HJNO Jul/Aug 2026

U.S. HEALTHCARE JOURNALS I  JUL / AUG 2026 23 JAMA NETWORK OPEN ORIGINAL INVESTIGATION / NEUROLOGY sports, military duty)?”on an online adapta- tion of the Ohio State University TBI Identi- ficationMethod. 37 Controls were matched to a subset of football players who completed the Geriatric Depression Scale 15 (GDS-15), Everyday Cognition Scale (ECog), and Cam- bridgeAutomated Neuropsychological Bat- tery Paired Associates Learning Test First Attempt Memory Score (PALFAMS) and Total Errors Adjusted (PALTEA), without missing demographic data based on pro- pensity score. The propensity score was based on age, race (eMethods 4 in Supple- ment 1), ethnicity, and educational level. Race categories includedAfricanAmerican or Black, Asian, Native American, White, multiple races, and other race (HITSS cat- egory); and ethnicity categories included Latino or not Latino. Race and ethnic- ity data were collected because these fac- tors are associated with dementia risk and incidence. 38,39 Matching is described in eMethods 5 and eFigure 1 in Supplement 1. The second substudy evaluated associations between RHI proxies and cognitive and neuropsychiatric outcomes among HITSS former football players only. Measures HITSS participants completed the Bos- ton University RHI Exposure Assessment, which evaluates sports participation and other RHI sources (eMethods 2 in Supple- ment 1). 40 We examined 4 measures of foot- ball exposure: (1) age of first exposure (AFE) to football (the age at which the participant began playing organizedAmerican football), (2) total years of football play (calculated by subtracting the AFE from the age at which the participant stopped playing football), (3) position of play (lineman or nonlineman), and (4) highest level of play (professional, college, high school, or youth). Professional level of play consisted of National Football League (NFL), Canadian Football League, Arena Football League, NFL Europe, and XFL. No semiprofessional leagues were included. Youth and high school were com- bined into one category due to small sample size of youth-only players (n=97). Outcomes HITSS includes remote, unsupervised, computerized cognitive tests and self-report questionnaires that assess domains affected by RHI, including executive function and episodic memory. 22,41-44 Only tests common to both the HITSS and BHR batteries (PAL- FAMS, PALTEA, GDS-15, and ECog) were examined in the analyses of players and nonplayers. Analyses of the larger football cohort included the Behavior Rating Inven- tory of Executive Function–Adult (BRIEF- A) Meta-Cognition Index (MI) and Behav- ioral Regulation Index (BRI) (eMethods 3 in Supplement 1). Statistical Analysis Football vs No RHI Analyses were performed using R, ver- sion 4.3.3 (R Foundation for Statistical Computing). We quantified associations of football-playing history (vs no RHI expo- sure) with each of the 4 outcomes assessed across HITSS and BHR (PALFAMS, PALTEA, ECog, and GDS-15), using mean differences between the groups in each outcome. We estimated these mean differences by fit- ting a separate regression model for each outcome. To account for the dependency structure introduced by the matching pro- cedure and the reuse of controls, we used multiway cluster-robust SEs clustered by both participant identification and matched pair membership. 45 We used a doubly robust estimation approach by including age as a covariate in all models to account for resid- ual imbalance after matching. Standardized β values were reported as the measure of effect size. RHI Proxies in the Football Cohort We characterized associations of the 4 RHI proxies among former football play- ers with each of the 6 outcomes assessed in HITSS, using several approaches. We fitted a multiple linear regression model for each RHI measure-outcome pair. Models were parameterized to estimate the mean differ- ence in each outcome: per each year older inAFE, per each additional year of play, and among linemen vs nonlinemen. Analysis of covariance (ANCOVA) com- pared outcomes across each pair of levels of play. Omnibus tests assessed the over- all effect of highest level of play; partial η 2 was reported as the measure of effect size. We estimated marginal means, including SEs and 95% CIs, across each level of play (eTable 2 in Supplement 1). Post hoc Tukey- adjusted pairwise comparisons examined differences in performance. Sample sizes varied across outcomes due to missing data (Table 1). Linear regression and ANCOVA model P values were adjusted by applying the Benjamini-Hochberg false discovery rate method independently for each RHI proxy. The total number of hypotheses was set to 4 for the cognitive domain (PALFAMS, PALTEA, MI, and ECog) and 2 for the neu- ropsychiatric domain (GDS-15 and BRI). An FDR-adjusted, 2-sided P ≤ .05 was consid- ered statistically significant. ECog was log and PALTEA was square root transformed in linear regressions and ANCOVA due to nonnormal distribution of residuals. To investigate clinical relevance of sig- nificant group effects observed in the linear models andANCOVA, we conducted binary logistic regressions using impairment cut points for the BRIEF-AMI and BRI (T-score, ≥65), ECog (T-score, ≥1.31 46,47 ), and GDS-15 (cut point, ≥5). 48 These analyses examined whether the RHI proxies were associated with increased odds of clinically meaning- ful elevations on these outcomes. All linear and logistic regression and ANCOVAmod- els adjusted for age, educational level, race, and vascular risk (participants classified as being at vascular risk if they endorsed having any of the following: heart disease, high blood pressure, high cholesterol, or diabetes). Sensitivity Analyses Four additional linear regression models withmultiway cluster-robust SEs compared performance of controls with each level of football player on all outcomes, adjust- ing for age. The number of participants for whom youth was the highest level of play among the football cohort was small (n=97).