This study utilizes the infrastructure developed in partnership with the State of Hawaiʻi to gather data from 2,200 adult Hawaiʻi residents in May 2022 (baseline) and 1,630 participants in November 2022 (follow-up survey). The research focused on adult individuals of working-age between 18 and 65 years old. We removed individuals with incomplete questionnaire responses, resulting in a sample size of 2,270 observations to 1,430 unique respondents over the two survey periods. The surveys contained over 100 questions covering demographics, employment status, vaccination status, attitudes toward vaccination, mental health, and food insecurity among other data.

The study was approved by the University of Hawaii Institutional Review Board under study number 2021–00989, and all participants gave informed consent. All methods were performed in accordance with relevant guidelines and regulations.

In this study, measures of socioeconomic determinants included demographic characteristics, education level, employment status, household income per person, and food insecurity. Demographic characteristics included race, age, sex, and civil status. We also included individual health outcomes and behaviors such as smoking and drinking, pre-existing health conditions, safety perception, and historical lingering effects from COVID-19. These metrics and those used in the empirical model are shown in Supplementary Table S1.

The metrics used in this study and the variables included in the empirical model are detailed in Supplementary Table S1. These metrics include the specific cut-off points used to define the mental health outcomes—depression, low self-esteem, and suicidal ideation—as binary variables. Each mental health symptom was coded as a binary variable, with a value of 1 indicating the presence of the symptom and 0 indicating its absence. The criteria for these binary classifications were established based on validated scales widely used in public health research. For example:

Depression was assessed using the Center for Epidemiological Studies Depression (CES-D) 10 item scale, with a score of greater than 10 indicating the presence and high risk of depressive symptoms and coded as 1. This threshold is consistent with clinical guidelines for identifying individuals at risk of depression (13–16).

Low self-esteem was measured using the Rosenberg Self-Esteem Scale (17, 18), with a cut-off score of 15. The individuals are classified as low self-esteem with the score of 15 or less, and high or normal self-esteem with the score of above 15. The threshold is based on prior studies that validated its sensitivity and specificity for detecting low self-esteem (19, 20).

Suicidal ideation was identified through a single-item question validated in previous research, with affirmative responses coded as 1. This standardized question has been used in surveys by the Centers for Disease Control and Prevention, including the “Adolescent Behaviors and Experiences Survey” (21) and the “Youth Risk Behavior Survey” (22).

These thresholds were selected to align with established clinical and research standards in mental health assessment. By using these criteria, we ensured that the binary representation of mental health outcomes accurately reflects meaningful distinctions between affected and non-affected individuals.

Furthermore, all predictor variables in the regression models, such as food insecurity, employment status, and community safety perceptions, were coded based on established definitions and thresholds from the literature (23). For instance, food insecurity was categorized using the Department of Agriculture’s 6 Item Food Security Module, which produce a score between 0 and 6 (24, 25). A score of 3 or above indicates the marginal, low, or very low food security and was coded as 1.

This rigorous approach to variable definition and coding was informed by peer-reviewed literature to ensure comparability with existing research and to provide robust, clinically relevant insights into the socioeconomic determinants of mental health.

This study evaluated three mental health symptoms: depression, low self-esteem, and suicidal ideation, represented as binary variables with 1 indicating presence and 0 indicating absence of each symptom. We employed parametric and non-parametric methods to analyze the impact of various social determinants and their correlations with mental health symptoms.

First, we used a probit model as the parametric method, specified by Equation 1:

where Y_ikjt represents the mental health symptom(s) of individual i, working in industry k and residing in county j at round t. The vector X_it includes individual-level characteristics, such as employment status, race, age, sex, civil status, long-COVID status, trust in official and unofficial information, community safety perceptions, and household factors (e.g., household size, income, food security). In addition, we controlled for fixed effects for the industry in which individuals work (η_k), the county of residence (ρ_j), and the time of data collection (ω_t) (i.e., spring or fall 2022); the error term is ε_ikjt.

For robustness, we also applied a non-parametric approach—the conditional inference decision tree. This method visually partitions mental health conditions based on predictors by testing for independence between predictors and symptoms. The method is built by performing a significant test on the independence between the predictors and symptoms (26). In order to use the conditional inference decision tree approach, we first used clustering analysis to group the individuals with similar mental health symptoms into homogeneous groups (27). In particular, individuals were clustered based on their mental health profiles: (i) no mental health symptoms, (ii) depression symptoms only, (iii) depression and low self-esteem symptoms together, and (iv) all three symptoms (depression, low self-esteem, suicidal ideation); 96.4% of individuals in this cohort matched one of these four profiles. We then generated a CTREE tree-shaped probability map (R software). This tree diagram, with branches determined by multiplicity-adjusted p-values based on Bonferroni’s criterion, illustrates the predictors’ non-linear effects on mental health symptoms (28).

Participants in this cohort study were recruited from four counties in Hawaiʻi, with the majority (70.9%) residing in Honolulu County, followed by Hawaiʻi County (15.1%), Maui County (8.9%), and Kauai County (4.8%) representative of the working-age state population 18–65 years old. Table 1 provides a summary of participant characteristics.

Most participants (88.4%) were employed. Demographically, 33.7% were male, with an average age of 45.4 years. Nearly 60% were married or living with a domestic partner, and the median education level was a bachelor’s degree. The cohort self-reported predominantly Asian (44.4%), followed by Caucasian (26.9%) and Native Hawaiian and Pacific Islander (19.5%), with 9.3% identifying as “other.”

In terms of health outcomes, 12.1% reported long-COVID, and 51.5% did not report any chronic health conditions. Alcohol consumption was reported by 25, and 12.5% reported smoking. The average household size was four people, with 27% of participants experiencing food insecurity. Trust in information was higher for official sources (average score of 0.72) than for unofficial sources (average score of 0.58), and 61% felt safe or very safe in their community against COVID-19. Self-reported mental health assessments revealed that 39.6% experienced depressive or highly depressive symptoms, 14.7% reported low self-esteem, and 4.2% reported suicidal ideation.

From clustering of behaviors analysis, we found that three mental health issues were significantly associated with each other, such that having one mental health issue was correlated with increasing the probability of having another mental health issue. In our cohort, 95% of individuals who reported suicidal ideation also experienced low self-esteem and depression Table 2. By using the non-parametric Chi-square test of independence (29), we found that mental health issues were mutually dependent on one another in the baseline sample. People who reported low self-esteem were also more likely to have reported depressive symptoms [χ21=218.76, p = 0.000] and people who reported suicidal ideation were more likely to have reported depressive symptoms [χ21=88.21, p = 0.000]. Individuals who reported having low self-esteem were also more likely to report suicidal ideation [χ21=114.8, p = 0.000].

We further analyzed the association between predictors and mental health conditions in the clustering sample by using a non-parametric approach of conditional inference trees (26). In conditional inference trees, only predictors that are statistically significant with p-values greater than 0.05 are displayed. Therefore, this method is informative and better at determining the true effect of the predictors. In particular, from the clustering sample of mental health conditions, we only focused on four dominant groups: A-None of the mental health symptoms (58.24%), B-One symptom (depression) (25.37%), C-Two symptoms (depression and low self-esteem) (10.26%), and D- Three symptoms (depression, low self-esteem, and suicidal ideation) (2.47%).

The results of conditional inference decision tree analysis are illustrated in Figure 2. Consistent with our probit estimation, participants who reported having food insecurity, pre-existing health conditions, long COVID, feeling unsafe in their community, and being unemployed have a higher probability of reporting mental health issues. Participants who are food insecure are more likely to have mental health issues than food-secure people. Asian people who are food secure, are married or living with their partners, and feel safe in their community are less likely to report mental health symptoms. Eighty-four percent of these people are classified into none of the symptoms group, and 13% of these individuals are in the group of one symptom (node 11). By contrast, non-Asian unemployed people aged 47 years old who are food insecure and have pre-existing health conditions are more likely to have mental health symptoms. Forty-five percent of these individuals are classified in the one-symptom group and 55% of them are in a three-symptom group (node 25). The absence of pre-existing health conditions and feeling safe in the community are considered as protectors of the food-insecure people (node 17).

Our findings reveal that food insecurity is a primary driver of mental health challenges, with strong associations across depression, low self-esteem, and suicidal ideation. Food insecurity was particularly impactful among unemployed individuals, where it exacerbated mental health disparities. Employment offered a strong protective factor only for those experiencing food insecurity, associated with lower depression rates, higher self-esteem, and reduced suicidal ideation. Additionally, food-insecure individuals reported a heightened negative impact of alcohol consumption on depression and self-esteem, further underscoring the vulnerability of this group. These results suggest that addressing food insecurity could yield substantial mental health benefits across multiple population segments in Hawaiʻi.

Our study supports prior research linking employment to improved mental health outcomes (30–36). Employment was associated with lower symptoms of depression, suicidal ideation, and higher self-esteem. Interestingly, employment also modified the effects of other risk factors: employed individuals facing food insecurity or chronic health conditions had fewer mental health symptoms than their unemployed counterparts. The disparities between employed and unemployed adults, especially those aged 30 to 60, are pronounced, suggesting that employment not only offers direct mental health benefits but also buffers against other socioeconomic challenges.

Significant sex differences emerged in mental health outcomes, with women reporting higher rates of depression but lower instances of suicidal ideation compared to men. This aligns with previous findings that stressors such as family responsibilities, multiple roles, and potential discrimination may contribute to depressive symptoms among women (37). Employment and supportive social networks appeared particularly beneficial for women, correlating with improved mental health and higher self-esteem (38, 39).

Mental health disparities across racial and ethnic groups reflected complex sociocultural influences (40). Asian participants reported lower depressive symptoms compared to Caucasians, Native Hawaiians, and Pacific Islanders (41). Food insecurity significantly increased depression and suicidal ideation across all racial groups. Notably, trust in official information sources was particularly protective against depression and low self-esteem among Caucasians, suggesting that interventions promoting trustworthy information may help reduce mental health risks in specific demographics.

Consistent with existing literature, our results show that marriage or long-term partnership is associated with improved mental health outcomes, likely due to increased psychological, social, and economic resources (42, 43). This protective effect was evident across both depression and self-esteem metrics, emphasizing the value of stable relationships for mental health resilience (44).

Our study examined trust in both official and unofficial information sources as potential mental health determinants using validated instruments for Hawaii’s populations (8–10). Both types of trust correlated with lower rates of depression, higher self-esteem, and decreased suicidal ideation. These findings reinforce the importance of trust as a supportive factor for mental health, suggesting that fostering reliable information channels may be beneficial in alleviating mental health symptoms (45–49).

COVID-19-related variables, including long-COVID and perceptions of community safety, were also significant. Individuals with long-COVID reported higher depression and suicidal ideation, consistent with findings on the extended impact of chronic symptoms on mental wellbeing. Conversely, feeling safe in the community was associated with fewer depressive symptoms, higher self-esteem, and lower suicidal ideation. This protective association aligns with research on community safety, indicating that a stable and safe environment mitigates mental health risks, particularly in uncertain times (50, 51).

This study has several limitations. First, self-reported survey data may underrepresent depressive symptoms, suicidal ideation, and behaviors like substance use due to social stigma, introducing potential bias. These self-reports differ from clinically-administered diagnoses, which may warrant further study to confirm consistency. Second, the study lacks detailed data on factors like financial hardship, coping strategies, and emotional intelligence that could influence mental health outcomes. Third, the sample size was limited to adults 18–65 and may not represent the State of Hawaii. Finally, although the conditional inference decision tree model has reasonable precision to detect complex associations between social determinants and mental health, some subtle effects would be neglected because of the stopping rules and the pruning procedures. Despite these limitations, this study offers valuable insights into how social determinants impact mental health in Hawaiʻi’s adult population, which may inform the development of targeted interventions to improve mental health outcomes.