Quantifying the Independent Impact of Social Determinants of Health and Posttraumatic Stress Disorder on Resource Utilization and Health Outcomes in Civilians

Abstract

Social determinants of health (SDoH) and posttraumatic stress disorder (PTSD) are well-documented risk factors for poor health outcomes, yet their independent contributions to healthcare resource utilization (HCRU) and patient-reported quality of life remain unclear. This retrospective study, using the All of US database, compared civilians with and without PTSD (n=747, matched on age, sex at birth, race, and ethnicity) and examined whether differences in SDoH, beyond the matched demographics, were associated with poorer health outcomes. Multiple linear regressions were conducted for twelve health outcomes related to HCRU and patient-reported well-being to separate the independent contributory effect of PTSD and SDoH. Across both HCRU and patient-reported well-being, PTSD patients had worse outcomes. SDoH factors accounted for 23.2% of Outpatient Visits and 50.1% of Emergency Room Visits among civilians, translating to a $8.82 billion health system burden. The SDoH domains of Employment Status, Stress, and Discrimination were significantly associated with increased HCRU. Further, SDoH factors contributed to more than half (52.4–84.5%) of poor self-rated well-being outcomes compared with PTSD itself. Self-reported outcomes were affected by a variety of SDoH domains, most prominently Employment Status, Education, and Relationships. These results offer insight into the potential significant independent contribution of unfavorable SDoH in worsening health outcomes among civilians with PTSD.

1. Introduction

The prevalence of PTSD in the civilian population is estimated at approximately 6% in the United States (US), translating to a 2019 population prevalence of 10.19 million (Davis et al., 2022; Schein et al., 2021). In the US, only about half of those with PTSD seek treatment, and even fewer are treated by a mental health professional, possibly due to broad access issues such as the high out-of-pocket cost of psychotherapy and limited availability of mental health services (Davis et al., 2022; Koenen et al., 2017).

The burden of PTSD has been estimated at $189.5 billion in the civilian population (driven by direct healthcare expenses and unemployment), costing more per individual than other conditions such as coronary heart disease, anxiety, and depression (Davis et al., 2022). These estimates encompass substantial non-healthcare and indirect costs, including unemployment, lost productivity, and other SDoH factors, representing a total burden of $106.5 billion for civilians (Davis et al., 2022). More effort is needed to measure these intangible costs and understand the broad impact of PTSD on society.

Social determinants of health (SDoH) as defined by the US Department of Health and Human Services, are “conditions in the environments in which people are born, live, learn, work, play, worship, and age that affect a wide range of health, functioning, and quality-of-life outcomes and risks” (US Department of Health and Human Services, 2021). With sufficient chronicity and severity, SDoH stressors like food and housing insecurity, unemployment, and discrimination may cumulatively exacerbate PTSD symptoms as much as the presence or absence of the disease itself (Alegría et al., 2018).

Prior research has indicated a connection between SDoH and the manifestation of PTSD symptoms in both veteran and civilian populations (Holder et al., 2022). Though the relationship between PTSD and SDoH is complex, it is not unidirectional, and SDoH can independently evolve and impact health outcomes in parallel with PTSD itself. The intersectional nature of SDoH has made it difficult to isolate their influence, but refinement of real-world evidence capabilities has made it possible to study such factors on a more granular scale.

The current study analyzes PTSD and SDoH to find their relative impacts on healthcare resource utilization (HCRU) and patient-reported well-being outcomes in the US civilian PTSD population. The study sought to distinguish the independent effect of PTSD on health outcomes and costs from that of myriad SDoH affecting civilian populations. By incorporating PTSD diagnosis and a wide range of SDoH as independent variables, this study aimed to disentangle the differences in SDoH between PTSD patients and their controls, enabling a more thorough understanding of their respective contributions.

2. Materials and Methods

2.1. Data Source and Participants

An important resource for studying SDoH and the primary data source of this study is the National Institute of Health’s All of Us Research Program. The All of Us database holistically captures a patient’s comprehensive and overall wellness by longitudinally tracking electronic health records (EHR), genetic data, lifestyle factors, environmental exposures, and a wide variety of SDoH. All of Us has enrolled a large number of participants diagnosed with PTSD, as reflected in surveys and electronic medical records. As of June 6th, 2022, the All of Us database has collected responses from 372,380 patients, with 10,860 enrollees reporting having PTSD.

2.2. Procedure

All analysis was conducted on a cloud-based workbench in the All of Us portal. The primary objective of this analysis was to compare civilians with PTSD to civilians without PTSD. These two civilian populations were ingested into the workbench for further cleaning, preprocessing, and analysis.

Preprocessing each subpopulation by observation involved applying a series of detailed inclusion and exclusion criteria (see Supplementary Tables S1 and S2) to each population. These criteria primarily consisted of diagnosis codes for PTSD that needed to be included in the EHR to confirm the medical diagnosis of PTSD. In addition, only All of Us participants who had fully completed the SDoH survey and had a linked EHR were included in the final populations for analysis, which resulted in a substantial reduction in population numbers, since many who have enrolled in the rapidly growing All of Us dataset have not yet fully completed the survey materials and data linkage processes. Finally, the date of diagnosis of PTSD had to be available, approximated by the first occurrence of a PTSD ICD code (ICD10CM: F43.10, ICD9CM: 309.81) before the survey completion date. The attrition process that resulted in our final populations of interest is displayed in Figure 1 below.

Figure 1.

Summary of Attrition after Inclusions, Exclusions, Data Integrity Checking, and Matching

2.3. Outcome Measures

Select measures of HCRU sourced from EHR data included Outpatient Visits, Inpatient Visits, Emergency Room Visits, Ambulatory Visits, Psychotherapy Visits, and Other Visits. A comprehensive breakdown of these HCRU measures is shown in Supplementary Table S3. All of Us obtains most of its HCRU data from one of 60 healthcare organizations within its network, while also allowing participants to provide additional data through their online portal (All of Us Research Program, 2022). In the US, 17.2% of Outpatient Visits are classified as Psychotherapy Visits for civilians (Ahn-Horst & Bourgeois, 2024). While Psychotherapy Visits were also analyzed independently, they are a subset of outpatient visits and, therefore, excluded from economic burden calculations to avoid double counting. A complete list of CPT codes used for the Psychotherapy Visits outcome measure is available in Supplementary Table S4. To avoid confounding effects from the 2020 COVID-19 pandemic, visits were sourced from 2019. As part of the analysis of HCRU, efforts were also undertaken to map visit count to economic burden. The count of Diagnosed Conditions for respondents was also calculated based on comorbidities that commonly occur in patients with PTSD; see Table 1 for the list of comorbidities included.

Table 1.

Well-being Outcome Measures

Variable Name	Item Definition
Rate Health (Self-Reported)	“In general, would you say your health is”:
Rate Quality of Life (Self-Reported)	“In general, would you say your quality of life is”:
Rate Physical (Self-Reported)	“In general, how would you rate your physical health?”:
Rate Mental (Self-Reported)	“In general, how would you rate your mental health, including your mood and your ability to think?”
Rate Social (Self-Reported)	“In general, how would you rate your satisfaction with your social activities and relationships?”
Average Pain (Self-Reported)	“In the past seven days, how would you rate your pain on average?”
Average Fatigue (Self-Reported)	“In the past seven days, how would you rate your fatigue?”
Emotional Problems (Self-Reported)	“In the past seven days, how often have you been bothered by emotional problems such as feeling anxious, depressed, or irritable?”
Diagnosed Conditions (EHR)	Additional diagnoses typically occurring in PTSD patients include:   • Irritable bowel syndrome (Oroian et al., 2021; Pacella et al., 2013)   • Hypertensive Disorder (Krantz et al., 2022)   • Crohn’s Disease (Oroian et al., 2021)   • Inflammatory Bowel Disease (Krantz et al., 2022)   • Rheumatoid Arthritis (Oroian et al., 2021)   • Parkinson’s Disease (Barer et al., 2022)   • Obesity (Oroian et al., 2021)   • Type 2 diabetes mellitus (Krantz et al., 2022)   • Hashimoto thyroiditis (Oroian et al., 2021)   • Alzheimer’s disease (Flatt et al., 2018)   • Bipolar disorder (Davis et al., 2022)   • Asthma (Allgire et al., 2021)

Note. Variables were coded as follows: Rate Health, Rate QoL, Rate Physical, Rate Mental, Rate Social: 1 (Poor) to 5 (Excellent); Average Pain: 0 to 10 for increasing level of pain; Average Fatigue: 1 (Very Severe) to 5 (None); Emotional Problems: 1 (Always) to 5 (Never); Diagnosed Conditions 0 to 12 for the existence of additional diagnoses.

This study also assessed patient well-being domain outcomes. These domains captured the mental, physical, social, and general quality of respondents’ lives. Patient-reported Pain, Fatigue, and Emotional Problems were also analyzed. These categorical well-being outcomes were recorded on a Likert scale. The full list of patient-reported well-being outcome measures is displayed in Table 1 below.

Ascertaining outcome differences between populations with and without PTSD was the first part of the hypothesis; the second was to gauge the relative influence of SDoH on these outcomes. These SDoH variables were selected a priori based on existing literature, including published studies and discussions with the study team. (Schein et al 2021; Davis et al 2022; Koenen et al 2017). Specific questions from the data source were removed to limit multicollinearity in the analysis. An overview of the 21 SDoH predictor domains and an exhaustive list of the 98 SDoH domain questions included in the analysis are listed in Supplementary Tables S5 and S6.

3. Data Analysis

3.1. Data Ingestion

All initially preprocessed populations were combined into a single “Master Group” workbench for additional cleaning and preprocessing of individual variables.

3.2. Data Cleaning

Each data field for each population was further processed for use in analysis. Binary (yes or no) questions were encoded as “1” for yes and “0” for no. For any ordinal question where more than two responses were possible, responses were re-coded to correlate with numerical values from 1 to n, with n being the number of possible responses. 1 (the lowest end of the numerical scale) was used to represent the most unfavorable outcome when considering health. Question-wording required that the directionality of some responses be flipped so that 1 consistently represented the worst health outcome. Dummy variables were created for nonordinal categorical data questions, in which a “1” is used to indicate the endorsed answer. All answer options that were not endorsed by the participant were marked as a “0.”

3.3. Multiple Imputation

Multiple imputation by chained equations is a common statistical technique used to address missingness in data to reduce bias and generate a truer dataset (Rubin, 2009). Some question responses suffered from incomplete, missing, or nonanswer options such as “Skip” or “Don’t Know”. To limit the attrition of participants and preserve data, five imputed datasets were created using the MICE package in R to address these nonresponses (Buuren & Groothuis-Oudshoorn, 2011). The results of the imputed datasets were combined using Rubin’s rule and outcomes were analyzed based on the pooled data results (Rubin, 2009). Supplementary Table S7 provides the percentage of questions that required data imputations for each outcome variable and SDoH domain. In total 5.0% of the PTSD population and 4.5% of the non-PTSD population had responses that required multiple imputations.

3.4. Propensity-Matching

Differences in SDoH characteristics are sensitive to age, sex at birth, race, and ethnicity. Many other studies have investigated the effect of these basic demographic features on various outcomes, some of which we refer to in the background. Since the purpose of this study was to investigate the specific SDoH factors associated with a given outcome, we wished to eliminate the confounding that would occur should the comparison groups differ by such immutable demographic features. Propensity score matching was used to create a matched control group for comparison with each PTSD cohort. For each of the PTSD subpopulations created during data ingestion, 1:1 matching using the R nearest neighbor MatchIt propensity matching function was performed on the control group (Greifer, 2023). This resulted in two equally sized comparison groups with nearly equal age, sex at birth, race, and ethnicity composition. The resulting pre-matched and post-matched population distributions are shown in Table 2.

Table 2.

Demographics Before and After Propensity Score Matching

Demographic	PTSD	Non-PTSD Original	Non-PTSD Matched
Age [Mean]	55.08	61.43	54.62
Sex at Birth [%]
Female	83.27	78.60	84.47
Male	>14.05	>18.72	>12.85
Other*	<2.68	<2.68	<2.68
Race [%]
White	73.76	84.45	75.37
Black/African American	10.31	5.31	10.98
Other**	15.93	10.24	13.65
Ethnicity [%]
Hispanic	10.58	6.21	10.04
Not Hispanic	85.54	>91.11	86.75
Other***	3.88	<2.68	3.21

Note. The Matchit function in R was used to propensity match using a nearest neighbor approach.

^*Includes: ‘Intersex,’ ‘None of These, ‘Prefer not to Answer’, and ‘Other’

^**Includes: ‘Asian’, ‘Middle Eastern/North African’, ‘Native Hawaiian and Pacific Islander’, ‘More than one population’, and ‘Other’

^***Includes: ‘Skip’, ‘None of These’, ‘Prefer Not to Answer’, and ‘No Matching Concept’

3.5. Significance Testing

For each outcome of interest variable, a paired t-test was conducted to test whether the means of the outcome variables of interest were nominally the same for the PTSD and control groups. Ordinal variables were compared via mean rating between the PTSD and comparative cohort. To isolate the SDoH and PTSD impact on outcomes measures, grouping PTSD and non-PTSD populations was required to add PTSD status as a discrete feature.

3.6. Regression Analysis

For each outcome of interest variable that showed a statistically significant difference (p < 0.05) between the PTSD population and the control in the paired t-test analysis, a multiple linear regression analysis was conducted. The linear regression models formulated equations with coefficients attached to each variable. These coefficients can be interpreted to understand the dependent variables’ predicted impact on each outcome. To ensure that PTSD effects were not biased by SDoH effects, both were included as independent covariates in the models, allowing for an estimation of the independent effects of PTSD on health outcomes while controlling for SDoH effects. The coefficient of the PTSD variable was weighted against the aggregated coefficients from the SDoH variables, providing insight into the relative influence of PTSD and SDoH on health outcomes. This comparison allows for a more nuanced understanding of the complex relationship between PTSD and SDoH, which is essential for informing policies and interventions aimed at improving health outcomes.

3.7. Mapping to Economic Burden

Using the prevalence of 10.19 million for the US civilian PTSD population in 2019, an exploratory analysis of the economic burden of PTSD and SDoH was independently calculated using a mixed methods approach from established visit cost data from the literature. Outpatient cost data was derived from literature using a top-down approach from cost estimates derived as spending per capita on each service divided by utilization per capita, resulting in an economic burden of $499 per Outpatient Visit (Moses et al., 2018). The calculated economic burden for an Emergency Room Visit is $1,082, defined as direct medical costs (deGraft-Johnson, 2023). Psychotherapy Visits were calculated using the Physician Fee Schedule using procedural costs, which were calculated at a cost of $137 per visit (License for the use of current procedural terminology, Fourth edition (“Cpt^®,” n.d.). All costs were pulled from 2019 data or adjusted for inflation to represent 2019 costs.

4. Results

4.1. Demographics

Table 2 displays the study population characteristics for each of the civilian PTSD cohorts. The civilian PTSD study population was 83.3% women and 73.8% white. The mean age of the civilian PTSD cohort was 55 years.

5.1. Significance Testing

As shown in Table 3, HCRU was significantly higher for the PTSD group for Outpatient Visits, Emergency Room Visits, and Psychotherapy Visits. Every self-reported well-being outcome was significantly worse among civilians with PTSD. Furthermore, civilians with PTSD were diagnosed with a greater number of comorbidities than those without PTSD. All outcomes that met the threshold for statistical significance were further analyzed in a regression analysis to quantify the independent effect of PTSD and SDoH on the individual outcomes. Of the 15 outcomes of interest, 12 showed statistically significant differences in mean between the PTSD population and the control. Inpatient Visits, Ambulatory Visits, and Other Visits failed to reach the threshold for statistical significance, so regression analysis was not performed on these outcomes.

Table 3.

Adjusted Mean T-Test Results Between Propensity-Matched Civilian PTSD and Non-PTSD

Population
	PTSD	Non-PTSD	Absolute Mean Difference	p-value
Outpatient Visits*	16.43	10.30	6.13	< .001
Inpatient Visits*	0.86	0.49	0.37	0.09
Emergency Room Visits*	0.57	0.29	0.28	< .001
Ambulatory Visits*	0.46	0.29	0.17	0.15
Other Visits*	0.91	0.80	0.11	0.41
Diagnosed Conditions	0.90	0.59	0.31	< .001
Psychotherapy Visits*	2.18	0.57	1.61	< .001
Rate Health	2.66	3.12	0.46	< .001
Rate Quality of Life	2.97	3.43	0.46	< .001
Rate Physical	2.62	3.01	0.39	< .001
Rate Mental	2.71	3.13	0.42	< .001
Rate Social Satisfaction	2.78	3.19	0.41	< .001
Average Pain	5.48	4.19	1.29	< .001
Average Fatigue	2.89	2.55	0.34	< .001
Emotional Problems	3.30	2.87	0.43	< .001

^*Visits with N/A replaced with a 0 to account for non-utilizers in the mean calculation

Note. Variables were coded as follows: 1 (Present); Outpatient Visits, Inpatient Visits, Emergency Room visits, Ambulatory Visits, Other Visits: Mean number of visits in 2019; Diagnosed Conditions ranging from 0 to 12 for the existence of additional diagnoses; Rate Health, Rate QoL, Rate Physical, Rate Mental, Rate Social: 1 (Poor) to 5 (Excellent); Average Pain: 0 to 10 for increasing level of pain; Average Fatigue: 1 (None) to 5 (Very Severe); Emotional Problems: 1 (Never) to 5 (Always).

5.2. SDoH and PTSD Independent Impact on Outcomes

PTSD and SDoH separately and independently contributed to incrementally poorer outcomes in PTSD groups. Table 4 below displays the independent impact of PTSD and SDoH on the HCRU of the civilian population. Within the civilian population, PTSD represented a larger impact on the additional healthcare visits than SDoH, with a contribution ranging from 49.9–101.0%, depending on the visit type. While not shown in the table below, SDoH contributed to 37.5% of the 0.31 additional diagnosed conditions in civilians with PTSD.

Table 4.

Impact of PTSD and SDoH on HCRU (2019) in the Civilian Population

	Base (Mean Visits in Control Group) (A)	PTSD Additive Effect (B)	Incremental SDoH Effect (C)	Mean Visits in PTSD Population (A+B+C)
Outpatient
Mean Visits	10.30	4.70	1.43	16.43
Additional Visit Contribution [%]		76.8	23.2
Cost per Patient ($)		2,345	714
Cost to System (in billions $)		23.90	7.27
Emergency Room
Mean Visits	0.29	0.14	0.14	0.57
Additional Visit Contribution [%]		49.9	50.1
Cost per Patient ($)		151	152
Cost to System (in billions $)		1.54	1.55
Psychotherapy
Mean Visits	0.57	1.62	−0.02	2.18
Additional Visit Contribution [%]		101.0	−1.0
Cost per Patient ($)		222	−3
Cost to System (in billions $)		2.26	−0.03

Table 5 shows the respective impact of PTSD and SDoH on self-reported well-being outcomes in civilians with and without PTSD. Compared to their counterparts without PTSD, civilians with PTSD rated themselves as having lower Health, Quality of Life (QoL), Physical Health, Mental Health, and Social Satisfaction, more Emotional Problems, as well as worse Average Fatigue and Average Pain. Unlike in HCRU, across well-being categories, SDoH contributed more to differences between civilians with or without PTSD than the disease itself (52.4–84.5% SDoH impact on population difference vs. 15.5–47.6% PTSD impact on population difference).

Table 5.

Effects of PTSD and SDoH on Patient-Reported Well-being in the Civilian Population

	Base (Mean Rating in Control Group) (A)	PTSD Additive Effect (B)	Incremental SDoH Effect* (C)	Mean Rating in PTSD Population (A+B+C)
Rate Health: Likert Scale of 1 (Poor) to 5 (Excellent)
Mean	3.12	−0.14	−0.31	2.66
% Contributed to Change		−31.2	−68.8
Rate QoL: Likert Scale of 1 (Poor) to 5 (Excellent)
Mean	3.43	−0.12	−0.34	2.97
% Contributed to Change		−36.4	−73.6
Rate Physical Health: Likert Scale of 1 (Poor) to 5 (Excellent)
Mean	3.01	−0.06	−0.32	2.62
% Contributed to Change		−15.5	−84.5
Rate Mental Health: Likert Scale of 1 (Poor) to 5 (Excellent)
Mean	3.13	−0.20	−0.22	2.71
% Contributed to Change		−47.6	−52.4
Rate Social Satisfaction: Likert Scale of 1 (Poor) to 5 (Excellent)
Mean	3.19	−0.08	−0.33	2.78
% Contributed to Change		−20.1	−79.9
Average Pain: Likert Scale of 0 to 10 for increasing level of pain
Mean	4.19	0.33	0.89	5.48
% Contributed to Change		26.7	73.3
Average Fatigue: Likert Scale of 1 (Very Severe) to 5 (None)
Mean	2.55	0.07	0.27	2.89
% Contributed to Change		21.3	78.7
Emotional Problems: Likert Scale of 1 (Always) to 5 (Never)
Mean	2.87	0.19	0.24	3.30
% Contributed to Change		−43.8	−56.2

Note: This table shows the sum of SDoH question coefficients from the regressions.

^*Incremental SDoH is based on the differences in SDoH between PTSD and non-PTSD patients. The base visits, by design, is effect by SDoH.

5.3. Individual SDoH Domain Impact on Outcomes

Among the various aggregated domains, some domains had clusters of questions that had a noticeably higher impact on the models than others. The SDoH domains Employment Status and Unable to Afford Care had the largest contributory effect on Outpatient Visits. Being discriminated against emerged as the most statistically significant factor driving an increase in Emergency Room Visits within the cohort. In contrast, the effect of SDoH on Psychotherapy Visits appeared minimal, as PTSD was the dominant factor influencing psychotherapy utilization, drowning out any SDoH-related contributions.

While SDoH domain significance varied by patient-reported outcomes, the most commonly significant domains were Employment Status, Stress, and Relationships. Income and Education Level also showed statistical significance, especially in Rate Quality of Life. While Diagnosed Conditions was heavily dependent on PTSD status, Discrimination and Treatment from Others were statistically significant predictors.

The civilian HCRU observed outcomes illustrate how costly PTSD is in the civilian population while demonstrating that SDoH has a separate and meaningful role. PTSD accounted for 49.9% to 101.0% of additional visits among civilians, and SDoH contributed −1.0% to 50.1%. Using a 2019 estimate of 10.19 million civilians living with PTSD, these visits amounted to $25.44 billion in direct additional costs due to PTSD, plus $8.82 billion in direct additional costs due to SDoH.

6. Discussion

This study is one of the first to link both SDoH and PTSD to economic costs and is unique in its inclusion of a large civilian sample. The research presented here quantified the independent impact of PTSD and SDoH on Outpatient Visits, Emergency Room Visits, and Psychotherapy Visits, as well as the per-patient and system costs of these additional visits. Furthermore, the effects of SDoH and PTSD on patient-reported measures of well-being were assessed for the following scales: Rate Health, Rate QoL, Rate Physical Health, Rate Mental Health, Rate Social Satisfaction, Rate Average Fatigue, Rate Emotional Problems, and Rate Average Pain. For both HCRU and self-reported well-being, the relative influence of various SDoH domains was measured. This analysis demonstrated that PTSD and SDoH independently contribute to poorer health and economic outcomes for civilians living with PTSD.

The results from this study reaffirmed previous literature on PTSD and SDoH, which has shown Employment, Stress, Support, and Relationships to be critical components of well-being (Holder et al., 2022). By evaluating the differential impact of SDoH and PTSD on health outcomes and costs, this study provides new insights into the independent contributions of various SDoH domains and PTSD. Our study confirmed the outsized relationship between employment on health and economic outcomes, which previous research has found to be predictive of PTSD remission post-treatment (Sibrava et al., 2019).

Importantly, our findings indicate that civilian PTSD is a substantial burden to the healthcare system. Although PTSD itself accounted for the majority of HCRU in civilians with PTSD, imbalances in SDoH factors between PTSD and non-PTSD participants also had a significant independent contribution to poorer outcomes beyond the disease itself. While the relative effect of SDoH varied across the well-being outcomes, perhaps our most important finding is that SDoH was even more influential than PTSD in shaping patient-reported outcomes. This suggests it will be critical to consider social and environmental factors in diagnosing, treating, and supporting civilians with PTSD.

Among the significant SDoH variables identified, Employment Status and Unable to Afford Care contributed most to civilians’ additional visits, whereas Employment Status, Support, Stress, and Relationships drove lower health ratings overall. Determining why these factors outweigh other SDoH domains is beyond the scope of this study. Employment Status is a logical determinant of HCRU since individuals who are unemployed or unable to work due to disability tend to have higher utilization rates. (Li et al., 2023; National Academies Press, 2018). It is also intuitive that employment – which provides a sense of purpose, financial stability, and a community outside of the home – would be central to well-being, particularly in the US, where individualism and self-sufficiency are highly valued. The role of Support may be explained by the lack of a formal treatment infrastructure for civilians, unlike veterans, who benefit from extensive PTSD-specific resources, such as group therapy and clinical care, available through the VA (US Department of Veterans Affairs, 2022c). Meanwhile, the outsized effect of Stress and Relationships may be due to their omnipresence; with no socioeconomic or demographic boundaries, essentially every person experiences some degree of stress or interpersonal highs and lows, with direct consequences for their self-perceived well-being. Simply put, stress and relationships are fundamental elements of the human experience. Other SDoH domains that showed statistical significance across multiple self-reported outcomes are Income, Education, Insurance, and Neighborhood Likeness. Among the Diagnosed Conditions outcome variable, Discrimination was the most significant SDoH predictor. This conclusion is supported by existing literature. In a longitudinal clinical study, Sibrava et al. evaluated associations between discrimination and PTSD among Latinx and African American adults with anxiety disorders (Sibrava et al., 2019). The reported frequency of discrimination accounted for 28.4% to 37.9% of the variance for PTSD diagnostic status but did not predict any other mood disorder. The frequency of discrimination experiences also correlated with increased comorbid disorders for both groups, while employment status significantly predicted remission rates after treatment in African Americans. Holder et al. observed similar correlations between SDoH and PTSD symptoms across veteran and civilian groups (Holder et al., 2022). Discrimination in research and medical practice creates gaps in care and understanding for minorities and underrepresented groups, which is a major area of focus that All of Us is combatting (Williams et al., 2019).

The findings from this study should be considered alongside some key limitations, which include common limitations to retrospectively collected data including representativeness of the sample. Given the limitation of the dataset in having diagnosis data available in ICD format and primarily prior to 2018, ICD-11 codes were unavailable as a search tool for capturing patients. Further, PTSD diagnosis by network analysis, which has become much more common in the past few years, could also not be used for this analysis as it would not have been available for most people at the time of diagnosis (Bovin et al., 2021). Future updates could address these issues.

The All of Us dataset is also limited by variable PTSD representation between states, cross-sectional surveys, and some attrition. Though all fifty states are represented, the number of patients with PTSD varies from state to state, requiring weighting by base respondent population size for a nationally representative cohort. Because the All of Us data collection methodology requires in-person testing at healthcare facilities, participants from rural areas may be unable to travel to the site, increasing the likelihood of missing information for these patients. A potential consequence is the over-representation of urban participants or those in closer proximity to healthcare resources. This bias complicates the proper scaling of the All of Us data to fully capture the US population’s diverse demographics. The online administration of the All of Us survey could also pose a barrier to participants without computer access, potentially skewing the socioeconomic distribution of participants in the sample.

It should also be noted that there was a considerable drop-off between All of Us enrollment and completion of the SDoH survey following PTSD diagnosis (90% attrition). While All of Us is a great resource for research in underrepresented populations, minority groups in the data have a much lower survey completion percentage than white participants, specifically for the Social Determinants of Health survey. Among participants that completed the Basics survey, 19.7% are black or African American, far higher than the 5.6% that completed the Social Determinants of Health survey. As completion of the SDoH survey was a critical inclusion criterion for the model, this drop-off may have led to results that skew to be more generalizable to the white population, rather than the country at large. Nonetheless, the richness of the SDoH survey offered detailed SDoH information not previously seen in other databases.

The high proportion of females in the AoU PTSD cohort is also noteworthy, albeit maybe unsurprising, given the AoU participation bias and prevalence data from the literature (Kambara et al., 2024). In the dataset, 72% of the civilian general population who completed the SDoH survey answered as female vs male. In addition, from literature female civilians are over twice as likely to be diagnosed with PTSD than their male counterparts (Lehavot et al., 2018). Future studies could involve stratification on sex at birth, race, and other demographic groups to assess whether specific subpopulations further exacerbate outcomes, but that was beyond the scope of this analysis. (Lehavot et al., 2018).

Beyond the limitation of PTSD diagnosis via ICD9 and ICD10, another limitation of the study is omitted variable bias, as there are likely factors that were not considered or not available in the data to include in the regression analysis. Any additional factors that were not considered would likely alter the percentage of burden that is attributed to SDoH and PTSD. While steps to address multicollinearity were undertaken, the sheer amount of SDoH variables included in the analysis likely contributed to multicollinearity on a smaller level, affecting the strength of some SDoH domains. Additionally, EHR as a source of data for assessing the number of visits does not capture the utilization of providers that are not integrated into EHR systems. When calculating the economic burden of PTSD among our populations of interest, a mixed methods approach was used, so additional adjustments should be explored for precise calculations of economic burden. If claims data were available for the subjects, more accurate estimates would have been possible. The economic burden calculations are likely underestimated as they do not include any indirect costs such as lost wages from unemployment, which are a large factor that is difficult to calculate given the nature of available data.

This research has shown that SDoH and PTSD individually influence economic and health outcomes in civilians. Having a deeper understanding of the independent role of SDoH and the specific domains at play can help inform a more holistic PTSD prevention strategy that considers the concurrent sociopolitical and environmental factors at play. This research suggests more expansive, systemic action is needed to address SDoH factors tied to and operating in parallel with PTSD. A paradigm shift is needed toward a trauma-informed care model, which incorporates the unique conditions of each patient’s history and delves into the external etiologies. This underscores the immense need for awareness, destigmatized public discourse, and intersectional care options that cater to civilians as a distinct PTSD subgroup. This approach will benefit from future research that further defines the dynamic relationship between PTSD and SDoH. An important next step will be to identify the most effective interventional measures in these specific domains. For example, pathway programs, resilience training, and marriage/relationship counseling may be promising areas for exploration. There is also the opportunity to replicate this research in non-U.S. populations to assess whether the importance of each domain differs across cultures and geographies. In addition, it will be important to assess whether our findings hold true for intergenerational trauma and, specifically, how evolving SDoH variables shape the epigenetic transfer of trauma. Finally, an analysis assessing the differences in outcomes between the civilian and veteran populations would elucidate the extent to which SDoH exacerbates HCRU. However, this is the subject of another analysis.

In conjunction, knowing the relative weight that PTSD has for civilians with varying SDoH may be useful for healthcare providers when comparing treatment options and engaging in shared decision-making with patients. Additionally, the limited treatment armamentarium matched with the direct costs of PTSD identified in the present study offers compelling motivation for public health officials, researchers, and manufacturers to explore new treatment options for what is a financially, emotionally, and physically devastating condition. Broader acknowledgment of the civilian PTSD population only underscores the urgent need to address PTSD in all its manifold forms.

Highlights.

• Civilians with PTSD showed worse outcomes than those without PTSD

• SDoH factors added an $8.82B burden to the health system via HCRU

• SDoH explained 52.4–84.5% of poor self-reported outcomes

• Employment, stress, and education impacted patient outcomes the most