Social Vulnerability and Racial Disparities in Depression Screening of US Adolescents, 2016 to 2021

Abstract

Objectives. To identify whether sociodemographic disparities exist by evaluating the association between rates of adolescent depression screening, social risk, and race.

Methods. We used a US nationwide sample of Highmark Health privately insured adolescents between 2016 and 2021 to estimate the odds of well-visit depression screening receipt as a function of social factors (Social Vulnerability Index [SVI], predicted race).

Results. Higher SVI was negatively associated with odds of screening and depended on predicted Black race (P for interaction < .01). Provider type was a strong predictor of screening, with pediatricians more likely to screen than other generalist specialties (odds ratio = 9.64; 95% confidence interval = 9.11, 10.2). Inclusion of variation at the practice location level attenuated the relationship with SVI and accounted for a large proportion of variability in screening (intraclass correlation coefficient = 0.76), although SVI remained significant ( P < .01).

Conclusions. Adolescents are more likely to receive a depression screen if they reside in lower social-risk locations or, if likely to identify as Black, higher social-risk locations. Ultimately, screening has strong ties to where an adolescent lives and the type of provider they access. (Am J Public Health. 2025;115(9):1436–1444. https://doi.org/10.2105/AJPH.2025.308149)

Adolescent depression is a significant and widespread public health concern, with a profound impact on various aspects of health, development, and natural life outcomes. Approximately 20% of adolescents experience a depressive episode before reaching adulthood,¹ which can impair academic performance, strain relationships, increase morbidity, and heighten risk of substance use disorders and suicidal behaviors.^2,3 Adolescent-onset depression is a strong predictor of recurrent depressive episodes and chronic mental health issues into adulthood, which pose further significant economic burdens from disability and loss of productivity.^4,5 The long-term impacts of adolescent depression underscore the need for early identification and intervention to mitigate extensive personal and societal costs.⁶

The connections between adolescent depression and socioeconomic status (SES) are complex and mediated by multiple factors. Adolescents from lower-SES backgrounds face higher rates of depression, which is attributable to a combination of chronic stressors, including financial and familial instability, limited access to mental health and educational resources, and higher exposure to violence.^1,2 Socioeconomic hardships can trigger some of these stressors, which then adversely affect adolescent mental health.² Taken together, low-SES adolescents routinely face barriers accessing quality and sufficient health care, which exacerbates the risk and severity of untreated or undertreated mental health disorders.³

The US Preventive Services Task Force (USPTF) recommends screening for major depressive disorder in adolescents aged 12 to 18 years,^4,5 ideally with support from systems that ensure accurate diagnosis, effective treatment, and proper follow-up. Health care professionals agree that adhering to screening guidelines is essential for effective identification, intervention, and management.⁶ Despite that, screening rates remain low, even among privately insured adolescents, highlighting gaps in health care delivery and suggesting that systemic barriers continue to hinder screening implementation.^7–9 Given the external barriers and risk of double burden among a lower-SES population, lagging screening rates may pose outsize harm through delayed identification and treatment to those at highest risk. This conflict underscores the need for more robust screening practices and systems to ensure that all adolescents are appropriately assessed for mental health issues.

However, data are lacking on how the intersection of social risk, race/ethnicity, and care team characteristics may influence screening. Furthermore, there is limited understanding of current screening rates more broadly, as recent literature typically relies on data from single locations.^6,10 Information regarding the relationship between screening rates and social vulnerability can inform policy for the provision of education and resources that will empower providers to uniformly screen and treat adolescents at risk. As such, we extracted information on adolescent members and depression screening from a large health plan claims database and linked these data to the Social Vulnerability Index (SVI) from the Centers for Disease Control and Prevention (CDC) to investigate the relationship between SVI and adolescent depression screening.¹¹

METHODS

The analytic cohort included adolescents aged 12 to 18 years across the United States who were enrolled in medical insurance through Highmark Health. To be eligible, adolescents had to be enrolled at some point between January 1, 2016 and April 30, 2021, with at least 1 well visit preceded by 6 months of enrollment. We identified well visits via outpatient claims with Current Procedural Terminology (CPT; American Medical Association, https://www.ama-assn.org/practice-management/cpt) codes (99381, 99382, 99383, 99384, 99385, 99391, 99392, 99393, 99394, 99395, 99461) or International Classification of Diseases, 10th Revision (ICD-10; Geneva, Switzerland: World Health Organization, 1992) codes (Z00121, Z00129, Z0000, Z0001). To avoid bias by member age or visit year, we selected a single random well visit as the index visit for each member. We identified depression screening from claims on the day of the index visit with the CPT/Healthcare Common Procedure Coding System (G0444, G8510, G8431, G8511, G8940, 96127) or ICD-10 code (Z1331) in any position. We excluded individuals based on discontinuous medical coverage, missing covariate data, baseline depression diagnosis, antidepressant prescription, psychotherapy treatment, or (because of the requirements of the analytic method) residing in a zip code with fewer than 3 other individuals. Most exclusions resulted from prior diagnosis or treatment of depression, or lack of sufficient continuous medical coverage. Few members were excluded because of missing covariate data (< 0.1%; for flow diagram, see Figure B, available as a supplement to the online version of this article at http://www.ajph.org). The final sample meeting these criteria was 178 972.

Measures

We extracted the SVI from the CDC’s 2018 estimates and appended it using geographic joining keys based on member residential address at index visit. The SVI is defined by 4 domains (Socioeconomic Status, Household Composition & Disability, Minority Status & Language, and Housing Type & Transportation) derived from 5-year averages of census tract data (2013–2018), and reported as an unweighted sum.¹¹ Although created to identify areas vulnerable to natural disaster, the SVI has frequently been implemented to demonstrate the importance of vulnerability in morbidity and mortality risk from chronic and noncommunicable disease.^12,13 We used the raw SVI in 2 ways: first as a categorical term with SVI broken into tertiles of low (SVI = 1.38–5.63), moderate (SVI = 5.64–7.21), and high (SVI = 7.22–13.1) social risk, and second as a standardized continuous term such that 1 unit represents 1 standard deviation. An additional version of SVI was used in select models, described where relevant in Methods.

Member characteristics were based on information from, or within, the 6-month look-back prior to the index well visit. This included sociodemographic factors, personal medical history, health care utilization (inpatient, emergency, and preventive encounters), and attributed provider specialty. Member age and gender were sourced from enrollment information, and medical history was quantified with the Agency for Healthcare Research and Quality’s Pediatric Medical Complexity Algorithm (PMCA)¹⁴ using all claims from the pre–well visit period. We calculated health care utilization based on unique service or admission dates. Well visit provider type was captured from the performing provider field on the index claim and categorized as “pediatric,” “family medicine,” or “other primary care” (e.g., multispecialty group, internal medicine) per specialty codes. We used identifiers for practice location of the performing provider to classify place of visit for sensitivity analysis.

Sociodemographic factors included predicted race and rurality of members’ residence. Because the claims database does not contain member racial or ethnic identification, we computed a proxy term for race using RAND’s Bayesian Improved Surname Geocoding (BISG) methodology. The BISG yields predicted probabilities for each of 5 race and ethnicity categories (White, Black, Hispanic, Asian, or Other) based on last name, age, gender, and census tract. We generated the BISG estimates for the cohort using the wru package.¹⁵ Prior work has validated the BISG method within national health plans and found that the estimates are highly correlated with self-identified race (average concordance = 0.93).¹⁵ For descriptive purposes only, predicted race was directly imputed based on the race or ethnicity category with the greatest predicted probability. However, all modeling incorporated the continuous predicted race probabilities, which better accounts for confidence and accuracy than direct categorical imputation. Because the BISG method includes Hispanic as its own category, predicted probabilities of race do not account for Hispanic ethnicity. Rurality designations were based on the US Department of Agriculture’s Food Access Research Atlas categorization.¹⁶ Residential zip code was additionally appended based on member address, per administrative claims, at time of index visit.

Statistical Analysis

We compared baseline characteristics by SVI tertile using Pearson’s χ² test for categorical terms and Kruskal–Wallis rank sum test for continuous terms. We set statistical significance at P < .05 for all analyses.

To assess the relationship between SVI and likelihood of receiving a depression screen at a well visit, we used a hierarchical model building approach focusing first on fixed effects and then incorporating random effects. Unlike a stepwise approach, the order of variable inclusion is driven by theory rather than the magnitude of association with the outcome. The starting model contained SVI and a quadratic term to capture the nonlinear association between SVI and depression screening (evidenced by visual inspection of raw data; see Figure 1). Then, theoretically important variables (e.g., provider type, rurality) and relevant member characteristics were added to the model one at a time. To be retained in the model, variable coefficients were required to be statistically significant and yield a significant improvement in the overall model based on a likelihood ratio test. Additionally, Akaike information criterion values were compared to ensure that significant improvements were also substantive, because negligible effects may be statistically significant given the large sample.

FIGURE 1—

Percentage of Members Being Screened for Depression as a Function of Their Percentile-Transformed Social Vulnerability Index (SVI) for (a) the Full Sample and (b) as a Function of Those Who Met Threshold for Likely Identifying as Asian, Black, Hispanic, or White: United States, 2016–2021

Note. Raw data are represented by thinner, more transparent lines, with LOESS (locally estimated scatterplot smoothing) smoothed estimates overlaid in bold lines.

Included in the variable selection process was an interaction between SVI and predicted race, based on visual inspection of these variables and proportion screened (Figure 1b). We incorporated predicted race into models as the BISG predicted probability estimate of Black race. We used this single probability estimate based on its distinct trajectory (all others had similar, negative slopes) and nonsignificance of other race categories in the variable selection process. For all models fitted with predicted race, we removed the minority status and language domain from composite SVI to avoid overfitting as they rely on the same underlying census data.

Finally, variability in screening across years and community level clustering was accounted for with random intercepts for screening year and residential zip code, respectively. Their inclusion was subject to the same variable retainment criteria described in this subsection. We calculated the intraclass correlation coefficient to quantify the variance in screening attributable to year and zip code as the proportion of variance in the data explained by the random effects, without regard to the fixed effects (i.e., the adjusted intraclass correlation coefficient) using the distribution-specific variance.

Additional analyses explored potential contributors to gaps in screening and differences in the effect of SVI across categorical predictors. First, we conducted a sensitivity analysis to quantify variability explained by practice location clustering; for this, we created a subset of the cohort, which was limited to members attending a practice that performed well visits for at least 3 members in the sample. The final model was refitted with this smaller subset of the cohort and an additional random intercept for practice location. Of the 23 545 distinct practice locations attended by the initial 178 972-member sample, 13 552 practice locations met the sensitivity sample requirement with 167 235 members. Next, differences in the effect of SVI were identified through subgroup analyses refitting the final model with data split into subsets based on different levels of categorical predictors. We included member age in subgroup analyses based on younger (< 15 years) versus older members (≥ 15 years). Finally, we ran a supplemental Poisson regression at the practice location level to identify any variation in the rate of change in screening over time across practice locations (i.e., if different practices adopted guidelines at different rates).

We conducted all analyses using RStudio Workbench (version 1.4.1717–3) running R version 4.0.0. Package reference details are included in the methods citation section of the online Appendix.

RESULTS

Cohort characteristics stratified by SVI tertile are presented in Table 1. Figure 1 shows the proportion of members screened as a function of SVI and key sources of variability in screening rates that were subsequently explored in modeling analyses. Of 178 972 total members, 23% had a depression screen at their well visit; screening rates were not uniform by SVI (Figure 1a), with a rate of 25% in low SVI areas, 21% in moderate, and 23% in high. Graphical inspection by predicted race category further illustrated that those likely to identify as Black accounted for the increased screening rates in the highest SVI areas (Figure 1b). As shown in Table 1, most of the cohort had visits occurring between 2017 and 2020, and in that time frame the rate of screening increased from less than 15% to almost 40%. This trend in increasing rates of screening over time was evident across all SVI tertiles.

TABLE 1—

Baseline Characteristics of Cohort Across Social Vulnerability Index Tertiles: United States, 2016–April 2021

Variable	Social Vulnerability Tertile, Mean ±SD or No. (%)			P a	Effect Sizeb
	Low (n = 58 937)	Moderate (n = 57 987)	High (n = 62 048)
Female	28 327 (48.1)	27 949 (48.2)	30 839 (49.7)	< .01	0.0
Age at well visit	15.0 ±2.0	15.0 ±2.0	14.9 ±2.0	< .01	0.0
BISGc race (direct imputation)				< .01	0.2
Asian	2 048 (3.5)	1 293 (2.2)	954 (1.5)
Black	1 263 (2.1)	2 688 (4.6)	10 804 (17.4)
Hispanic	2 700 (4.6)	3 863 (6.7)	10 733 (17.3)
Other	33 (0.1)	40 (0.1)	382 (0.6)
White	52 893 (89.7)	50 103 (86.4)	39 175 (63.1)
Probability Black	0.1 ±0.1	0.1 ±0.2	0.2 ±0.3	< .01	0.0
Probability White	0.8 ±0.3	0.8 ±0.3	0.6 ±0.4	< .01	0.1
Probability Asian	0.0 ±0.2	0.0 ±0.1	0.0 ±0.1	< .01	0.0
Probability Hispanic	0.1 ±0.2	0.1 ±0.2	0.2 ±0.3	< .01	0.0
Probability other	0.0 ±0.0	0.0 ±0.0	0.0 ±0.1	< .01	0.0
Urban	42 793 ±72.6	34 077 ±58.8	43 630 ±70.3	< .01	0.1
Year of index visit				< .01	0.0
2016	6 700 (11.4)	6 668 (11.5)	5 814 (9.4)
2017	10 828 (18.4)	11 133 (19.2)	10 316 (16.6)
2018	11 466 (19.5)	11 851 (20.4)	13 838 (22.3)
2019	12 087 (20.5)	11 926 (20.6)	14 053 (22.6)
2020	13 577 (23.0)	12 481 (21.5)	13 284 (21.4)
2021d	4 279 (7.3)	3 928 (6.8)	4 743 (7.6)
Census region				<.01	0.1
Midwest	9 702 (16.5)	6 777 (11.7)	6 863 (11.1)
Northeast	30 470 (51.7)	31 064 (53.6)	21 752 (35.1)
South	15 961 (27.1)	17 698 (30.5)	29 601 (47.7)
US territories/other	0 (0.0)	4 (0.0)	0 (0.0)
West	2 804 (4.8)	2 444 (4.2)	3 832 (6.2)
Emergency department	0.1 (0.3)	0.1 (0.3)	0.1 (0.4)	< .01	0.0
Inpatient stayse	3.6 (70.8)	3.9 (68.8)	5.2 (90.3)	< .01	0.0
Preventive care	0.2 (0.5)	0.2 (0.7)	0.2 (0.6)	0.5	0.0
Depression screen at visit	14 996 (25.4)	12 233 (21.1)	14 549 (23.4)	< .01	0.0
Medical complexity				< .01	0.0
Complex chronic	1 986 (3.4)	2 065 (3.6)	2 506 (4.0)
Nonchronic	46 771 (79.4)	45 798 (79.0)	47 793 (77.0)
Noncomplex chronic	10 180 (17.3)	10 124 (17.5)	11 749 (18.9)
Provider type				< .01	0.1
Family medicine	6 325 (10.7)	9 846 (17.0)	8 976 (14.5)
Other primary care	32 773 (55.6)	28 600 (49.3)	33 538 (54.1)
Pediatric	19 839 (33.7)	19 541 (33.7)	19 534 (31.5)

Note. The study cohort includes a sample of members of a national private payor. The total sample size was n = 178 972.

^a Pearson’s χ² test; Kruskal–Wallis rank sum test; Fisher’s Exact Test for count data with simulated P value (based on 2000 replicates).

^b Eta squared for Kruskal–Wallis, Cramer’s V for χ².

^c BISG = Bayesian Improved Surname Geocoding, a method for imputing race/ethnicity based on first/last name, date of birth, and census tract.

^d Partial year, only runs through April 2021.

^e Represented as rate per 1000 members.

The initial unadjusted model (i.e., a linear and quadratic term for SVI and no random effects) had a very poor fit with R² 0.002 and an area under the curve of 53%, although SVI was statistically significant (P < .01) and suggested a nonlinear relationship such that the odds of a screen initially decrease with increasing SVI up to the mean (SVI odds ratio [OR] = 0.94; 95% confidence interval [CI] = 0.93, 0.95), then increase again (quadratic SVI [OR] = 1.07; 95% [CI] = 1.06, 1.08). The iterative addition of other independent variables thought to be significant per baseline analysis (Table 1) included age, gender, predicted race, provider type, PMCA category, and rurality. All terms were statistically significant except for gender, but because of its theoretical importance, it was left in the model. Interaction terms were tested and, ultimately, a term for predicted Black probability × SVI was kept for its statistical significance, magnitude of association, and improvement in overall model fit. Finally, following the inclusion of all other covariates and random effects, the quadratic term for SVI no longer met criteria for inclusion and was ultimately removed from the model.

The significance and magnitude of the association between SVI and depression screening was maintained in the fully adjusted model despite the significance of most individual and environmental predictors incorporated, as well as the random intercepts for year of screening and zip code. The interaction between race and SVI indicated that the relationship between SVI and depression screening depended on the likelihood of predicted Black race, where those likely to identify as Black in high SVI areas had higher predicted odds of screening than those unlikely to identify as Black in high SVI areas (P  < .01; see Figure 2 for a depiction of the moderated association).

FIGURE 2—

Predicted Probability of Being Screened for Depression as a Function of a High (P[Black] = 1) or Low (P[Black] = 0) Predicted Probability of Black Race From Generalized Linear Mixed Effects Models (a) Without (Fully Adjusted) and (b) With (Sensitivity) Terms to Control for Variability by Practice Location: United States, 2016–2021

Note. SVI = Social Vulnerability Index. Predictions are based on the midpoint of the study period for both models (year = 2019); for the sensitivity model, a random provider location whose percentage of members with depression screenings was at the average for that year (mean = 20%) was chosen.

Subgroup analyses overall revealed little variation in the magnitude of the association between SVI and screening by individual member characteristics such as gender, age, medical complexity, or rurality (Figure B, available as a supplement to the online version of this article at http://www.ajph.org). However, models suggested differences by provider type such that SVI was not significantly associated with screening within family medicine providers, and was more strongly associated with reductions in screening within less specific provider specialties. Ultimately, the sensitivity model revealed that a large proportion of the variability in screening likelihood could be explained by clustering at the practice level: the addition of the random intercept representing practice location increased the intraclass correlation coefficient from 0.38 to 0.78 (see online Table A for full comparison of random effects variances across models). This addition attenuated both SVI and predicted Black race, although they remained statistically significant (Table 2). Ancillary Poisson analyses further demonstrated significant variability in the yearly trend (likelihood ratio test to reduced model P < .01) across practice locations (random slope, 0.18; full model output in online Table C).

TABLE 2—

Odds of Depression Screening Receipt in the Context of a Well Visit as a Function of SVI and Other Predictors: United States, 2016–April 2021

	OR (95% CI)
Characteristic	Fully Adjusted	Sensitivity
SVI	0.85 (0.84, 0.87)	0.91 (0.89, 0.94)
Provider type
Family medicine (Ref)	1	1
Other primary care	2.17 (2.04, 2.31)	3.39 (2.92, 3.93)
Pediatric	9.64 (9.11, 10.2)	5.42 (4.59, 6.40)
Age	0.96 (0.95, 0.96)	1.00 (0.99, 1.01)
Rurality (Ref = urban)	1.16 (1.11, 1.20)	1.07 (1.02, 1.13)
BISG probability Black	1.40 (1.29, 1.52)	1.30 (1.16, 1.47)
Medical complexity
Nonchronic (Ref)	1	1
Noncomplex chronic	1.05 (1.01, 1.08)	1.01 (0.96, 1.06)
Complex chronic	1.01 (0.94, 1.08)	0.80 (0.72, 0.88)
Gender (Ref = female)	0.98 (0.95, 1.00)	1.01 (0.97, 1.05)
SVI × BISG Probability Black	1.18 (1.10, 1.26)	1.16 (1.06, 1.27)

Note. BISG = Bayesian Improved Surname Geocoding (a method for imputing race/ethnicity based on first/last name, date of birth, and census tract); CI = confidence interval; OR = odds ratio; SVI = Social Vulnerability Index. The fully adjusted model includes the full sample cohort (n = 178 972); the sensitivity model includes a restricted sample of those attending a practice with at least 3 cohort members (n = 167 235). Study cohort includes a sample of members of a national private payor.

DISCUSSION

The goal of this study was to describe patterns of adolescent depression screening at annual well visits by measures of social vulnerability. Our findings show that screening remains relatively low (approximately 23% of members), and that there is a nuanced relationship between the odds of screening and SVI. Although SVI is negatively associated with screening, the relationship depends on Black race, such that those most likely to identify as Black residing in areas with the highest social risk had some of the greatest odds of screening across the SVI distribution. Of note, pediatricians far outpaced other provider types in screening. Ultimately, the sensitivity model clarified that much of the variability in depression screening across race and social risk can be explained by differences in group behaviors at the practice level.

The finding that provider type and degree of adherence to guidelines varied with geographic measures of social risk suggests that depression screening receipt is, in part, an issue of access and provider ability. Furthermore, the increased screening rates within higher SVI areas seen in the unadjusted data ultimately depended on a higher likelihood of identifying as Black. Members who were likely Black and resided in urban areas were seen by pediatricians at rates higher than any other race category (online Table B). It is possible that the density of urban areas allows for both greater proximity and greater availability of pediatric providers, and thus higher screening rates. The same providers may also be more empowered to screen, diagnose, and treat, as some pediatricians serving a primarily urban, high social risk case mix report higher rates of comanagement with pediatric mental health providers.¹⁷

Although increased screening rates among those likely to identify as Black is a positive sign, it is important to note that minority groups may also be disproportionately disadvantaged by living in less connected areas relative to their White peers. A study of pediatric access, availability, race, and mental health screening found that although Black, lower-SES children benefited from urban access, relatively short increases in drive time to a pediatrician had the strongest negative influence on screening rates among Black and Hispanic children.¹⁸ Furthermore, in our sample, most of those unlikely to identify as Black in higher SVI areas had far lower model predicted odds of screening receipt, indicating persistent disparities by geography.

Evaluation of screening rates by broader indicators of social risk is uncommon, although there is existing evidence that general geographic factors like pediatrician access and availability are associated with screening.¹⁸ Others have shown that screening completion can vary by individual factors such as native language, child demographics, insurance status, parental beliefs, or provider consistency, although reports are mixed on the direction and magnitude of association.^7,8,19 Our results showed that the bulk of variability in screening rates is primarily determined by practice-level tendencies, which may be more modifiable. Although education and training likely play a large role in these tendencies, published research suggests that resources may also be key, as providers cite time constraints, disagreement on the role of general pediatric providers in mental health management, and concern with a lack of available follow-up.^20–22 In the context of these findings, solutions that reduce direct time burdens on providers to perform screening, and ensure that providers have viable mental health care referral sources, could help improve screening guideline compliance. However, these solutions are speculative and would require direct study.

Although screening remains lower than ideal across studies, it appears that the USPTF guidelines have had a positive impact. A report on rates in the mid- to late 2010s found minimal screening overall, with just 1.8% of adolescents with a documented screen across a 5-year follow-up.⁸ More recent analyses, as well as rates reported in this study (Table 1), show evidence of marked increases over time, although most existing estimates are often limited in scope.^6,10 Our supplemental analysis to identify variation in the rate of change in screening over time across providers revealed significant variation in adopting screening guidance, with some quick to change and others trailing behind. Although this study was not designed to identify specific provider tendencies that may influence screening, this finding and our primary results demonstrate the importance of the provider in closing screening gaps and the need for more research on the interface of provider behavior, geography, and individual determinants. Empowering physicians to better manage adolescents with depression and improving competency in pediatric mental health will be necessary to meet the needs of adolescents and avoid significant burden over the life course.

Limitations

This work presents a robust examination of the nuanced relationship between social risk, race, provider, and depression screening. Still, there are limitations. Although the BISG algorithm presents an opportunity to consider race where otherwise impossible, it is subject to misclassification. We attempted to account for the uncertainty in members’ predicted race by incorporating continuous predicted probabilities as model terms rather than a direct imputation, although this can make interpretation more complex. Because members’ index well visits occurred anywhere between 2016 and 2021, and all SVI data were from 2018 five-year estimates (spanning 2013–2018), there is a possibility that the SVI estimates may not reflect true area-level vulnerability across the study span. However, given that changes in geographic social risk are slow, it seems unlikely that using the 2018 estimates would yield substantive differences.

It is also important to note that claims-based measures of health status and utilization are subject to billing requirements, accuracy of data entry, and practitioners’ time to comprehensively chart all elements of visits. As such, the estimates herein may reflect underestimates of actual use and experience. However, we used measures of health status validated for use with administrative data where possible. Although we cannot account for screens that occurred and were never recorded, we investigated whether excluding depression screenings at other visit types may have affected our findings by including any screen within 18 months of a member’s randomly chosen well visit. The overall rate of screening increased only marginally (from 23% to 25%), and the pattern of results remained the same.

Despite these limitations, the opportunity to leverage administrative claims data from a large national payor allowed for the inclusion of adolescents and providers across a large geographic area and more accurate accounting of screening rates than single health system analyses or provider self-report.

Public Health Implications

Early identification of adolescent depression is needed to mitigate the burdens of and risks to adolescents’ current health and longer-term impacts into adulthood; this is particularly important among those at higher risk because of social determinants or systemic inequalities. The USPTF recommends routine screening for depression in annual well visits to address this need, although our data show that uptake is still relatively low. Geographic-based social vulnerability is associated with a lower likelihood of screening, even after accounting for patient, geographic, and provider-level factors, although the nature of the association is complex and moderated by race. Ultimately, practice-level tendencies explain much of the variability in screening rates. Further study is needed to understand provider barriers and facilitators of screening across locations of varied social risk, which may lead to equitable solutions for improving identification and intervention.