A prospective observational cohort study was conducted with seven Gallagher Premiership men’s teams and six Premier 15 s/Premiership Women’s Rugby teams, who participate in the highest level of rugby union competition in England. Data were collected during 132 men’s and 69 women’s regular in-season training sessions during the 2022–23 (26 training sessions from four men’s teams, and 8 sessions from two women’s teams), and 2023–24 (106 training sessions from six men’s teams, and 61 training sessions from five women’s teams) seasons. Ethical approval was obtained from the Research Ethics Approval Committee for Health (EP 20/21 088). Across the 13 teams sampled, > 99% provided informed consent to take part in the study.

Participants were fitted with a custom Prevent Biometrics iMG (Minneapolis, MN, USA) from 3D digital dental scans taken by an experienced dentist. The iMGs are equipped with an accelerometer and gyroscope, with measurement ranges of ± 200 g and ± 35 rad/s and a sampling rate of 3.2 kHz. Embedded in the iMG were infrared proximity sensors that detect coupling to the upper dentition to start monitoring head kinematics. The Prevent Biometrics iMG has been validated in both laboratory and on-field studies [19,20,21], with a concordance correlation coefficient of 0.98 for the laboratory accuracy of head kinematic measures, and positive predictive value of 0.94 and sensitivity of 0.75 for on-field video verification validation [19]. Only HAEs that exceeded both 5 g (at the head centre of gravity) and 400 rad/s2 were used per previous rugby iMG studies [3, 11,12,13,14,15]. This threshold helps to remove non-contact HAEs (e.g., running and jumping) which might be captured, with a positive predictive value of 0.99 (95% CI 0.97–1.00) [3].

A sensor acceleration event, which is the iMG datum that approximates HAEs at the head’s centre of gravity [22], was recorded by the iMG when one of the axes on the medial–lateral, anterior–posterior, and vertical planes exceeded 8 g, capturing 50 ms of head kinematics (10 ms pre-trigger; 40 ms post-trigger). The peak linear acceleration (PLA; g) and peak angular acceleration (PAA; rad/s2) of each event was calculated from resultant values. To account for tangential and acceleration effects [23], linear head kinematics were transformed to the 50th percentile male head centre of gravity from the iMG location. Prevent Biometrics’ in-house algorithm categorised signal noise for each HAE recorded as minimal (0; n = 5660), moderate (1; n = 374), or severe (2; n = 270). A 4-pole, zero-phase, low-pass Butterworth filter was employed, with class 0, 1, and 2 signals using 200, 100, and 50 Hz cut-off frequencies, respectively. This is consistent with previous research [3, 11,12,13,14,15].

Integrating video in wearable sensor studies is recommended to support data windowing when researching on-field head kinematics [5]. Consequently, a member of the research team attended each of the participating clubs for three consecutive training weeks to capture tight video angles from height at the central point of the pitch of all field-based training sessions. As clubs had access to iMGs for the entire season and collected data routinely, teams were asked to share their analyst recorded video footage of conventional, regular-season training sessions to supplement data collected by the researcher. All angles were timestamped with the real-world time that the video recording began in HHMMSS format. This allowed temporal data windowing [5] to take place. Training activities were determined by the lead author from the video of training and were allocated into one of the activity types in Table 1, adapted from Starling et al. [18]. Training activities fitted one of three defined levels of contact; full contact (live collisions between players in an uncontrolled environment), controlled contact (play in a controlled environment at reduced speed and/or intensity, either between players or through the incorporation of pads, shields, bags, and/or crash mats), and non-contact (unopposed play with no intensity or force in contact or collisions between players). Game-based training drills, characterised by 15v15 match play simulation, were observed at all three levels of contact. Repetition-based training drills (live breakdown, live tackle drill, wrestling drill; all without the use of pads, and breakdown bags opposed, breakdown bags unopposed, tackle bag drill; all with pads) were all categorised as controlled-contact level, apart from passing drills (non-contact). The real-world start and end times of drills, defined as the duration between the start of the first set or repetition and the end of the last, were recorded as used as the training exposure duration. Simultaneously occurring activities were assessed separately to recognise potential differences in level of contact and drill type. For example, a three-station contact skills rotation was separated into nine training drills. Only training drills that were captured on video for their entirety during sampled training sessions were included in the analysis. Data from training sessions where there was no video shared, or where the video timestamp did not align with HAEs in video, were excluded from the study.

Players were asked to wear their iMG from the start of training, and refrain from removing their iMG until the end of training drills, to ensure all potential HAEs were captured for each training activity (Fig. 1). Therefore, iMG adherence when actively participating in training activities was visually inspected in a sample of players and training videos used in this study to determine whether all contact events were monitored for HAEs. A sample of 47 players (24 men [16 forwards and 8 backs] and 23 women [18 forwards and 5 backs]) were followed through 51 men’s and 45 women’s player-training sessions across four men’s and three women’s clubs, contributing to 368 men’s and 237 women’s player drill exposures. Some players were observed to not wear their iMG at all for some training drills during the training sessions sampled. However, during the training drills in which players did wear their iMG, only five men’s player-drill exposures (< 1% of player-drill exposures) were observed where a player did not wear their iMG for all contact events they engaged in. Players otherwise only removed their iMG during training drills when they were not actively participating in the drill (e.g., between repetitions). A similar visual inspection of iMG adherence behaviours was also tested in another study of HAE exposure during elite men’s and women’s rugby union training using iMGs on a smaller sample of the same population [15]. Roe et al. [15] reported that 100% of player-drill observations inspected saw iMGs worn for all contact events, and thus generalised iMG adherence behaviours to their whole study sample. Therefore, they assumed that a player had worn their iMG for their whole involvement in the training drill if they had at least one time-on-teeth log in its duration. Due to the overwhelming proportion of player exposures observed to have worn their iMG for their involvement during a training drill in this study and previously in the literature [15], the same assumption was made, and players with at least one time-on-teeth log during a training drill were included in the training exposure calculation. Training exposure was defined as the sum of drill durations. This was assessed on a player-by-player basis as their inclusion in calculations depended on whether they wore their iMG in each training drill. A custom-written MATLAB script was used to visualise time-on-teeth during training drills to identify the number of player exposures for each training activity.

An example visualisation of iMG time on teeth throughout a training session. Vertical shaded areas indicate the type of training drill and when it occurs during the training session. Horizontal zigzagged bars indicate when the iMG is being worn by players, and the gaps between them when it is not, during the training session. A player is assumed to have worn their iMG for all their involvements in a drill, and therefore all potential HAEs are captured if they have at least one time-on-teeth log in a drill. They are included in the training exposure calculation. HAEs head acceleration events, iMG instrumented mouthguard

The incidence of HAEs was calculated per player minute for each player during each contact level and each drill type that they participated in by dividing the total HAE count by total minutes of training exposure. The overall HAE incidence for each contact level and drill type was calculated by averaging the rates for all players. This was to reduce the bias of individual player variability in HAE count and training exposure and is consistent with HAE rugby match exposure research [3] and the consensus statement for epidemiological data reporting [24]. The incidence of HAEs was calculated for those at the minimum threshold (≥ 5 g and 0.4 Krad/s2) and for ≥ 25 g and ≥ 1.5 Krad/s2, to allow comparison with the literature for higher magnitude HAEs. The HAE and training exposure data were organised in Microsoft Excel (version 16.89.1) and read into R (version 4.3.1) for analysis in R Studio (version 2023.12.1 + 402). Confidence intervals at 95% confidence were estimated using bootstrapping, and resampling 1000 times as player exposures for different drills varied from 0.4 to 641.1 min.