In-clinic event related potentials after sports concussion: A 4-year study

2.1 Subjects

The study followed 364 individuals aged 17–23 over the course of up to 4 sports seasons at 4 different sites. These subjects were participants in NCAA Div. 1 men’s football (172 players, representing all players from a single team), NCAA Div. 1 woman’s soccer (29 players, representing all players from a single team), men’s high school football (142 players, representing all seniors from a single team), and Junior A men’s ice hockey (20 players, representing all players from a single team).

The study was approved by the Solutions Institutional Review Board and written informed consent was obtained from the participants or their guardians before study intake.

2.2 Procedure

After completing a short 1-page intake form, participants were given an EEG test that included the oddball audio P300 component. Reaction times were also measured by asking the participant to click the track pad or mouse upon hearing the oddball tone. Touch screen Trail Making Tests A and B were added in the second year.

2.3 EEG acquisition and preprocessing

The EEG was recorded using the WAVi TM Research Platform (WAVi Research, Boulder, CO, USA) sampled at 250 Hz and bandpass filtered between 0.5–30 Hz. The electrodes were placed according to the International 10–20 system using caps with 19 tin electrodes (both with the WAVi Headset and Electro-cap International Inc., Eaton, OH, USA). Linked reference electrodes were placed at the earlobes.

The test administrators were instructed to keep the electrode impedances below 20 kOhm for ground-to-EEG locations and below 10 kOhm for the ground-to-ear locations where possible. These targets are well below the 1 GOhm input impedance of the WAVi amplifiers, are practical regarding preparation time in the sports setting (including certain “sport-related” hairstyles) and produced sufficient yield [46].

To be consistent with the goal of testing a simplified platform, a continuous 4 minute 2-tone audio oddball eyes- closed P300 protocol was used. Here, 200 common tones (1000 Hz) and 40 rare tones (2777 Hz) were delivered in random order over the span of 4 minutes, creating a 0.95 s interstimulus interval with a 50 ms tone length. The tones were delivered using Skullcandy TM over the ear headphones, at 65 dB.

While many methods exist for removing movement and other artifacts, we have chosen a technique of automatic artifacting that helps minimize test-retest variance in subjects to achieve better reliability. Here EEG segments with excessive amplitude and/or frequency activities in any of the 4 common EEG frequency bands (delta, theta, alpha, and beta) were automatically excluded from analysis on a channel-by-channel basis. Data were then visually inspected for noise to verify the accuracy of the automated artifacting.

ERP features can also be quantified by a variety of methods, including the extraction of P300 parameters at the extremum, taking voltage differences between common and rare at the extremum, and calculating areas under curves. Here, again to be consistent with our goal of investigating the simplest approaches that produce acceptable test-retest variance, we report P300 components measured by identifying the positive extremum in the latency range of 240–500 ms. The depth (P300V) is then extracted from the mean amplitude of all stimuli and the latency (P300T) is the delay recorded for that depth. These independent ERP epochs were baseline-corrected using the 100ms pre-stimulus period.

P300 parameters are typically extracted from the Cz, or Pz, or the average of various sites. Here we report a P300  μ V that is the highest amplitude from the 6 C-P scalp sites (C3, Cz, C4, P3, Pz, P4), and the fastest P300 time (smallest latency) from these same 6 C-P sites. These sites both produce an acceptable test-retest variance but have also been noted as the most useful for mTBI identification in previous studies [23].

2.5 Statistics

Categorical group comparisons were analyzed using unpaired sample t-tests for these sufficiently normal distributions. While most studies suggest that P300 amplitude should decrease after trauma, amplitudes may increase in some cases. Because the direction is unclear, we use a 2-tailed t-test to compare injured and non-injured groups. We set our p value cutoff to a more stringent alpha of 0.005, a value recommended in a recent study to address an apparent lack of reproducibility in some medical studies [47].

Successful group comparisons do not necessarily result in clinical utility and so we also constructed receiver operator characteristic (ROC) curves as a function of percent change from baseline between concussed and non-concussed groups for both reaction time and P300 amplitude.

Caution is needed when attempting to create a diagnostic marker for concussion, however, because concussion itself is defined clinically and not physiologically (or electro-physically) and must therefore rely on expert diagnosis as a standard. Furthermore, the P300, as with most concussion assessments, is itself a non-specific marker. While we do report sensitivities and specificities here, the goal of the study is not to simply develop another concussion diagnostic, but also to test the utility of a quantitative method of documenting post-injury electrophysiological change. One of these targets is to investigate changes in P300 parameters at return to play, comparing how individual P300 amplitudes normalize with respect to reaction times, neuropsychological results, and symptoms.

3.1 Reference group test-retest variability

Changes in P300 after an event form the basis of this study and so variations in a non-concussed reference group from the same population need to be understood. We can conceptualize variance in P300 metrics as a combination of trait (stable characteristics of a person) and of state (representing aspects of a subject’s psychological or physiologic state, such as concussion). To separate these traits from states (i.e., is there a concussion?) and for longitudinal tracking it’s therefore important to determine the within-person variance in routine clinical settings. Table 2 shows changes from baseline for the reference groups as a standard deviation (SD, or 1 σ ), into which 68% of the players fell. This creates a “normal” expectation. The variation in P300 parameters here is consistent with previous studies for a normal population [37, 38, 48]. Variations in reaction and Trail Making times of Table 2 are larger than P300 variations as they include systematic errors from variations in posture, in computer interfaces (e.g., track pad vs mouse), and practice effects.

In order to provide context to compare P300 to other methods of physiologic assay such as HDL or cholesterol, it’s useful to investigate P300 variance as coefficients of variation (CV), calculated as 100*SD/Average for each person and then averaged over each group. Table 3 shows the CV values of this study (which are comparable to previous P300 studies) to be lower than other common clinical assays, suggesting the clinical utility of longitudinal tracking [48, 49]. It also suggests that the P300 extraction methods of this study produce sufficient yield.

OPEN IN VIEWER

It should finally be noted that measuring change as a percentage change from baseline is one of many methods that can be employed. Other methods include investigating the voltage change themselves. Using a Bland-Altman method, a change of approximately ± 7  μ V covers a 95% limit [50, 51]. This method doesn’t change the results of this paper except that information from the lowest baselines is lost.

To investigate the general performance of P300 in relation to concussion, Fig. 1 shows average P300 values for non-injured groups (BASE and BASE for those who later became part of the CN group) and the injured group (NCN, CN, CRTP, CEND, and NEXT). Here, significant P300 voltage differences from BASE are seen after the event for both CN and CRTP groups ( 0.001, effect size Cohen’s d = 1.1/0.8 respectively). No differences from BASE are seen for the NCN group, and latency differences are not significant in any group (Fig. 1). The values for the NCN group at return to play were also not altered in a significant way from baseline, which otherwise would beg an interesting question regarding a different category of concussion.

Figure 2 plots P300 amplitude changes from initial baseline alongside performance measures for the various groups. These between-group statistics show significant P300 voltage, reaction time, and Trail A time changes after concussion ( 0.001, effect sizes Cohen’s $d>$ 1.1 for all of these measures) with only P300 voltage changes remaining significant at the beginning of return to play ( 0.001, d = 1.3) where reaction time and Trail Making had returned to baseline. This shows a clear trend with P300 amplitude normalizing at a slower rate than performance measures, which is consistent with the above-mentioned studies demonstrating the persistence of altered brain activity following concussion.

OPEN IN VIEWER

Note that the large change in the CN group with Trail B is not significant at the 0.005 level ( 0.05 d = 0.6) partially due to the larger within-person variation for this test coupled with the fact that Trails were added in the second year, reducing the number of test subjects.

The significant group differences seen between baseline and concussion normalize for all markers for the CRTP group except for P300V. The next question is how these translate on an individual basis. We explore this question below for the 2 strongest parameters: reaction time and voltage.

OPEN IN VIEWER

Figure 3 shows the ROC curve for the following conditions: (i) reaction time changes from baseline for REF versus CN groups in bins of + 12% change from baseline (AUC = 0.80); (ii) P300 voltage changes from baseline for REF versus CN in bins of 10% change from baseline (AUC = 0.92); (iii) a combined P300 voltage OR reaction time change from baseline for REF versus CN (AUC = 0.95).

For reaction time, 68% of the concussed players slowed by more than 1.5 σ compared to 23% in the non-concussed reference group. The resulting sensitivities and specificities are similar to those reported previously for reaction time, where the trading of a higher specificity for higher sensitivity was deemed appropriate for concussion screening [40]. Note that the 23% in the reference is higher than expectation because of a skew introduced by systematic errors associated with in-clinic use, (e.g., a group of administrators choosing the track pad over the mouse even though the mouse was the suggested method).

Regarding P300 voltage, 88% of the concussed players showed a greater than 1.5 σ change, compared to 16% in the reference group which followed normal expectation. Adding reaction time with P300, we see that 100% of the players had either a 1.5 σ change in voltage OR reaction time at concussion, compared to 29% in the reference group, dropping to 88% true and 10% false at the 2 σ level. If the clinician thinks the tradeoff of higher sensitivity is indeed appropriate in concussion screening, for example in youth sports, these markers can all provide screening information.

While this study shows significant P300V and reaction time changes after concussion, as hypothesized, an objective of this study is to test these parameters at the beginning of return-to-play protocols where clinical standards are less clear, particularly for the general practitioner often tasked with making these decisions.

OPEN IN VIEWER

OPEN IN VIEWER

As with the group statistics, Fig. 4 clearly shows slower P300V recovery than for standard symptom- and performance-based return-to-play assessments in some players, confirming expectations that electro-physical changes can linger after symptom resolution. It should be noted that of the players who had yet to normalize at return to play, roughly one half of those were from a subgroup whose symptoms were not attributed to a single event but rather to a series of sub-acute events such as those experienced by the offensive line in football. These players also seemed more prone to multiple year concussions, though more statistics would be needed to confirm this result.

The topographs of Fig. 6 illustrate a range of changes to be expected over the course of an NCAA football career for 3 players measured at BASE (pre-career) and post-career. These players were from three groups including one from a non-contact position, one from the CN group, and one from the NCN group. While the magnitude of the highest P300 voltage may vary, it appears that people retain the same general pattern (left-right or front-back) over time, particularly in the central-parietal region of interest. This finding reflects those of previous studies where a person’s specific P300 morphology (waveforms of Fig. 5) shows little variation over recording sessions or experiments [37].

4.2 Sample injury progression reports

The topographs of Fig. 7 illustrate a range of changes seen after a concussion event for 3 players: one whose P300 voltage recovery matched that of the standard return protocols (symptom and performance), one whose P300 voltage recovery was delayed but normalized by post season, and one whose voltage never returned to baseline. This latter player was a part of the sub-acute group, had concussions in three consecutive seasons, and never returned to initial baseline in any of those seasons.

The recovery pattern seen in player a of Fig. 7 was common, with clear P300 changes at CN that had normalized to within 1.5 σ by the time the graded return-to-play protocol began. Roughly 40% of the players, however, showed patterns of players b or c who were clinically cleared by existing standards of care but still had lingering P300 deficits. Roughly 20% of these were clinically cleared with P300 anomalies that were still $>$ 3 σ reduced from initial baseline. Finally, there were six players who had CN events repeated in consecutive seasons, and four of these failed to return to less than 2 σ of the P300 voltage baseline the season before.