We conducted a retrospective matched cohort study using deidentified EHR data accessed through the TriNetX Research Network, a federated platform comprising multiple health care organizations across the United States [32]. The network contains longitudinal inpatient and outpatient data, including demographics, diagnoses, and encounter information. All analyses were performed within the secure analytic environment provided by TriNetX using deidentified data in accordance with institutional policies.

The study was reviewed by the institutional review board and determined to be exempt.

The study population included young adults aged 18 to 22 years with clinical encounters recorded between January 1, 2019, and December 31, 2020. The 2019 to 2020 index period was selected to allow sufficient longitudinal follow-up of up to 5 years within the available TriNetX data while capturing a contemporary cohort of young adults receiving routine care. This period was not selected to coincide with the COVID-19 pandemic but rather reflects the most recent interval permitting extended follow-up. To ensure engagement with the health care system, patients were required to have at least one qualifying ambulatory visit during the study period.

Patients were followed longitudinally for the development of incident anxiety diagnoses through December 31, 2025, allowing for up to 5 years of potential follow-up.

The exposed cohort consisted of patients with a diagnosis of a dermatologic condition, defined by the presence of International Classification of Diseases, Tenth Revision (ICD-10) diagnosis codes corresponding to common dermatologic disorders. The index date for exposed patients was defined as the date of the first qualifying dermatologic diagnosis during the study period.

The comparison cohort consisted of patients without any recorded dermatologic diagnosis at any time prior to or during the study period. For these patients, the index date was defined as the date of the first qualifying ambulatory visit occurring within the same calendar period.

The primary outcome was incident anxiety, defined as a new diagnosis of an anxiety disorder recorded after the index date, identified using ICD-10 diagnosis codes for anxiety disorders.

To focus on incident anxiety, patients with any recorded anxiety diagnosis during the 1 year before the index date were excluded at the cohort definition stage. In addition, outcome analyses excluded patients with anxiety diagnoses occurring prior to the start of each analytic time window to ensure that all patients included in the risk set were at risk for incident anxiety during the specified follow-up period.

Outcome time windows began 1 day after the index date to reduce misclassification of anxiety diagnoses recorded during the same clinical encounter as the dermatologic diagnosis or index visit.

Baseline covariates included age, sex, race, and ethnicity, as well as measures of health care use prior to the index date, including the presence of ambulatory visits. Covariates were selected a priori based on clinical relevance and their potential role as confounders of the association between dermatologic diagnoses and subsequent anxiety.

Anxiety diagnoses were not included as matching covariates because patients with prior anxiety were excluded from the study population, and postindex variables were not incorporated into propensity score estimation.

To reduce confounding, propensity score matching was performed to create balanced cohorts of patients with and without dermatologic diagnoses. Propensity scores were estimated using logistic regression based on baseline demographic characteristics and health care use measures.

Patients were matched one-to-one using nearest neighbor matching without replacement. Balance between cohorts was assessed using standardized mean differences, with values less than 0.1 indicating acceptable balance.

All outcome analyses were conducted using the matched cohorts.

Patients were followed from the index date until the earliest of the following events: first recorded anxiety diagnosis, end of the specified follow-up window, or last available encounter in the dataset.

Separate analyses were conducted using follow-up windows of 1, 3, and 5 years following the index date. As follow-up duration increased, fewer patients contributed data to later time horizons due to censoring at the end of available follow-up.

Cumulative incidence of anxiety was calculated for each cohort at 1, 3, and 5 years following the index date. Absolute risk differences and relative risks with 95% CIs were reported.

Time to incident anxiety was additionally evaluated using Kaplan-Meier methods, with differences between cohorts assessed using the log-rank test. Hazard ratios with 95% CIs were estimated to quantify relative differences in the rate of anxiety diagnosis over time. Kaplan-Meier analyses were performed as supportive time-to-event analyses to assess consistency with cumulative incidence findings.

All statistical tests were 2-sided, and statistical significance was defined as P<.05.

Sensitivity analyses were conducted by comparing effect estimates across multiple follow-up horizons to assess the robustness and temporal consistency of findings.

This study was reviewed by the Institutional Review Board of UMass Chan Medical School and determined to be exempt from human subjects research (STUDY00002783). The research used deidentified EHR data accessed through the TriNetX Research Network. Because the data were deidentified prior to analysis and posed minimal risk to participants, no direct identifiers were available to the investigators, participants were not contacted, and no attempts were made to reidentify individuals, informed consent was not required in accordance with the Common Rule provision for exempt secondary research [33,34] and the HIPAA (Health Insurance Portability and Accountability Act) Privacy Rule standard for deidentification.

Data were analyzed within the secure TriNetX analytics environment, and results were reported only in aggregate form. No participant compensation was provided, as this study did not involve direct participant contact.

Figure 1 summarizes cohort assembly. Among young adults aged 18 to 22 years with a qualifying dermatologic diagnosis between January 1, 2019, and December 31, 2020, eligible individuals were matched 1:1 to controls without dermatologic diagnoses using propensity scores. After matching, 169,720 individuals were retained in each cohort.

Baseline demographic characteristics before and after matching are shown in Table 1. Prior to matching, modest differences were observed across race, ethnicity, sex, and ambulatory visit history. After propensity score matching, covariates were well balanced between cohorts, with standardized mean differences below 0.01 for all matched variables.

Young adults aged 18 to 22 years were identified from the TriNetX Research Network between 2019 and 2020. Patients were categorized based on the presence or absence of a qualifying dermatologic diagnosis (including acne, psoriasis, or dermatitis or eczema), and those with a prior diagnosis of anxiety were excluded. Propensity score matching was applied to create balanced cohorts with and without dermatologic conditions. The final matched cohorts were followed for incident anxiety diagnoses at 1-, 3-, and 5-year intervals after the index date.

Table 2 presents the cumulative incidence of newly diagnosed anxiety disorders at 1, 3, and 5 years following the index date. Across all time horizons, young adults with dermatologic conditions experienced a consistently higher incidence of anxiety compared with matched controls. The absolute difference between cohorts increased over time, while the relative increase in risk remained modest across follow-up periods.

Figure 2 illustrates these cumulative incidence estimates over time, demonstrating a parallel increase in anxiety diagnoses in both cohorts with a persistently higher absolute risk among individuals with dermatologic conditions.

Time-to-event analyses using Kaplan-Meier methods further supported these findings. Individuals with dermatologic conditions exhibited a modestly increased hazard of incident anxiety across follow-up periods.

The hazard ratio for incident anxiety was 1.135 (95% CI 1.088‐1.185) at 1 year, 1.121 (95% CI 1.094‐1.149) at 3 years, and 1.096 (95% CI 1.074-1.119) at 5 years, consistent with the cumulative incidence findings in Table 2; log-rank tests were significant across all analyses (all P<.001).

Despite statistical significance, Kaplan-Meier curves showed substantial overlap between cohorts, reflecting the modest magnitude of the association and reinforcing the importance of interpreting relative measures alongside absolute risk differences.