The Football Players Health Study (FPHS) is a transdisciplinary strategic initiative devoted to the health and wellbeing of former professional American-style football players who formerly played professional football after 1960 [21]. Using a Community-Based Participatory Research model, the FPHS regularly engages former professional football players and family members to guide research objectives, player recruitment, results translation, and dissemination of findings to former players, families, clinicians, researchers, and the general public. Starting in 2015, all eligible former players (those who signed a contract with a professional football league after 1960) received electronic and residential mail study invitations; contact information was supplied by the NFL Players Association. A follow-up survey invitation coupled to a financial incentive was similarly disseminated starting in 2019. This study has been approved by the Institutional Review Board at the Harvard T.H. Chan School of Public Health, protocol number IRB18-1365. The study was performed in accordance with the ethical standards as laid down in the 1964 Declaration of Helsinki and its later amendments.

All data were collected as part of the initial survey unless specifically noted that they were extracted from the follow-up survey. Demographic factors (e.g., age, race, height, weight) were self-reported. AFE to football was assessed with the question “How old were you when you began to play organized football?” for which participants provided their age in years. Participants also reported the number of years they actively practiced/played professional football and their primary field position.

Psychological and cognitive health at the time of survey completion were assessed using validated self-report questionnaires. The Patient Health Questionnaire (PHQ-4) is a brief self-report scale that consists of a two-item depression scale (i.e., the PHQ-2) and a two-item anxiety scale (GAD-2) that assesses symptoms over the past two weeks [22], with higher scores indicating more symptoms. Screening positively for depression and anxiety was operationalized as total PHQ-2 ≥ 3 and GAD-7 ≥ 3, respectively. Two Quality of Life in Neurological Disorders (Neuro-QoL) scales [23] were used to assess perceived cognitive difficulties (Neuro-QoL Applied Cognition—Cognitive Concerns; V2.0 Short Form) and irritability (Neuro-QoL Emotional and Behavioral Dyscontrol; Short Form). Each questionnaire contained eight items (e.g., “I had trouble concentrating,” “I had trouble controlling my temper”). Participants rated the frequency of the difficulties over the past 7 days (e.g., “Never” to “Always”). For each questionnaire, individual items are summed and the questionnaire raw score was converted to a T score to understand how the participants compared to a standard reference sample with a mean score of 50 and standard deviation (SD) of 10, consistent with standard practice [24]. Worse perceived cognitive functioning is associated with lower raw and T scores, while worse irritability is associated with higher raw and T scores.

Participants also indicated if they were ever recommended/prescribed medication by a medical provider for the following conditions (yes/no): headaches, pain, anxiety, depression, memory loss, attention-deficit/hyperactivity disorder (ADHD), hypertension, and low testosterone. In addition, participants were asked if a health care provider had previously rendered a diagnosis for any of the following health conditions (yes/no): dementia/Alzheimer’s disease (AD), vascular dementia (follow-up only), dementia other than AD/vascular dementia (follow-up only), and chronic traumatic encephalopathy [25].

Consistent with prior studies [7, 20], AFE was examined as a continuous variable, as well as dichotomized to lower than 12 years old (AFE < 12) or 12 years or greater (AFE 12 +). For continuous demographic and outcome variables, Kruskal–Wallis rank sum tests were used to determine univariable statistical significance with accompanying effect size η2. Chi-squared tests were used to examine potential differences in proportions of the AFE groups for categorical variables, with effect size Cramer’s V. The η2 effect sizes were interpreted as: < 0.01, negligible; 0.01–0.05, small; 0.06–0.013, medium; and ≥ 0.14, large [26]. Cramer’s V [degrees of freedom (df) = 1] effect sizes were interpreted as follows: 0.00–0.09, negligible; 0.10–0.29, small; 0.30–0.49, medium; and ≥ 0.50, large [26].

A series of multivariable linear regressions estimated associations between dichotomized AFE (0 = AFE 12 + ; 1 = AFE < 12) and each outcome. Similar to prior studies [25, 27], covariates included age, race, body mass index (BMI), playing position, total number of professional seasons, and concussion signs and symptoms score quartile (an approximation of prior concussion burden based on the number of times they recalled clinical concussion signs/symptoms following head impacts during football) [27]. Generalized linear regressions were used to determine whether there were linear associations between continuous AFE and each outcome. To investigate whether there were significant non-linear relationships between continuous AFE and each outcome that would not be detected by standard linear regression models, we implemented penalized spline regression in generalized additive models. The application of a penalty for the addition of each knot reduces the likelihood of overfitting and balances the model fit with the data. Model diagnostics were run to identify model misspecification. All model types (AFE as binary, AFE as a continuous linear term, and AFE as a spline) were additionally run using only age and race as covariates.

No adjustments for multiple comparisons were performed to optimize the identification of all possible significant results at the accepted risk of type 1 statistical errors. Statistics were calculated using R Language for Statistical Computing [28].