A total of 13 male contact sports participants who had suffered from a SRC (7 American football, 6 Rugby) were recruited from local clubs. To be eligible, participants needed to be aged 13–35 y and take part in competitive contact sports, (e.g., Rugby or American Football), at least twice a week. Participants were excluded from the study if they had: (i) a neurological or psychiatric condition (ii) previously suffered from epilepsy, febrile convulsions in infancy, seizures, or had recurrent fainting spells (iii) undergone a neurosurgical procedure (including eye surgery) (iv) any of the following fitted: heart pacemaker, cochlear implant, medication pump, surgical clips, neurostimulator (v) metal in the brain/skull (except titanium). Participants completed a standardised pre-participation questionnaire for underlying health issues (PAR-Q) and TMS suitability (Rossi et al., 2011), as well as providing written informed consent. This study was carried out following ethical approval from the local ethics committee and is in keeping with the latest iteration of the Declaration of Helsinki.

Players with a suspected concussion were encouraged to take part in the study by coaching staff. Concussions were not diagnosed by a physician, players were removed from the pitch if any visible signs of concussion (e.g., dazed look, confusion, balance impairments) were observed, according to guidance set out by Sport Scotland (2018). This approach reflects what happens in an amateur context, as most teams at this level do not have access to team physicians to diagnose any injury. The first stage of the graduate RTP protocol requires 7 days complete rest, following which players can progress through stages 2–6 every 24 h if no symptoms appear. We believe that participants in our study only experienced mild concussions as they were able to progress between stages without additional delays. Participants were tested at 4 time points throughout their recovery, within 24 h (18 h ± 4) of injury (day 0) and then at days 7, 9, and 11 post-injury, corresponding to stages 2, 4, and 6 of their graduated return to play following the Scottish Sport Concussion Guidelines (Sport Scotland, 2018; Figure 1). Participants were asked to refrain from exercise, caffeine and alcohol use 48 h before the sessions. Participants underwent the same protocol during each testing session, completing the SCAT5 (neurological function) and TMS (corticomotor inhibition) evaluation (Figure 1). Data were collected by a single individual trained by members of a research team with >20 years experience in neuromuscular assessments (Di Virgilio et al., 2019).

The SCAT5 is comprised of multiple domains: symptom evaluation, cognitive and neurological screening, and the modified balance error scoring system (mBESS). Symptom evaluation requires the participant to rate their well-being on 22 different questions at the time of testing. Each question scales from 0 (non-existent) to 6 (severe), giving a maximum possible score of 132, with a higher overall score indicating more symptoms and increased severity. The cognitive screening section is made up of the standard assessment of concussion (SAC), which tests orientation, immediate memory, concentration, and delayed recall. The neurological assessment covers multiple domains (see Supplementary Material 1) giving either yes or no responses to assess for any abnormal conditions. The mBESS requires participants to hold 3 stances (on two feet, one foot and in tandem stance) for 20 s with eyes closed. A score is obtained for each stance by adding 1 point for each error made during the 20 s. An error was recorded when the athletes moved their hands from the iliac crest, opened their eyes, made a step, had a fall, displayed hip abduction or flexion beyond 30°, lifted their forefoot or heel from the testing surface or remained out of the proper testing position for more than 5 s. The mBESS can be subjective in nature, with some of its tasks showing low interrater and intrarater variability (Finnoff et al., 2009). For the purposes of our study all errors were recorded by the same individual at all times and only the tasks found to have an acceptable intrarater variability were used.

All assessments took place indoors. Participants were seated in either a customised load cell, if assessed at the club training ground, or isokinetic dynamometer (Biodex System 4, New York, NY, United States), if assessed in our neurophysiology laboratory, with their knee at 60° flexion (full leg extension 0°). Participants undertook a standardised warm-up of 3 × 50% then 3 × 70% of their perceived maximum voluntary contraction (MVC) with 30 s recovery between exertions. Participants then performed 3 MVCs of 5 second duration each, from which 20% of MVC was calculated for the TMS. Electromyography (EMG) was used to measure surface amplitude of the rectus femoris muscle. Pilot testing and previous work within our laboratory (Di Virgilio et al., 2016, 2019) has shown that the rectus femoris is the best muscle in the quadriceps femoris for the elicitation and recording of inhibitory/excitatory responses. Responses were recorded using a wireless system (Biopac Systems, Inc. Goleta, CA, USA) and disposable Ag/AgCl surface electrodes (Vermed, Devon, UK). Following the surface EMG for non-invasive assessment of muscles (SENIAM) recommendations (Hermens et al., 2000), electrode positions were selected, shaved, and cleaned with an alcohol wipe. Electrodes were then positioned 2 cm apart. Data were sampled at 2 kHz and filtered using 500 Hz low and 1.0 Hz high band filters.

Corticomotor inhibition was assessed in the rectus femoris of participants' self-reported dominant leg by applying TMS contra-laterally over the primary motor cortex (M1). Single pulse magnetic stimulations of 1 ms duration were applied using a magnetic stimulator and 110 mm double cone coil (Magstim 2002 unit, The Magstim Company Ltd., Whitland, UK). Participants were instructed to extend their leg at ~20% of their MVC guided by visual feedback of force production. Optimal coil position for generating motor evoked potentials (MEPs) in the rectus femoris was identified by placing the coil lateral to the vertex and moving by 1 cm steps. The coil position producing the largest MEP was marked with indelible ink and used for the rest of the study (Goodall et al., 2009). Active motor threshold (aMT) was identified through participants holding 20% of their MVC whilst single pulse stimulations were applied. Intensity started at 20% of the stimulator's maximum output, increasing by 5% increments if four out of five stimulations did not elicit a visible MEP (Wilson et al., 1995). To assess corticomotor inhibition, participants were instructed to perform a 5s MVC during which a stimulation of 130% aMT was applied. Participants were instructed to maintain the contraction through the stimulation and briefly after. This was repeated 3 times with a rest period of 30 s between contractions. The cortical silent period (CSp) duration was defined as the time from the stimulus artefact to the resumption of normal voluntary EMG activity and was manually assessed (Figure 2).

CSp, SCAT-5 and balance data were analysed in IBM SPSS v27. Repeated measures ANOVAs were used to investigate the pattern of recovery after a SRC with Time (days) being the within variable; post-hoc comparisons were corrected with the Bonferroni method. In the case of non-normally distributed data, non-parametric alternative tests were used (Friedmann tests with Wilcoxon signed-rank as post-hoc—please refer to the Supplementary Material for details of normality checks). Statistical significance was set at p < 0.05.

We also used the R package RStanarm (Goodrich et al., 2020) to carry out the Bayesian alternative to the frequentist analyses described above and used the posterior distributions to examine the probabilities of the presence of an effect (Swinton et al., 2018). Small sample sizes are commonplace in concussion research, with the majority of existing TMS studies reporting data from a limited number of athletes (Miller et al., 2014; Pearce et al., 2015; Edwards and Christie, 2017). Bayesian analyses provide direct probabilistic comparisons between groups/treatments and are suited to sport science studies as they provide more defence against over-confidence in small sample sizes (Mengersen et al., 2016; Borg et al., 2018) compared to frequentist techniques. The RStanarm package provides an interface to the Stan probabilistic programming language, which uses Hamiltonian Markov Chain Monte Carlo (MCMC) method to generate MCMC chains used to characterise the posterior distribution (Hoffman and Gelman, 2014). After defining the model and priors, the other values set in the call to RStanarm were 4 MCMC chains with 2,000 iterations each and a burn-in period of 200 iterations. The priors used can be found in the Supplementary Material 1. Briefly the prior for the CSp was a normal distribution centred on the average CSp found in healthy participants in our laboratory and a standard deviation 2.5 times the standard deviation from healthy participants. Backward difference coding (found in Supplementary Material 1) was used to compare the mean of the dependent variable (CSp) for one level of the independent variable (Time in days) to the mean of the dependent variables for the prior adjacent level; day 7 compared to day 0, day 9 compared to day 7; day 11 compared to day 9.

Thirteen athletes completed the study. All athletes were male with a mean age of 20.5 (±4.5) years and were playing for local amateur rugby and American football clubs (Table 1). One of the 13 concussed participants was identified as an outlier and excluded from the analyses because the CSp values immediately after concussion were physiologically inconsistent (abnormally high) and more than 3 IQRs from the median of the sample (Supplementary Material 1).

The CSp of the concussed group significantly declined from day 0 to day 11 [F_{(3, 33)} = 7.80, p < 0.001]. Specifically, inhibition decreased between the day 0 and day 9 (p < 0.001) and day 11 (p = 0.002) post-concussion time points (Table 2).

The number of concussion symptoms significantly decreased [χ(3)2 = 30.44, p < 0.01] between the immediate (day 0) and day 7 (p = 0.002), 9 (p = 0.002) and day 11 (p = 0.002) timepoints post-concussion (Table 2); when corrected for multiple comparisons with the Bonferroni method the differences remained significant. Similarly, symptom severity significantly decreased [χ(3)2 = 25.75, p < 0.001] between day 0 and day 7 (p = 0.003), day 9 (p = 0.002), and day 11 (p = 0.003) post-concussion (Table 2); the uncorrected comparison between days 7 and 9 was significant (p = 0.011) but did not remain so after Bonferroni correction. The cognitive tasks of SCAT5 failed to reach levels of statistical significance (p > 0.05; see Supplementary Material 1). Lastly the number of errors on the mBESS decreased significantly [F_{(3, 33)} = 19.55, p < 0.001] between the day 0 and days 7 (p < 0.001), 9 (p < 0.001) and 11 (p < 0.001) post-concussion; all other comparisons were non-significant.

We used Bayesian methods to generate posterior distributions so we could assess a proportion of response (Swinton et al., 2018) for CSp and mBESS over the recovery period. For CSp values the posterior distributions with a 90% highest density interval (HDI) provide evidence that CSp duration is decreased at day 7 compared to day 0 after concussion and from day 7 to day 9 post-concussion (Figure 3). More than 90% of each of these posterior distributions lies below a difference of zero (Figure 3). The posterior distribution for the difference in CSp duration between days 9 and 11 is centred on zero suggesting no difference in CSp between these timepoints (Figure 3). Concerning the mBESS in SCAT5, the posterior distributions with a 90% HDI provide evidence that the number of errors is decreased at day 7 compared to immediately after concussion (Figure 4). The posteriors distributions provide evidence that the number of balance errors remains similar between the days 7, 9 and 11 post-concussion (Figure 4).

One major limitation of this study is the lack of a pre-concussion baseline measure. Athletes playing in local American football or rugby teams were asked to participate to the study upon suffering an SRC and having a baseline measure for all local athletes was not a feasible option. Further, from a methodological standpoint future studies would benefit from assessing inhibitory responses from whole muscle groups as opposed to single muscles, in order to understand the overall effects of SRC on the motor system. Another limitation of our study is the relatively small sample size; however, as the sample size in this case could not have been predetermined (unclear how many athletes will suffer a SRC) we believe that Bayesian statistics provide an elegant solution to address this problem, allowing us to draw inferences from smaller amounts of data (Mengersen et al., 2016; Borg et al., 2018). Moreover, the findings of this study should be interpreted with caution since the athletes assessed belonged to a specific demographic [20.5(±4.5) y.o., Caucasian males].