Table 5.

Association between exposure measures and CTE severity (mild versus severe)

	Odds ratio (95% CI)	p-value	BIC	Cross-validation mean error	AUC	p-valuea
Concussion number	1.00 (1.00–1.00)	0.49
Position, non-speed vs speedb	1.37 (0.79–2.36)	0.26
Duration of play per year	1.14 (1.08–1.20)	4.5 × 10−7	399.9	0.138	0.668
CHII per 1,000 hits	1.13 (1.05–1.22)	9.2 × 10−4	418.2	0.144	0.674	0.75
CHII-G per 10,000 g	1.19 (1.12–1.26)	1.1 × 10−8	390.2	0.134	0.717	8.5 × 10−7
CHII-R per 1,000,000 rad/sec2	1.20 (1.13–1.28)	1.7 × 10−8	391.8	0.134	0.726	7.6 × 10−6

Separate logistic regressions were run for each exposure measure due to multicollinearity to determine odds ratios and p-values. For models with significant exposure measures, BIC and the mean error resulting from a 10-fold cross-validation analysis are reported to determine relative model performance. All models had the outcome of CTE severity (mild vs severe) and were adjusted for age at death

AUC receiver operating characteristics area under curve, BIC Bayesian information criterion, CHII cumulative head impact index representing estimated number of head impacts per donor per 1,000 hits, CHII-G cumulative head impact index representing estimated cumulative g-force experienced by each donor per 10,000 g, CHII-R cumulative head impact index estimated cumulative rotational force experienced by each donor per 1,000,000 rad/sec², CTE chronic traumatic encephalopathy

^aAUC p-value represents results of bootstrap analysis with 2000 replicates drawn from the sample to determine if there was a true difference between the AUCs for models examining CTE status and duration of play compared to other exposure measures

^bPresented as dichotomous non-speed (offensive and defensive lineman) vs speed (all other positions) for all football players with a single known position