HJAR Sep/Oct 2023

HEALTHCARE JOURNAL OF ARKANSAS I  SEP / OCT 2023 35 characterize exposure 46 . This involved three steps: (1) identifying published helmet accel- erometer studies reporting hits per season, linear acceleration, and rotational accelera- tion specific to position and level of play; (2) abstracting measures of central tendency (e.g., mean or median hits per season) from each report; and (3) computing summary means of hits per season, linear accelera- tion, and rotational acceleration, specific to position and level of play, with each study’s estimate weighted in proportion to its sam- ple size. First, data were compiled frompreviously published helmet accelerometer studies in football players that reported either the number of impacts sustained per season, average linear acceleration sustained each season, and/or average rotational accelera- tion sustained each season 8–41 . Specifically, a literature review was conducted using PUBMED to identify articles published prior to 2021 with the search terms: “head impact telemetry system,” “football helmet accel- erometer,”“football helmet linear accelera- tion,”and “football helmet rotational accel- eration”. These articles were reviewed and included in the positional exposure matrix (PEM) if they fit the following criteria: • Head impacts were measured across practices and games for the entire season. • Level of play (youth, high-school, col- lege) was identified. • Mean ormedian head impact frequen- cies, linear acceleration, or rotational acceleration were reported by posi- tion played (only for high-school and beyond, given that no studies reported results for youth by position played). • Any impact event with a peak linear acceleration <10 g was excluded from analysis. A minimum cutoff of 10 g ensures the elimination of nonimpact events (e.g., jumping) from the calcula- tion of head impact frequency. Based on these criteria, 34 articles were identified. We next compiled values that were either directly reported in a specific paper or that we derived from the data reported (e.g., a paper might have reported total partici- pants and cumulative hits across all par- ticipants; mean frequency of hits per sea- son was derived by dividing cumulative hits per season by number of participants). Values were derived using arithmetic if not directly reported. The contributing studies reported either the mean or median values, both were included in the PEM and treated similarly. When studies grouped multiple positions together in their results (e.g., sim- ply reporting results for “speed” and “non- speed”positions), the aggregate information provided for each group was applied to all positions within that group, for that study. Cumulative head impacts, linear accelera- tion, and rotational acceleration for a single player across a single season were reported and derived if necessary. Mean or median impact frequencies, linear acceleration, and rotational acceleration across a sin- gle season were weighted by each study’s sample size. These weighted averages are the impacts experienced per position per season at the different levels of play (youth, high school, college). Finally, these data were used to develop the PEM. The PEM aggregated the weighted mean annual numbers and intensities (lin- ear and rotational acceleration) of expo- sures to head impacts across all reported positions and levels of play. For any miss- ing information in the PEM based on lack of available helmet sensor data (e.g., no reports of the rotational acceleration experienced by collegiate defensive backs), the average data for that level across all positions was used (e.g., the mean rotational acceleration across all collegiate positions weighted by study sample size). Because there is cur- rently no helmet sensor data for semi-pro- fessional or professional football players, collegiate data from the PEM were used to approximate these levels of play. Table 7 provides a summary of values obtained from key variables in the PEM, with the cor- responding data reported in Supplementary Source Data, per PRISMA guidelines 69 . Statistical methods Separate logistic regression models were fitted to determine the association between each exposure measure (concussion num- ber, position at highest level, duration of football play, CHII, CHII-G, and CHII-R) and CTE status (absent or present). Among those with CTE, we fitted parallel models using CTE severity (low or high) as the out- come. Because we sought information about individual-level prediction, receiver oper- ating characteristics (ROC) curves were plotted for all significant exposure mea- sures ( p < 0.05) to observe the relationship between each exposure measure and both

RkJQdWJsaXNoZXIy MTcyMDMz