Skip to main content

This is a preprint.

It has not yet been peer reviewed by a journal.

The National Library of Medicine is running a pilot to include preprints that result from research funded by NIH in PMC and PubMed.

Research Square logoLink to Research Square
[Preprint]. 2024 Dec 3:rs.3.rs-5536130. [Version 1] doi: 10.21203/rs.3.rs-5536130/v1

Application of Intersectionality Framework andArea-level Indicators in Machine Learning Analysisof Depression Disparities in All of Us ResearchProgram Data

Dmitry Scherbakov 1, Michael T Marrone 2, Leslie A Lenert 3, Alexander V Alekseyenko 4
PMCID: PMC11643291  PMID: 39678323

Abstract

Background/objective:

Depression is a complex mental health disorder influenced by various social determinants of health (SDOH) at individual and community levels. Area-level factors and intersectionality framework, which considers overlapping personal identities, are used in this paper to get a nuanced picture of depression disparities.

Methods

This cross-sectional study uses electronic health records data from the All of Us research network. Our study cohort includes 20,042 individuals who completed the SDOH surveys in All of Us and had at least one in-patient visit, with 27.3% diagnosed with depression since 2020. We used depression diagnosis as an outcome, while independent variables include US Religious Census and American urvey responses, area-level variables, sociodemographic characteristics: age group, income, gender, sexual orientation, immigration status, marital status, and race/ethnicity – and the interactions of the latter with each other and with other variables. The association between depression diagnosis and the variables is reported by fitting the logistic regression model on the subset of variables identified by LASSO method.

Results

The analysis revealed that area-level indicators, such as religious adherence and childbirth rates, significantly influenced depression outcomes when interacting with personal identity variables: area-level religious adherence was associated with increased depression odds for women (OR 1.33, 95% CI 1.15–1.54) and non-binary individuals (OR 3.70, 95% CI 1.03–13.31). Overlapping identities, such as younger adults unemployed for less than a year and never married Middle Eastern and North African participants showed higher depression odds (OR 2.3, 95% CI 1.06–4.99, and OR 3.35, 95% CI 1.19–9.45, respectively).

Discussion/Conclusion:

The findings underscore the importance of considering all types of factors: individual, area-level, and intersectional in depression research.

Keywords: Depression, Mental health disparities, Intersectionality, Social Determinants of Health, LASSO, All of Us Research Program, Area-Level Indicators

INTRODUCTION

Depression is a multifaceted mental health disorder influenced by various social determinants of health (SDOH) that encompass individual, community, and societal factors. Research has consistently shown that sociodemographic factors, social support, and adverse life experiences significantly contribute to the prevalence and severity of depressive symptoms across diverse populations [1–3].

Intersectionality theory, which was originally developed to acknowledge the combined effects of sex, gender, race, and ethnicity on life outcomes, [4] can be projected into the domain of human health to show that different intersections of identity, social position, processes of oppression or privilege, and policies or institutional practices contribute to health outcomes. The intersectionality framework provides researchers with a perspective to view individuals’ overlapping identities that interact to shape their experiences, in addition to traditional demographic characteristics [5]. In the mental health gender disparities area, the framework may reveal the well-known difference between females and males in depression prevalence rates [6], among other more intricate underlying inequalities created by social structures. For instance, these structures may favor women homemakers and male breadwinners [7]. Intersectionality methods allow us to understand the nature of mental health disparities by analyzing separately working and non-working men and women, moving away from the more general “females” vs. “males” comparisons.

Fehrenbacher & Patel argue that integrating intersectionality into quantitative and mixed-methods research can help address persistent health inequalities [8]. A study by Persmark et al. emphasizes the importance of contextualizing health issues within an intersectional framework in the context of the U.S. opioid crisis, where individual risk factors often overshadow underlying systemic causes of health disparities [9]. This perspective is echoed by Ogungbe et al., who highlight the need for intersectionality to address implicit biases in healthcare that contribute to disparities [10].

The intersectionality framework has been applied to various health contexts, revealing how specific groups experience compounded disadvantages. For example, a study on disability trajectories illustrates how race, ethnicity, and gender intersect to influence health outcomes, suggesting that health disparities are not uniform but rather contingent upon the specific identities individuals hold [5]. This is further supported by research that examines how sexual orientation and race intersect to affect health risk factors, indicating that individuals with multiple marginalized identities face greater health challenges [11], with specific instruments developed to quantify the intersectional effects, for example, related to discrimination [12].

The practical implementation of intersectionality research in healthcare is hindered by specific modeling challenges – namely, the growth of the number of independent variables as a researcher accounts for the interaction effects of a significant number of demographic and social variables. For example, a model with 10 sociodemographic independent variables, each having only 3 levels, results in 200 binary encoded variables when all possible interactions are considered. The growth of the number of variables leads to issues, such as model not being generalizable when number of observations per each variable in minority class is small [13].

Thus, statistical modeling of intersectionality requires special techniques. This paper explores the LASSO variable selection method [14] in order to reduce the dimensions of the design matrix and highlight the most significant intersectional variables that affect depression. LASSO is an efficient method to subset variables and shows superior performance to stepwise variable selection methods [15].

In addition to individual-level SDOH operates at the community level. A secondary goal of this publication is to explore the simultaneous influence of area-level predictors on depression. It is known that lower individual SES is associated with an increased risk of depression due to factors such as financial stress, limited access to healthcare, and reduced educational opportunities [16, 17]. The correlation of household income with depression rates is apparent when two maps are compared side by side (Figs. 1A and 1B), with regions with high depression (e.g., Appalachian) correlating with lower-income regions. At the same time, living in areas where median income is lower than the national average can affect depression through indirect pathways, even if individual or family income is on par with national averages. For example, a middle-class family living in a deprived area might be exposed to higher levels of neighborhood crime and illicit drug exposure, factors associated with lower-income areas [18].

Figure 1.

Figure 1

US County-level data for A: Depression prevalence, %; B: Median household income in USD; C: Percentage of women who gave birth in intact family; D: Percentage of religious adherents.

Area-level factors related to income, family structure, percentage of women who give birth, and religious activity are explored in this paper. Positive religious coping is known to provide emotional support and a sense of meaning, which may buffer against depressive symptoms [19, 20]. At the same time, studies highlight that individuals with strong social networks and supportive relationships tend to experience lower levels of depressive symptoms [21, 22]. Differences in family compositions and religious adherence form distinct geographic spots, shaping areas that can be called “family islands” and “religious islands” (Fig. 1C and Fig. 1D).

To the best of our knowledge, prior studies didn’t investigate how these area-level indicators, together with individual-level predictors, and the interaction between predictors influence depression prevalence rates. The present study combines individual-level sociodemographic indicators from The All of Us Research Program and several area-level predictors derived from All of Us 3-digit ZIP code level data, the American Community Survey (ACS), and the 2020 US Religion Census. We aim to identify which of these variables are statistically significant in predicting depression diagnosis in electronic health records (EHR). Individual-level factors such as age, race/ethnicity, depression diagnosis, and other relevant individual data are obtained from the All of Us platform. The All of Us program’s commitment to diversity and inclusion is essential to capture a broad range of intersectional health outcomes in this study [23].

METHODS

Our general approach follows these steps: 1. Identify individuals in the All of Us dataset diagnosed for depression on or after January 1st, 2020 and those never diagnosed with depression; 2. Link area-level indicators to individual data using a 3-digit ZIP code; 3. Perform LASSO variable selection; 4. Fit a logistic regression model using variables selected by LASSO and plot statistically significant variables.

Identification of people with depression was performed using the All of Us platform phenotyping algorithm, which suggests diagnosis based on the term entered (“depression”). We limited our cohort to patients with at least one in-office visit, who have EHR data present, and who have completed the SDOH survey at any point in time. The SDOH survey contains a range of questions related to social and community context, economic stability, education, neighborhood and built environment, health, and health care [24].

We marked patients as having depression if they had a diagnostic code suggested for the term “depression” (and minimally filtered to exclude unrelated conditions, e.g., myocardial depression, see Table 1) in their EHR, and the diagnosis date was January 1st, 2020, or later. This cut-off date was selected because surveys in All of Us were completed by participants after 2020, and thus, they re ect only recent participant information.

Table 1.

Depression phenotype in All of Us platform.

Diagnosis OMOP ID SNOMED Code
Acute depression 37016718 712823008
Bipolar affective disorder, current episode depression 439254 191627008
Bipolar disorder, most recent episode depression 35624743 767631007
Bipolar I disorder, most recent episode depression 35624748 767636002
Chronic depression 4103574 192080009
Depression screening positive 762504 428181000124104
Depressive disorder 440383 35489007
Depressive episode 3656234 871840004
Major depression in full remission 4269493 63412003
Major depression in partial remission 4148630 30605009
Major depression in remission 4176002 42810003
Major depression single episode, in partial remission 4323418 70747007
Major depression with psychotic features 37111697 726772006
Major depression, single episode 4282096 36923009
Major depressive disorder 4152280 370143000
Mild depression 4149320 310495003
Mild major depression 4336957 87512008
Mild major depression, single episode 4195572 79298009
Mild recurrent major depression 4228802 40379007
Minimal depression 36713698 718636001
Mixed anxiety and depressive disorder 4338031 231504006
Moderate depression 4151170 310496002
Moderate major depression 4307111 832007
Moderate major depression, single episode 4049623 15639000
Moderate recurrent major depression 4077577 18818009
Moderately severe depression 36717092 719593009
Positive screening for depression on PHQ-9 (Patient Health Questionnaire 9) 43021839 112001000119100
Postpartum depression 4239471 58703003
Presenile dementia with depression 377527 191455000
Psychosis and severe depression co-occurrent and due to bipolar affective disorder 35622934 765176007
Reactive depression (situational) 4314692 87414006
Reactive depressive psychosis 435520 191676002
Recurrent depression 4098302 191616006
Recurrent major depression 4282316 66344007
Recurrent major depression in full remission 4263748 46244001
Recurrent major depression in partial remission 4141454 33135002
Recurrent major depression in remission 433991 68019004
Severe depression 4149321 310497006
Severe major depression 42872722 450714000
Severe major depression with psychotic features 4250023 73867007
Severe major depression without psychotic features 4327337 75084000
Severe major depression, single episode 42872411 251000119105
Severe major depression, single episode, with psychotic features 438406 430852001
Severe major depression, single episode, without psychotic features 441534 76441001
Severe recurrent major depression 43531624 281000119103
Severe recurrent major depression with psychotic features 4154309 28475009
Severe recurrent major depression without psychotic features 435220 36474008
Single episode of major depression in full remission 4025677 19527009

In addition, we considered childbirth (since 2020) as the independent variable and a similar phenotyping process was done to capture deliveries and childbirth.

Area-level indicators are described in Table 2. Because All of Us provides individual participant location using a 3-digit ZIP code, we had to aggregate ACS and US Religion Census data into this higher geographical level using the mean values approach. We also binned all area-level indicators into “Low” (less than one SD from the mean), “High” (higher than one SD from the mean), and “Average” (in between Low and High).

Table 2.

Area-level indicators.

Indicator Source Variable name (in the source)
Median income in the area, US$ All of Us 3-digit ZIP code level data median_income
Women who gave birth in the past year, % 5-year (2018–2022) ACS B13004_002
Women who gave birth in the past year, with a spouse present, % 〃 B13004_003
Women who gave birth in the past year, in an unmarried partner household, % 〃 B13004_004
Women who gave birth in the past year, in other living arrangement, % 〃 B13004_005
Share of religious adherents (all denominations/religious groups), % US Religion Census 2020 TOTRATE_2020

We used these area-level predictors, together with individual predictors from the All of Us platform, and intersectional variables created by specifying variable interactions of a select number of individual predictors known to be correlated with depression prevalence (Age group; Per capita income, calculated as Household income divided by number of people in household; Gender identity; Marital status; Sexual orientation; Race; and Birthplace, In- or outside the US) with each other and with other predictors. The individual, area-level, and variables selected for interaction, corresponding number of participants, and univariable odds ratios are described in Table 3.

Table 3.

Sociodemographic profile of study cohort.

Variable No Depression, N (%) Depression, N (%) OR (univariable)
Total N (%) 14564 (72.7) 5478 (27.3) -
Age groupt Adults (40 to 69 years old) 6048 (41.5) 2838 (51.8) -
Older adults (70 + years old) 7560 (51.9) 2096 (38.3) 0.59 (0.55–0.63, p < 0.001)
Younger adults (< 40 years old) 956 (6.6) 544 (9.9) 1.21 (1.08–1.36, p = 0.001)
Home Ownership Own 11316 (77.7) 3440 (62.8) -
Other Arrangement 652 (4.5) 363 (6.6) 1.83 (1.60–2.09, p < 0.001)
Rent 2596 (17.8) 1675 (30.6) 2.12 (1.97–2.28, p < 0.001)
Education Level College Graduate 4345 (29.8) 1426 (26.0) -
Advanced Degree 5819 (40.0) 1570 (28.7) 0.82 (0.76–0.89, p < 0.001)
College One to Three 3096 (21.3) 1727 (31.5) 1.70 (1.56–1.85, p < 0.001)
Unfinished school 148 (1.0) 100 (1.8) 2.06 (1.58–2.67, p < 0.001)
Never Attended 1 (0.0) 2 (0.0) 6.09 (0.58–131.19, p = 0.140)
Twelve Or GED 1155 (7.9) 653 (11.9) 1.72 (1.54–1.93, p < 0.001)
Sexual Orientation† Straight 13621 (93.5) 4833 (88.2) -
Non-heterosexual 943 (6.5) 645 (11.8) 1.93 (1.73–2.14, p < 0.001)
Do you speak a language other than English at home No 12898 (88.6) 4916 (89.7) -
Yes 1666 (11.4) 562 (10.3) 0.89 (0.80–0.98, p = 0.018)
Marital Status† Married 9082 (62.4) 2638 (48.2) -
Divorced 1872 (12.9) 1089 (19.9) 2.00 (1.84–2.18, p < 0.001)
Living With Partner 619 (4.3) 302 (5.5) 1.68 (1.45–1.94, p < 0.001)
Never Married 1768 (12.1) 921 (16.8) 1.79 (1.64–1.96, p < 0.001)
Separated 163 (1.1) 137 (2.5) 2.89 (2.29–3.64, p < 0.001)
Widowed 1060 (7.3) 391 (7.1) 1.27 (1.12–1.44, p < 0.001)
Gender Identityt Man 6096 (41.9) 1557 (28.4) -
Non-binary 46 (0.3) 57 (1.0) 4.85 (3.28–7.21, p < 0.001)
Woman 8422 (57.8) 3864 (70.5) 1.80 (1.68–1.92, p < 0.001)
Birthplace† USA 13217 (90.8) 5126 (93.6) -
Outside USA 1347 (9.2) 352 (6.4) 0.67 (0.60–0.76, p < 0.001)
Within the past 12 months were you worried whether your food would run out before you got money to buy more Never true 13558 (93.1) 4351 (79.4)
Often true 174 (1.2) 261 (4.8) 4.67 (3.85–5.69, p < 0.001)
Sometimes true 832 (5.7) 866 (15.8) 3.24 (2.93–3.59, p < 0.001)
Living Situation - People Under 18 0 12424 (85.3) 4339 (79.2) -
1 or 2 1880 (12.9) 956 (17.5) 1.46 (1.34–1.59, p < 0.001)
3 or more 260 (1.8) 183 (3.3) 2.02 (1.66–2.44, p < 0.001)
Employment Status Employed For Wages 5244 (36.0) 1784 (32.6) -
Homemaker 379 (2.6) 222 (4.1) 1.72 (1.44–2.05, p < 0.001)
Out Of Work Less Than One 214 (1.5) 152 (2.8) 2.09 (1.68–2.59, p < 0.001)
Out Of Work One Or More 324 (2.2) 233 (4.3) 2.11 (1.77–2.52, p < 0.001)
Retired 6474 (44.5) 1861 (34.0) 0.84 (0.78–0.91, p < 0.001)
Self Employed 1100 (7.6) 249 (4.5) 0.67 (0.57–0.77, p < 0.001)
Student 144 (1.0) 85 (1.6) 1.74 (1.32–2.27, p < 0.001)
Unable To Work 685 (4.7) 892 (16.3) 3.83 (3.42–4.29, p < 0.001)
Stable House Concern No 13877 (95.3) 4685 (85.5) -
Yes 687 (4.7) 793 (14.5) 3.42 (3.07–3.81, p < 0.001)
Race† White 12190 (83.7) 4551 (83.1) -
Asian 305 (2.1) 63 (1.2) 0.55 (0.42–0.72, p < 0.001)
Black 1035 (7.1) 458 (8.4) 1.19 (1.06–1.33, p = 0.004)
Hispanic 814 (5.6) 315 (5.8) 1.04 (0.91–1.18, p = 0.601)
MENA 92 (0.6) 33 (0.6) 0.96 (0.64–1.42, p = 0.844)
NHPI 10 (0.1) 4 (0.1) 1.07 (0.29–3.20, p = 0.907)
Other 118 (0.8) 54 (1.0) 1.23 (0.88–1.68, p = 0.218)
Recent childbirth No 14248 (97.8) 5390 (98.4) -
Yes 316 (2.2) 88 (1.6) 0.74 (0.58–0.93, p = 0.012)
Median income in the area Average 11018 (75.7) 4392 (80.2) -
High 2635 (18.1) 585 (10.7) 0.56 (0.51–0.61, p < 0.001)
Low 911 (6.3) 501 (9.1) 1.38 (1.23–1.55, p < 0.001)
Area deprivation index Average 9869 (67.8) 4197 (76.6) -
High 2690 (18.5) 778 (14.2) 0.68 (0.62–0.74, p < 0.001)
Low 2005 (13.8) 503 (9.2) 0.59 (0.53–0.65, p < 0.001)
Estimate Total Women who had a birth in the past 12 months (total) Average 10736 (73.7) 3711 (67.7) -
High 1971 (13.5) 1052 (19.2) 1.54 (1.42–1.68, p < 0.001)
Low 1857 (12.8) 715 (13.1) 1.11 (1.01–1.22, p = 0.024)
Estimate Total Women who had a birth in the past 12 months Has a spouse present Average 7935 (54.5) 3172 (57.9) -
High 2739 (18.8) 1075 (19.6) 0.98 (0.90–1.07, p = 0.660)
Low 3890 (26.7) 1231 (22.5) 0.79 (0.73–0.85, p < 0.001)
Estimate Total Women who had a birth in the past 12 months Partner in an unmarried partner household Average 9963 (68.4) 3576 (65.3) -
High 2506 (17.2) 1136 (20.7) 1.26 (1.17–1.37, p < 0.001)
Low 2095 (14.4) 766 (14.0) 1.02 (0.93–1.12, p = 0.691)
Estimate Total Women who had a birth in the past 12 months Other living arrangement Average 9925 (68.1) 3985 (72.7)
High 2727 (18.7) 902 (16.5) 0.82 (0.76–0.90, p < 0.001)
Low 1912 (13.1) 591 (10.8) 0.77 (0.70–0.85, p < 0.001)
Rate of religious adherence county level Average 9962 (68.4) 3450 (63.0) -
Low 2828 (19.4) 1005 (18.3) 1.03 (0.95–1.11, p = 0.536)
High 1774 (12.2) 1023 (18.7) 1.67 (1.53–1.81, p < 0.001)
How often do you go to religious meetings or services None/Little 7656 (52.6) 2994 (54.7) -
Often 3477 (23.9) 1204 (22.0) 0.89 (0.82–0.96, p = 0.002)
I am not religious 3431 (23.6) 1280 (23.4) 0.95 (0.88–1.03, p = 0.230)
In your day to day life how often are you treated with less respect than other people None/Little 10829 (74.4) 3325 (60.7) -
Often 1005 (6.9) 883 (16.1) 2.86 (2.59–3.16, p < 0.001)
A few times a year 2730 (18.7) 1270 (23.2) 1.52 (1.40–1.64, p < 0.001)
How often are you treated with less respect than other people when you go to a doctor s office or other health care provider None/Little 13358 (91.7) 4493 (82.0) -
Often 1206 (8.3) 985 (18.0) 2.43 (2.22–2.66, p < 0.001)
Per capita household income† High 3260 (22.4) 609 (11.1) -
Low 2852 (19.6) 2044 (37.3) 3.84 (3.46–4.26, p < 0.001)
Medium 8452 (58.0) 2825 (51.6) 1.79 (1.63–1.97, p < 0.001)
Household size Two person household 7868 (54.0) 2336 (42.6) -
Single person household 2885 (19.8) 1379 (25.2) 1.61 (1.49–1.74, p < 0.001)
Three and more person household 3811 (26.2) 1763 (32.2) 1.56 (1.45–1.68, p < 0.001)
†

denotes variables selected for modeling of interactions (for intersectional analysis).

The categorical variables were encoded using one-hot encoding into binary predictors. The LASSO regression method is then used for variable selection. We use the logistic regression model on the set of predictors identified by the LASSO to calculate and then present our main result – the impact of area and individual level predictors on depression – through an odds ratio diagram where we display only significant predictors (p-value < .05).

The supplement provides information about the AIC criterion of the regression model, and demonstrates that the model created with LASSO variables provides the best fit compared to the model with all predictors and the model created by fitting only statistically significant predictors separately (Supplemental Table 3).

RESULTS

The matching All of Us cohort had 20,042 individuals; their demographics, along with individual and area-level variables used in the analysis, are presented in Table 3. In our cohort, 27.3% had a depression diagnosis since 2020.

The initial list contained 1,003 binary independent variables (that includes 27 categorical variables transformed using one-hot encoding and the specified interactions between them). LASSO method identified 181 important variables. The logistic regression modeling results on the variables selected by LASSO method are displayed in Fig. 3, which shows only the predictors with p-value < .05 (and hence confidence intervals not overlapping with 1 on the odds ratio scale).

Figure 3.

Figure 3.

Odds ratio analysis for predictors with p-value < 0.05. Values greater than one indicate a higher chance of occurrence of depression diagnosis. The border around the column indicates the interaction between area and individual variables. 95% CI displayed. “x”-shaped points indicate unadjusted odds ratio (Supplemental Table 1).

The variable interaction with the highest odds was Race - Other : Share of women who had a birth (area-level) - High (OR 6.41, 95% CI 1.56–26.38), followed by Gender Identity - Non-binary : Rate of religious adherence (area-level) - High (OR 3.70, 95% CI 1.03–13.31). Marital Status - Never Married : Race - MENA (OR 3.35, 95% CI 1.19–9.45) and Education Level - Unfinished school : Per capita income - Medium (OR 2.40, 95% CI 1.21–4.78) also increased the odds of the outcome. Other significant positive factors included Employment - Unable to Work (OR 1.99, 95% CI 1.58–2.51), and Younger adult : Share of women who had a birth (area-level) - High (OR 1.98, 95% CI 1.44–2.74). Notably, Younger adult : Marital Status - Divorced (OR 1.96, 95% CI 1.05–3.65) and Older adult : Area deprivation index Low (OR 1.68, 95% CI 1.38–2.04) were also associated with an increase in the outcome.

On the other hand, the variables with the lowest odds included Race - Hispanic : Recent childbirth - Yes (OR 0.22, 95% CI 0.05–0.86), additionally, Do you speak a language other than English at home? - Yes : Race - Other (OR 0.38, 95% CI 0.15–0.98) showed a negative association, as did Home Own - Rent : Race - Asian (OR 0.43, 95% CI 0.22–0.88). Sexual Orientation - Non-heterosexual : Rate of religious adherence (area-level) - Low (OR 0.66, 95% CI 0.48–0.89) and Marital Status - Widowed : Household - Three and more person household (OR 0.63, 95% CI 0.42–0.95) were also associated with lower odds. Other variables with negative associations included Education Level - College One to Three Years : Race - Black (OR 0.70, 95% CI 0.54–0.91), Employment - Self Employed (OR 0.73, 95% CI 0.62–0.85), and Home Own - Other Arrangement : Marital Status - Living With Partner (OR 0.55, 95% CI 0.35–0.89).

Supplementary Table 1 provides uni and multi-variate modeling of all predictors selected by LASSO, not just statistically significant, while Supplementary Table 2 provides a similar analysis for base variables without interactions.

DISCUSSION

Our statistical learning analysis using LASSO and logistic regressions reveals interesting patterns related to intersectionality and interactions with area-level SDOH. Below, we will comment on some of the major findings.

In general, area indicators were comparable in strength and, in some cases, stronger than similar individual-level predictors in our model. For example, individual religious adherence was not a significant predictor in the model, while the interaction of the variable reflecting area religious adherence with gender and sexual orientation was statistically significant. While it is known that women and non-binary individuals face higher rates of depression [25–27], the present study highlights the influence of area factors. Religious organizations can exert pressure on their members, and this can explain the significance of this area predictor.

Similarly, living in an area with a relatively higher number of childbirths was an important predictor of depression for the younger population. In addition, having one, two, or more children in the household was a predictor of depression, but specifically for women, this effect was strong and statistically significant. This finding seems to be confirmed by other studies that suggest that approximately 10% of postpartum women experience major depressive symptoms, with higher rates observed among minority and low-income populations [28, 29].

Belonging to Middle Eastern and North African race and having never married was related in our cohort (which consists mostly of middle-aged and older adults) to higher depression incidence. It is possible that cultural factors related to religious pressures to become married can influence this effect [30].

Unemployment was a strong predictor of depression independent of interaction with demographics, but being out of work for less than one year seemed to be an especially important predictor for the Hispanic subgroup. Similarly, being unemployed was another strong predictor for those born outside of the US. Indeed, research shows that immigrants and historically marginalized groups have limited wealth, and unemployment can lead to quick depletion of resources necessary for sustenance and maintaining the living standard [31, 32]. These specific intersectional effects were much stronger in magnitude than, for example, a simple predictor of having low per capita income.

Some intersectional and interactional effects reduced depression incidence, for example, Asian home renters, individuals of other races who speak a second language at home, and the Hispanic subgroup who had recent childbirth episodes. These effects, while being harder to interpret, were still statistically significant and present an interesting observation.

Another interesting finding is that windowed people living with other people (3 or more people households) seem to have reduced odds of depression. This is echoed by research that shows that older widows living alone are particularly vulnerable to depressive symptoms, suggesting that loneliness plays a critical role in their mental health outcomes [33].

Some individual SDOH variables, although less strong, were still significant in our model, with the most important variables related to per capita income, discrimination, food, and housing insecurity – factors that have long been associated with elevated depression rates [34, 35].

Limitations

Not all possible variable interactions were explored in this analysis. The initial design matrix had only a curated list of demographic and social variables, interactions for which were explored. Adding a larger number of interactions was possible, but it implied significant computation time in the All of Us platform and was not performed. Although the comparison of magnitudes is difficult because some confidence intervals are too wide, we think that our discussion points are still valid because we can observe the direction of effect for each variable and compare it to previously published results.

Conclusions

In this work, we fit the LASSO regression to highlight significant area-level and intersectional predictors for depression. This study highlights the importance of environmental and social factors for historically marginalized populations. This study underscores the importance of modeling the overlapping identities through variable interactions when analyzing mental health and other health outcomes in EHR research networks and databases. Our findings also suggest that specific area parameters, such as rate of childbirth and religious adherence (“family” and “religious islands”), have an unexpected influence on some historically marginalized groups.

Acknowledgments

We gratefully acknowledge All of Us participants for their contributions, without whom this research would not have been possible. We also thank the National Institutes of Health’s All of Us Research Program for making available the cohort data examined in this study.

Funding

Dr. Scherbakov is supported by grant T15 LM013977, Biomedical Informatics and Data Science for Health Equity Research (SC BIDS4Health) Training Program. Dr. Alekseyenko is support by grant T15 LM013977.

The All of Us Research Program is supported by the National Institutes of Health, Office of the Director: Regional Medical Centers: 1 OT2 OD026549; 1 OT2 OD026554; 1 OT2 OD026557; 1 OT2 OD026556; 1 OT2 OD026550; 1 OT2 OD 026552; 1 OT2 OD026553; 1 OT2 OD026548; 1 OT2 OD026551; 1 OT2 OD026555; IAA #: AOD 16037; Federally qualified Health Centers: HHSN 263201600085U; Data and Research Center: 5 U2C OD023196; Biobank: 1 U24 OD023121; The Participant Center: U24 OD023176; Participant Technology Systems Center: 1 U24 OD023163; Communications and Engagement: 3 OT2 OD023205; 3 OT2 OD023206; and Community Partners: 1 OT2 OD025277; 3 OT2 OD025315; 1 OT2 OD025337; 1 OT2 OD025276.

Funding Statement

Dr. Scherbakov is supported by grant T15 LM013977, Biomedical Informatics and Data Science for Health Equity Research (SC BIDS4Health) Training Program. Dr. Alekseyenko is support by grant T15 LM013977.

The All of Us Research Program is supported by the National Institutes of Health, Office of the Director: Regional Medical Centers: 1 OT2 OD026549; 1 OT2 OD026554; 1 OT2 OD026557; 1 OT2 OD026556; 1 OT2 OD026550; 1 OT2 OD 026552; 1 OT2 OD026553; 1 OT2 OD026548; 1 OT2 OD026551; 1 OT2 OD026555; IAA #: AOD 16037; Federally qualified Health Centers: HHSN 263201600085U; Data and Research Center: 5 U2C OD023196; Biobank: 1 U24 OD023121; The Participant Center: U24 OD023176; Participant Technology Systems Center: 1 U24 OD023163; Communications and Engagement: 3 OT2 OD023205; 3 OT2 OD023206; and Community Partners: 1 OT2 OD025277; 3 OT2 OD025315; 1 OT2 OD025337; 1 OT2 OD025276.

Footnotes

Ethics statement and approval

This research aligns with the ethical principles set forth by the All of Us Research Program, including respect for persons, beneficence, and justice. This involves treating participants as autonomous agents, safeguarding their well-being, and ensuring fair distribution of research benefits and burdens.

The Institutional Review Board of the All of Us Research Program approved the collection of survey data and the procedures of sharing the data with verified and qualified researchers. The authors who conducted the analysis have gained at least registered tier rights per All of Us policy.

Conflicts of interest statement

The authors have no competing interests to declare.

Supplementary Files

This is a list of supplementary les associated with this preprint. Click to download.

Contributor Information

Dmitry Scherbakov, Medical University of South Carolina.

Michael T. Marrone, Medical University of South Carolina

Leslie A. Lenert, Medical University of South Carolina

Alexander V. Alekseyenko, Medical University of South Carolina

Data and Code availability

The data underlying this article is available for registered users in All of Us platform. The code required to fetch the data and produce the results in this paper will be made available via the GitHub https://github.com/scherbakovdmitri/AllofUsDepressionDisparitiesLASSO/

References

  • 1.Brown C et al. (2017) Social Determinants of Depression and Suicidal Behaviour in the Caribbean: A Systematic Review. BMC Public Health, 17(1) [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Califf RM et al. (2021) Importance of Social Determinants in Screening for Depression. J Gen Intern Med 37(11):2736–2743 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Liang Y et al. (2012) Social Determinants of Health and Depression: A Preliminary Investigation From Rural China. PLoS ONE 7(1):e30553. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Crenshaw K (2013) Demarginalizing the intersection of race and sex: A black feminist critique of antidiscrimination doctrine, feminist theory and antiracist politics, in Feminist legal theories. Routledge, pp 23–51 [Google Scholar]
  • 5.Warner DF, Brown TH (2011) Understanding How Race/Ethnicity and Gender Define Age-Trajectories of Disability: An Intersectionality Approach, vol 72. Social Science & Medicine, pp 1236–1248. 8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.Salk RH, Hyde JS, Abramson LY (2017) Gender differences in depression in representative national samples: Meta-analyses of diagnoses and symptoms. Psychol Bull 143(8):783. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Sinha S (2017) Multiple roles of working women and psychological well-being. Industrial psychiatry J 26(2):171–177 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Fehrenbacher AE, Patel D (2019) Translating the Theory of Intersectionality Into Quantitative and Mixed Methods for Empirical Gender Transformative Research on Health, vol 22. Culture Health & Sexuality, pp 145–160. sup1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Persmark A et al. (2019) Intersectional Inequalities and the U.S. Opioid Crisis: Challenging Dominant Narratives and Revealing Heterogeneities. Crit Public Health 30(4):398–414 [Google Scholar]
  • 10.Ogungbe O, Mitra AK, Roberts J (2019) A Systematic Review of Implicit Bias in Health Care: A Call for Intersectionality. Imc J Med Sci 13(1):005 [Google Scholar]
  • 11.Hsieh N, Ruther M (2016) Sexual Minority Health and Health Risk Factors. Am J Prev Med 50(6):746–755 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Scheim AI, Bauer GR (2019) The Intersectional Discrimination Index: Development and validation of measures of self-reported enacted and anticipated discrimination for intercategorical analysis, vol 226. Social Science & Medicine, pp 225–235 [DOI] [PubMed] [Google Scholar]
  • 13.Doll JC, Lauer TE (2013) Bayesian estimation of age and length at 50% maturity. Trans Am Fish Soc 142(4):1012–1024 [Google Scholar]
  • 14.Fridgeirsson EA et al. (2024) Comparing penalization methods for linear models on large observational health data. J Am Med Inform Assoc 31(7):1514–1521 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Zhou DJ et al. (2024) Comparison of lasso and stepwise regression in psychological data. Methodology 20(2):121–143 [Google Scholar]
  • 16.Shen B et al. (2013) The Correlations of Disease Activity, Socioeconomic Status, Quality of Life, and Depression/Anxiety in Chinese Patients With Systemic Lupus Erythematosus. Clinical and Developmental Immunology, 2013: pp. 1–6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Zhang G et al. (2022) Depressive Symptoms and Socioeconomic Status Among the Labor Force: Evidence From China’s Representative Sample. PLoS ONE 17(8):e0272199. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Karriker-Jaffe KJ (2011) Areas of disadvantage: A systematic review of effects of area level socioeconomic status on substance use outcomes. Drug Alcohol Rev 30(1):84–95 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Bosworth HB et al. (2003) The Impact of Religious Practice and Religious Coping on Geriatric Depression. Int J Geriatr Psychiatry 18(10):905–914 [DOI] [PubMed] [Google Scholar]
  • 20.Ng GC et al. (2016) Anxiety and Depression in Cancer Patients: The Association With Religiosity and Religious Coping. J Relig Health 56(2):575–590 [DOI] [PubMed] [Google Scholar]
  • 21.Mubangizi I et al. (2020) Prevalence and Factors Associated With Depression Among Adults With Sickle Cell Disease at Mulago Hospital, Uganda: A Cross Sectional Study. [Google Scholar]
  • 22.Oddone CG et al. (2011) Social Support Modifies the Relationship Between Personality and Depressive Symptoms in Older Adults. Am J Geriatric Psychiatry 19(2):123–131 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Mapes B et al. (2020) Diversity and Inclusion for the All of Us Research Program: A Scoping Review. PLoS ONE 15(7):e0234962. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Tesfaye S et al. (2024) Measuring social determinants of health in the All of Us Research Program. Sci Rep 14(1):8815. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Kneale D, Bécares L (2021) Discrimination as a Predictor of Poor Mental Health Among LGBTQ + People During the COVID-19 Pandemic: Cross-Sectional Analysis of the Online Queerantine Study. BMJ Open 11(6):e049405. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Ormiston CK, Williams F (2022) LGBTQ Youth Mental Health During COVID-19: Unmet Needs in Public Health and Policy. Lancet 399(10324):501–503 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Price-Feeney M, Green AE, Dorison SH (2020) Understanding the Mental Health of Transgender and Nonbinary Youth. J Adolesc Health 66(6):684–690 [DOI] [PubMed] [Google Scholar]
  • 28.Ertel KA et al. (2012) Racial Discrimination, Response to Unfair Treatment, and Depressive Symptoms Among Pregnant Black and African American Women in the United States. Ann Epidemiol 22(12):840–846 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Nowak A et al. (2020) How Depressive Symptoms Among African American Women Relate to Measures of Social Disorder in Her Childhood and Pregnancy Neighborhood. J Urb Health 97(1):26–36 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Abuelezam NN et al. (2021) Understanding Differences Within Ethnic Group Designation: Comparing Risk Factors and Health Indicators Between Iranian and Arab Americans in Northern California. BMC Public Health, 21(1) [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.González HM et al. (2010) The Epidemiology of Major Depression and Ethnicity in the United States. J Psychiatr Res 44(15):1043–1051 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Singh V, Haynes PL, Quan SF (2022) Assessing Depression and Suicidality Among Recently Unemployed Persons With Obstructive Sleep Apnea and Socioeconomic Inequality. Southwest J Pulmonary Crit Care Sleep 24(5):81–88 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Jeon G-S, Choi K, Sung-Il C (2017) Impact of Living Alone on Depressive Symptoms in Older Korean Widows. Int J Environ Res Public Health 14(10):1191. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Gou Y et al. (2022) Province- And Individual-Level In uential Factors of Depression: Multilevel Cross-Provinces Comparison in China. Front Public Health, 10 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Lorant V et al. (2003) Socioeconomic Inequalities in Depression: A Meta-Analysis. Am J Epidemiol 157(2):98–112 [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Data Availability Statement

The data underlying this article is available for registered users in All of Us platform. The code required to fetch the data and produce the results in this paper will be made available via the GitHub https://github.com/scherbakovdmitri/AllofUsDepressionDisparitiesLASSO/


Articles from Research Square are provided here courtesy of American Journal Experts

RESOURCES