity by participants [13].

In this review, we used NOS, a 9-point scoring system to evaluate the study quality. Accordingly, all the included studies were good quality studies (the NOS score for the included studies ranges between eight and nine from the total 9 points) (see Additional file 1: Table S1).

Figure 2 shows the forest plot indicating the relative risk and 95% CI of each study as well as the overall pooled relative risk. To account for the observed heterogeneity between the included studies (I² = 84.88%; Q = 39.69; df = 6; P < 0.0001), we employed a random effect model. The meta-analysis of seven studies showed that the risk of developing depression was significantly higher in those PLWHA who were food insecure as compared to those who were food secure [RR 2.28 (95% CI 1.56–3.32)].

In our subgroup analysis by the study design, the risk of developing depression based on cohort studies was 2.06 (95% CI 1.36–3.12)), whereas based on cross-sectional studies was 3.24 (95% CI 1.85–5.68). The heterogeneity was significant for cohort studies (I² = 87.62%; Q = 42.30; df = 4; p < 0.0001), but not for cross-sectional studies (I² = 0.00%; Q = 0.23; df = 1; p = 0.630) (see Fig. 3).

In our stratified analysis by the level of food insecurity, the risk of depression was 1.83 (95% CI 1.09–3.07)), 1.95 (95% CI 0.85–4.52)), 2.59 (95% CI 1.03–6.48)) respectively for mild, moderate and severe food insecurity. The observed differences in risk estimates between the groups were not statistically significant (P = 0.813) (see Table 2).

The funnel plot and Egger’s regression tests (B = 2.60, SE = 0.96, P = 0.042) provided evidence of substantial publication bias for the association between food insecurity and the risk of depression in PLWHA (Fig. 4).

To identify the possible effects of drug use, we conducted stratified analysis by restricting the analysis to the studies that adjusted for the potential confounding effects of drug use. In this analysis, an increased risk of depression was observed in PLWHA who were food insecure in studies that accounted for the effects of drug use [RR 1.63 (95% CI 1.27–2.10)] as it was in the studies with no adjustment [RR 2.71 (95% CI 1.46–5.00)]. In this analysis, the observed difference in risk estimates of depression (RR) between the groups was not significant (P = 0.134). The reported heterogeneity across the studies was also not significant in studies that adjusted for the confounding effects of drug use (I² = 0.00%; Q = 0.034; df = 1; p = 0.850) but it was significant in those that did not adjust for the effects of drug use (I² = 89.83%; Q = 39.30, df = 4, P < 0.001) (see Table 2).

We further conducted the sensitivity analysis by restricting the analysis to studies that adjustment for confounding effects of lack of social support. We found an increased risk of depression in PLWHA who were food insecure in studies that accounted for the effect of social support [RR 2.21 (95% CI 1.18–4.16)] as well as in studies that did not account for confounding effects of social support [RR 2.32 (95% CI 1.48–3.63)]. This analysis resulted in no significant heterogeneity between studies that adjusted for social support (I² = 18.02%; Q = 1.22; df = 1; p = 0.269) but it was significant in those that did not adjust for social support (I² = 89.07%; Q = 36.61, df = 4, P < 0.001) (see Table 2).

Finally, we performed the analysis by restricting the analysis to studies that accounted for potential confounding effects of ART drug treatment because evidence suggests that antiretroviral therapy side effects were positively associated with depression in PLWHA in previous studies  [33]. We found an increased risk of depression in PLWHA who were food insecure in studies that accounted for the effect of ART drugs [RR 1.96 (95% CI 1.17–3.28)] as well as in studies that did not account for the effect of ART drugs [RR 2.43 (95% CI 1.32–4.48)]. The heterogeneity between studies that adjusted for ART was not significant (I² = 51.67%; Q = 2.06; df = 1; p = 0.150) but it was significant in those studies that did not adjust ART drugs (I² = 88.93%; Q = 36.14, df = 4, P < 0.001) (see Table 2).

We also conducted a leave-one-out sensitivity analysis for further examining the possible cause of heterogeneity in the analysis of food insecurity and depression in PLWHA. This analysis resulted in a pooled estimated relative risk (RR) ranging between 1.88 (95% CI 1.41–1.70) and 2.58 (95% CI 2.52–4.29) after the deletion of a single study. This finding indicates that our findings were robust and not dependent on a single study (Additional file 2: Table S2).

This systematic review and meta-analysis assessed the risk of depression in PLWHA who were food insecure across five cohorts and two cross-sectional studies. Our final analysis demonstrated that there was a positive and significant association between food insecurity and greater risk of depression in PLWHA (RR = 2.28) that was unaffected by the level of adjustment for ART drug use, the degree of social support, as well as substance use (drug use). When we limit the analysis by the level of food insecurity, the risk was higher for severe food insecurity (RR = 2.59) followed by moderate (RR = 1.95) and mild food insecurity (R = 1.83), which supports the robustness of the main analysis. This finding suggests the necessity of the application of early screening and intervention strategies of food insecurity in PLWHA.

However, the included cross-sectional studies, the sample sizes, and the level of adjustment for the potential confounding factors must be considered. The level of adjustment factors was inconsistent in the included seven studies. Drug use (2 studies), age of participants (4 studies), on ART (2 studies), and social support (2 studies) were the most common potential confounders taken into account in the included studies. Only one study accounted for the possible confounding effects of the previous history of depression [27]. This study found a significant and week association for moderate food insecurity and moderate association for severe food insecurity, but the association was not significant for mild food insecurity. This result suggests the possibilities that the association seen in studies with a lower level of adjustment could be due to chance or the effects of confounding. Supporting this view, a substantial body of research showed a greater risk of depression in those PLWHA who substance users were, had a previous history of depression, as well as poor social support. Thus, in studies, which did not account for the effect of the above factors, the observed association between food insecurity and greater risk of depression could be due to the confounding effect of unmeasured drug use, ART drugs as well as lack of social support.

In fact, the robustness of an increased risk of depression in PLWHA with food insecurity was supported by our s analysis that we conduct in the current study: Firstly, the robustness of the observed association between food insecurity and depression was supported by our dose–response analysis. In this analysis, we found a greater risk of developing depression in those participants with severe food insecurity followed by moderate and mild food insecurity. These findings suggest the possible causal association between food insecurity and depression. Secondly, the robustness of the association observed in the current study was also supported by the sensitivity analyses that we conducted restricting the analysis to studies that controlled the confounding effects of drug use, ART drugs, and social support. In this analysis, we found the increased risk of depression in those PLWHA studies which accounted for the possible effects of drug use [RR 1.63 (95% CI 1.27–2.10)] social support [RR 2.21 (95% CI 1.18–4.16)] as well as those studies that adjusted for effects of ART drugs [RR 1.96 (95% CI 1.17–3.28)].

Even though our meta-analysis resulted in a significant association between food insecurity and depression in PLWHA, the observed differences across the seven studies led to a significant level of between-study heterogeneity in our final analysis. The stage of the disease, the country, and the exposed population differed on a number of characteristics which may have contributed to the variance in the risk of depression in those PLWHA who were food insecure. Nevertheless, our leave-one-out sensitivity analysis indicated that the risk of developing depression remained virtually unchanged from the main analysis. This finding suggests that our findings were strong and not significantly influenced by a single study. Additionally, our subgroup analysis and sensitivity analysis by the degree of food insecurity as well as the level of adjustment to a range of confounding factors supported the robustness of our findings. Moreover, in order to make the finding of the current meta-analysis meaningful, we have used a random-effects model to pool risk estimates from the individual studies. It is widely held that the summary effect estimates in the random effect model meta-analysis are more conservative than fixed effect summaries in the epidemiological meta-analysis, suggesting the robustness of our findings [20].

In our study, two of the evidence came from cross-sectional studies therefore the possibility that depression in PLWHA may lead to rather than precede the diagnosis of food insecurity might be considered. This is because epidemiologic evidence reported a reciprocal relationship between food insecurity and depressive symptoms [34]. Supporting our view, scientific studies have found that depressive symptoms were significantly associated with food insecurity [34‐36]. In fact, this may not be a major concern in our study because when we limit our analysis to cohort studies where exposure (food insecurity) certainly precede the outcome (depression), we found a significant association between food insecurity and the increased risk of depression [RR 2.06 (95% CI 1.36–3.12)].

There is a range of explanations for the associations between food insecurity and increased risk of depression in PLWHA. Firstly, food insecurity is linked with incomplete HIV viral load suppression and less immune reconstitution in PLWHA, which in turn linked with a higher risk of depression [37‐39]. Secondly, food insecurity is also associated with a significant reduction in CD4 count, which has been consistent, associated with a greater risk of depression in PLWHA in previous studies [40‐42]. Thirdly, the rates of underweight are higher in food-insecure PLWHA as compared to food-secure people and underweight has been associated with a higher risk of depression in several epidemiologic studies [43, 44]. Furthermore, food insecurity is associated with a higher risk of opportunistic infections and other comorbid conditions, which are among the major risk factors of depression among PLWHA [45, 46]. Finally, food insecurity is associated with reduced social capital, and higher levels of (or increased level of) social isolation, stigma, stress, and loneliness, which are in turn linked with increased risk of depression in those food-insecure people [47, 48].

This systematic review and meta-analysis have several strengths: Firstly. First, we utilized a predesigned search strategy, data abstraction, quality assessments, analysis, data extraction, and quality assessment to minimize the reviewer’s bias. Secondly, we performed a sensitivity and subgroup analysis based on drug use, level of social support, and use of ART drugs to identify the small study effect and the risk of heterogeneity in analyses of the risk of depression and those with food insecurity. Thirdly, we use a standard tool to evaluate the quality of the included studies (NOS) and our evaluations showed that the methodologic qualities of the included studies were good.

However, the current review also has some limitations: (1) our meta-analysis resulted in significant associations between food insecurity and depression, but the degree of adjustments was inconsistent across the studies and important confounding variables including the effects of drug use, social support, previous history of depression, as well as ART drugs, were not adjusted in most studies. (2) Our final analysis showed significant heterogeneity between the studies. However, the observed variations between the studies may not be a major issue because our sensitivity and subgroup analysis resulted in a greater risk of depression in those food-insecure participants in both cohort 2.06 (95% CI 1.36–3.12)) and cross-sectional 3.24 (95% CI 1.85–5.68) studies. (3) The inclusion of a relatively low number of studies in our subgroup and sensitivity analysis is the other limitation of this review, which might reduce the precision of the reported estimate. (4) Pertinent studies published in a language other than English may have been missed. Finally, we found a significant publication bias the possibility of publication bias indicating the possibilities of remained studies due they may not be reported or published based on their findings or other factors. (5) We have not conducted the literature search in PsycINFO to identify studies relevant to the current review, which is one of the authoritative databases in mental health. In fact, we have conducted our systematic search for relevant studies in the three commonly used databases in mental health such as EMBASE, PubMed, and Scopus. We have also tried to include all the potential articles relevant to the review by our manual search and we believe that we have included the required important articles to estimate the pooled relative risk for the association between food insecurity and the risk of depression in people living with HIV/AIDS.