Training-related concussions in men’s English Premiership Rugby Union: an eight-season prospective cohort study of 153 243 player-weeks

Abstract

Objective

To describe the incidence, severity, burden and risk factors of training-related concussion injuries in elite men’s Rugby Union.

Methods

Training-related concussions recorded as part of the Professional Rugby Injury Surveillance Project were examined between the 2016/2017 and 2023/2024 seasons. Incidence rates (per 1000 hours) were calculated with 95% CIs. Poisson regression and incidence rate ratios described season-to-season changes in concussion incidence. A generalised linear mixed model assessed the influence of time in season, match minutes played in the previous 28 days, dominant training type and previous concussion as fixed effects, accounting for weekly training exposure. Random intercepts included player and team.

Results

The incidence of concussions in training was 0.23 per 1000 hours (95% CI 0.22 to 0.29). The mean severity (days absent) of concussions was 18 days (95% CI 15 to 21). The burden of training concussions was 4.16 days absent per 1000 hours (95% CI 3.57 to 4.85). The tackle (46%) and accidental collisions (26%) were the most common inciting events. The incidence of training-related concussions was significantly higher in the 2021/2022, 2022/2023 and 2023/2024 seasons compared with the 2016/2017 season. Preseason (incidence rate ratio (IRR) 1.80 (95% CI 1.24 to 2.60), p=0.01) and full-contact training (IRR 2.20 (95% CI 1.17 to 4.12), p=0.01) were associated with a higher risk of concussion compared with the final 3 months of the season and non-contact training.

Conclusion

Full-contact training and the preseason period are associated with an increased risk of training-related concussion. Accordingly, full-contact and preseason training should be strategically planned to minimise risk.

WHAT IS ALREADY KNOWN ON THIS TOPIC.

WHAT THIS STUDY ADDS

• While concussion incidence during training is substantially lower than in matches, concussion risk is elevated during preseason periods and full-contact sessions.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• Club and governing bodies should consider strategies and policy that encourage a graduated return to contact during the preseason, and delay the introduction of full-contact training in addition to a more detailed video analysis of the factors associated with concussion during in-season full-contact training sessions.

Introduction

Rugby Union (hereinafter rugby) match injury incidence is relatively high, with concussion the most common injury.^1–3 Concussion surveillance is of importance because of a higher risk of subsequent concussion,^4–6 as well as musculoskeletal injuries,⁷ and concerns of long-term brain health⁸ compared with players who have not sustained a concussion.

Most rugby concussion research has focused on matches. In elite rugby training, concussion incidence has previously been reported as low (0.09 per 1000 hours) between 2002 and 2018.⁹ Players spend, on average, 6.75 hours in training per week, with 84 min (21%) in contact training,⁹ compared with an 80-minute match. As a more modifiable environment with potentially higher control compared with matches, training may be a target for injury prevention strategies.¹⁰

There have been both increased awareness and improvements in concussion diagnosis over time. Therefore, there is a need to quantify current concussion rates in elite rugby training, and associated risk factors. Concussion risk might be expected to be higher during full-contact training compared with controlled-contact training, for example. Training content and structure change throughout the season,¹¹ and the season periods might possess different training concussion risk. Players with minimal match exposure during a season were found to have elevated match concussion incidence rates, but not training-related concussion incidence rates, in the proceeding season.¹² However, the influence of short-term match exposure (eg, previous 28 days) on training risk has not been investigated. Additionally, while concussion history is associated with subsequent concussion risk,^4–6 it is unknown whether this explicitly applies to training-related concussions.

The aim of this study was to describe the incidence, severity and burden of training concussions, and examine risk factors associated with training-related concussion injury risk in the English Premiership over eight seasons.

Methods

Participants

Data were collected through the Professional Rugby Injury Surveillance Project (PRISP) from all English Premiership clubs, the highest men’s league in England. Several changes in concussion policies have been implemented, with these strategies coinciding with an increase in match concussion incidence between 2009/2010 (3.9 per 1000 hours) and 2016/2017 (20.9 per 1000 hours).¹³ This has subsequently levelled off. As such, training concussion data from 2016/2017 to the 2023/2024 season were included in this study. Data were collected from first-team players across all English Premiership teams. Players provided informed consent to participate each season irrespective of consent status in previous seasons. During the eight-season period, <1% did not provide consent.

Procedure

Club medical staff recorded injury data on the electronic medical note-keeping system ‘Rugby Squad’ (The Sports Office, UK). An injury and its severity were defined as per Fuller et al,¹⁴ and the concussion definition aligned with Concussion in Sport Group international consensus statements.¹⁵ Orchard Sports Injury and Illness Classification System codes V.10¹⁶ were used, with codes HNCX (concussion), HNCA (acute concussion) and HNCO (acute concussion with visual symptoms) to report concussions. For each reported concussion, practitioners entered information including the training activity in which the injury occurred and the inciting injury event. Baseline data including players’ main playing position were collected at the start of each season. Data completion, validation and accuracy checks were conducted prospectively during data collection according to the IOC consensus statement.¹⁷ Concussion data entry on ‘Rugby Squad’ was cross-checked with data from World Rugby’s SCRM app (World Rugby, Ireland), an online platform where all Head Injury Assessment results are recorded by the practitioners conducting them, to ensure all concussions were recorded.

Club sports science and strength and conditioning staff reported weekly training exposure on a bespoke electronic platform, ‘EliteHub’. Training exposure was captured at team-level and calculated as the number of training minutes per week multiplied by the average number of players attending training that week. Each training week, practitioners reported the type of training activity, the number of first team players and the duration. Each season, quality checks were made to ensure exposure data accuracy.

Statistical analysis

Data were downloaded into Microsoft Excel (V.16.101.2), read into R (V.4.5.1) and analysed in RStudio (V.2025.09.1+401). Injury incidence was calculated by dividing the total number of concussions by total training exposure, expressed per 1000 hours. To account for player-level clustering,¹⁸ 95% CIs were calculated using the sandwich package.¹⁹ Injury incidence and 95% CIs were calculated for overall training, activity types (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; non-contact: unopposed play with no intensity or force in contact or collisions between players; conditioning: skill, endurance, speed or weight conditioning based activities), and by season period (preseason, first 3 months, second 3 months, last 3 months). The mean and median (95% CI) severity of concussions were calculated. Concussion burden was calculated by summing days lost and dividing by total training exposure, expressed as days absence per 1000 hours. Bootstrapping with player-level clustering was used to calculate CIs.²⁰ Poisson regression with incidence rate ratios (IRRs) and 95% CIs was used to assess season-to-season changes in concussion incidence, with statistical significance set at p<0.05.

As training exposure data were collected per week, players were assigned a binary outcome for the week (no concussion/concussion). As such, all players provided an observation each training week of every season under surveillance. It was assumed that, unless injured, every player took full part in all their team’s training activities reported for their respective positional group that training week. Player-week observations were excluded if the player was absent due to any time loss injury, with the period of unavailability calculated using the injury onset date reported and the total days absence reported. It was assumed training exposure was evenly spread across a 5-day training week. If a player returned from time loss injury during the training week, their player-week observation of training exposure was calculated proportionally. Training exposure was adjusted for consenting players.²¹ The number of match minutes played at the start of each player-week observation in the 28 days prior¹² was included as a risk factor, calculated using electronic match card submissions to the Rugby Football Union Game Management System for domestic competitions and match coding by Oval Insights (Oval Insights, London, UK) for European and International fixtures. For each player-week observation, training exposure in each activity type was calculated and dominant training activity identified as where players spent most of their training time. Concussion history was included in the model, defined as a player who had been diagnosed with a match or training concussion in the previous 12 months, identified using the PRISP data.

A generalised linear mixed model (GLMM) with binomial outcome using a complementary log-log link was fitted to assess risk markers for concussion. The model was estimated using maximum likelihood estimation with the Laplace approximation, as implemented in the R package glmmTMB.²² Fixed effects included time in season, match minutes in the previous 28 days, dominant training type and previous concussion history, with log-transformed total weekly exposure as an offset. Random intercepts for player and team were included to account for clustering.

Model validation was performed using graphical and statistical diagnostics using the performance package²³ (online supplemental figure 1). Model fit was evaluated using information criteria, specifically Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC), provided by the glmmTMB package.²² Results are presented as incidence rate ratios (IRR) with 95% CIs.

Equity, diversity and inclusion

All first-team players in the men’s English Premiership were eligible for participation. All participants were therefore male and represented a range of ethnicities. The authorship group is comprised of one woman and eight men, spanning multiple career stages. The lead author is an early-career researcher.

Patient and public involvement

Patients and/or public were not involved in the design, conduct, reporting or dissemination plans of this research.

Results

Incidence, severity and burden

Over eight seasons, 1056 unique players were exposed to training. On average, players were sampled during 4±2 seasons. There were 257 training concussions recorded (18.7% of all concussions) during 1 102 070 hours of training exposure (table 1). During the study period, 208 (19.7%) participants had at least one training concussion, 25 (2.4%) players had two training concussions, 9 (0.9%) players had three training concussions and 2 (0.2%) players had four training concussions. The incidence of concussions in training was 0.23 per 1000 hours (95% CI 0.22 to 0.29). The mean severity of training concussions was 18 days (95% CI 15 to 21). The median severity of training concussions was 12 days (95% CI 11 to 13). The burden of training concussions was 4.16 days absent per 1000 hours (95% CI 3.57 to 4.85). Training concussion incidence was 0.25 per 1000 hours (95% CI 0.23 to 0.33) (table 1) for forwards and 0.21 per 1000 hours (95% CI 0.18 to 0.29) for backs.

Table 1. Concussion count, incidence, total days absence, mean and median severity, and burden in training, by position, period in season and training activity (95% CI).

		Concussion (N)	Training exposure (hours)	Incidence (N per 1000 hours)	Days absent (N)	Average severity (days absent)	Severity range (days absent)	Median severity (days absent)	Burden (days absent per 1000 hours)
Training		257	1 102 070	0.23 (0.22 to 0.29)	4580	18 (15 to 21)	7–167	12 (11 to 13)	4.16 (3.57 to 4.85)
Position	Forward	157	633 338	0.25 (0.23 to 0.33)	2996	19 (15 to 23)	7–167	13 (11 to 15)	4.73 (3.84 to 5.78)
	Back	100	468 732	0.21 (0.18 to 0.29)	1584	16 (13 to 19)	7–123	10 (8 to 14)	3.36 (2.64 to 4.21)
Period in season	Preseason	99	266 390	0.37 (0.30 to 0.45)	1722	17 (14 to 21)	7–123	12 (10 to 14)	6.46 (5.15 to 8.01)
	First 3 months	62	297 119	0.21 (0.16 to 0.26)	1017	16 (12 to 21)	7–142	11 (9 to 13)	3.42 (2.64 to 4.50)
	Second 3 months	61	303 120	0.20 (0.15 to 0.25)	929	15 (12 to 19)	7–60	12 (9 to 13)	3.06 (2.52 to 3.67)
	Last 3 months	35	235 441	0.15 (0.10 to 0.20)	912	26 (15 to 37)	7–167	14 (9 to 18)	3.87 (2.47 to 5.58)
Training activity	Full contact	128	50 524	2.53 (2.10 to 2.99)	2296	18 (14 to 22)	7–167	12 (9 to 13)	45.44 (36.64 to 55.56)
	Controlled contact	106	182 149	0.58 (0.47 to 0.70)	1925	18 (14 to 22)	7–142	12 (10 to 14)	10.57 (8.45 to 13.04)
	Non-contact	12	204 929	0.06 (0.03 to 0.09)	153	13 (8 to 17)	7–33	12 (8 to 13)	0.75 (0.52 to 1.04)
	Conditioning	2	392 958	0.01 (0.00 to 0.01)	24	12 (8 to 16)	10–14	10 (10 to 14)	0.06 (0.05 to 0.07)

The incidence of training concussions was higher in the preseason (0.37 per 1000 hours (95% CI 0.30 to 0.45)), compared with the first 3 months (0.21 per 1000 hours (95% CI 0.16 to 0.26)), the second 3 months (0.20 per 1000 hours (95% CI 0.15 to 0.25)) and final 3 months (0.15 injuries per 1000 hours (95% CI 0.10 to 0.20)) of the season (table 1) (figure 1).

Figure 1. Training concussion incidence and severity in the English Premiership between 2016/2017 and 2023/2024 during periods of the season. The preseason incidence error bars do not overlap with other periods in the season, inferring a meaningful difference.

The incidence of training concussions was 4.4 times higher in full-contact training (2.53 per 1000 hours (95% CI 2.10 to 2.99)) compared with controlled-contact training (0.58 injuries per 1000 hours (95% CI 0.47 to 0.70)) (table 1) (figure 2). There were nine training concussions recorded with the training activity ‘unknown’.

Figure 2. Training concussion incidence and severity in the English Premiership between 2016/2017 and 2023/2024 during different training activities. The full-contact and controlled-contact incidence error bars do not overlap with other training activities, inferring a meaningful difference.

Training concussions were most associated with the tackle (46%), followed by accidental collisions (26%). The incidence of training concussions was the same for the tackler and ball carrier (both 0.05 per 1000 hours (95% CI 0.04 to 0.07)).

Trends in training concussion

Using the 2016/2017 season as a reference, there were no statistically significant differences in training concussion incidence between 2017/2018 and 2020/2021 (table 2). However, training concussion incidence was significantly higher in later seasons, with rates more than twice as high in 2021/2022 (IRR 2.27 (95% CI 1.34 to 3.82), p=0.002), 2022/2023 (IRR 2.50 (95% CI 1.46 to 4.30), p<0.001) and 2023/2024 (IRR 2.48 (95% CI 1.45 to 4.25), p<0.001). There were no statistically significant season-to-season changes in mean concussion severity (F(7 248)=0.61, p=0.75, adjusted R²=−0.01).

Table 2. Incidence, severity and IRRs with significance set at p≤0.05 and 95% CIs for season-to-season trends in training concussion.

Season	Incidence/1000 hours	Mean severity	IRR	95% CI lower	95% CI upper	P value
2016/2017 (intercept)	0.14	16	Reference	Reference	Reference	Reference
2017/2018	0.21	16	1.48	0.85	2.56	0.16
2018/2019	0.21	19	1.46	0.86	2.49	0.16
2019/2020	0.19	15	1.32	0.76	2.27	0.32
2020/2021	0.17	20	1.21	0.64	2.30	0.56
2021/2022	0.32	17	2.27	1.34	3.82	<0.01*
2022/2023	0.35	18	2.50	1.46	4.30	<0.01*
2023/2024	0.35	23	2.49	1.46	4.25	<0.01*

^*Indicates statistical significance.

IRR, incidence rate ratios.

Risk factors associated with training concussion

The mixed effects binomial regression (cloglog link) model included 153 243 player-week observations. Model performance indices indicated adequate fit (AIC=3791; BIC=3910; marginal R²=0.06), with a log-loss of 0.01 and root mean squared error of 0.04. The model showed underdispersion (0.61). Model diagnostics of residuals indicated distributional assumptions were met, with no severe violations suggesting misspecification, non-linearity or heteroscedasticity (online supplemental figure 1). No obvious influential outliers biased inferences.

Player-level random effects demonstrated large variance (intraclass correlation coefficient (ICC)=0.59), compared with team-level random effects (ICC=0.00). The model identified significant factors associated with concussion risk after accounting for cumulative exposure, and player and team clustering. Using the last 3 months of the season as a reference category, the preseason exposure period was associated with greater training concussion risk (IRR 1.80 (95% CI 1.24 to 2.60), p=0.01). Concussion risk in the first 3 months (IRR 1.28 (95% CI 0.87 to 1.87), p=0.20) and second 3 months (IRR 1.20 (95% CI 0.81 to 1.77), p=0.37) did not differ from the last 3 months. Using non-contact training exposure as a reference category, full-contact training exposure was associated with greater training concussion risk (IRR 2.20 (95% CI 1.17 to 4.12), p=0.01), but not controlled-contact training (IRR 1.41 (95% CI 0.91 to 2.17), p=0.13) and conditioning training (IRR 1.07 (0.74 to 1.54), p=0.72). Concussion history was not associated with an increased risk of training concussion (IRR 0.90 (95% CI 0.49 to 1.65), p=0.74) compared with players without a concussion history. Match minutes played in the previous 28 days were not associated with an increased risk of training concussion (IRR per minute 1.00 (95% CI 1.00 to 1.00), p=0.12).

A summary of our key findings is presented in figure 3.

Figure 3. Central figure summarising key findings from the present study, including the low training-related concussion risk compared with matches, but the increased risk observed during preseason and full-contact training. IRR, incidence rate ratio.

Discussion

This study aimed to describe the incidence, severity and burden of training-related concussions and the risk factors associated with such concussion risk. Training-related concussion incidence was relatively low compared with matches but significantly higher in the latter three seasons compared with 2016/2017. Preseason and full-contact training were associated with significantly higher training concussion risk, while match minutes played (previous 28 days) and concussion history were not. Player-level random effects showed large variance in the GLMM.

The incidence of training concussions is approximately 80 times lower in training compared with matches. Training-related concussions are relatively uncommon with 80% of players not sustaining a concussion in training over the eight seasons. Assuming an average Premiership player spends ∼6.75 hours in training per week,⁹ each player would experience one training concussion every 639 weeks (12 years). During the study period, on average, there was one concussion observed every 17.8 weeks within each club or one every 1.4 weeks across all teams in the league. This is lower than the 4.7 concussions per round of match fixtures between 2016 and 2019.³ Furthermore, the average incidence of match concussions between 2016 and 2019³ was 106 times higher than training in the present study (19.75/1000 hours vs 0.23/1000 hours). While training concussion incidence in 2021–2024 was twice as high compared with earlier seasons in the present study and previously reported,⁹ it remains much lower than matches.

Full-contact training was associated with a higher concussion incidence than other training activities. Full-contact training has the highest incidence of overall training injuries in the same cohort (9.6 per 1000 hours) but where players spend the lowest proportion of their total training time (24 min/player/week).⁹ Furthermore, head acceleration event (HAE) incidence, the biomechanical mechanism of concussion injuries,²⁴ is highest in full-contact training.²⁵ Concussion is more likely during full-contact training, during which player collisions are uncontrolled, simulating match scenarios. However, concussion incidence in full-contact training is still ∼8 times lower than that in matches.³ Full-contact training appears to comprise elements, events or behaviours that pose lower injury risk compared with similar contact events in matches.

Preseason training volume, where a significantly increased training concussion risk was observed, is higher compared with in-season.⁹ Following an off-season rest period, players are reintroduced to periodised contact training, initially focusing on contact skill development.²⁶ Mid- to latter stages of preseason focus on rugby skills training. A graded exposure to contact is required to condition players, support proficiency and safe execution.²⁷ It remains unclear whether increased preseason concussion risk is driven by higher training loads and longer full-contact sessions, inefficient graded exposure or an inherent degree of risk associated with reintroducing contact.

Player-level random effects showed large variance (ICC=0.59), compared with the team-level (ICC=0.00), implying meaningful heterogeneity between players. One-fifth of players who experienced a training concussion in the present study sustained at least two. Interplayer variability has been demonstrated in HAE incidence both in matches^{28 29} and training.²⁵ Future research should explore player-level training intensity and exposure, particularly during the preseason.

Although very low compared with matches, concussion incidence in the latter 3 years of the present study (2021–2024) was significantly higher than 2016–2021. The incidence of training concussions reported between 2007 and 2018 in the same cohort was lower (0.09 per 1000 hours).⁹ Training-related concussion incidence has followed a trajectory similar to matches, with an initially stable rate preceding a growth period.¹³ The reasons for higher concussion incidence between 2021 and 2024, compared with the first five seasons of the present study, remain unclear. Improvements in concussion recognition during matches and the coinciding rise in reported match concussions¹³ may have been less pronounced in training. This could reflect a prevailing focus on match-day concussions and the diagnostic processes and technologies specifically applied in that context. Consequently, awareness and recognition of concussions during training may have lagged but now caught up. It is also possible that the intensity, composition or duration of contact training activities has changed since they were last reported.⁹ Future research should examine contact training exposure in greater detail to better contextualise increasing incidence of training-related concussions.

Concussion history was not associated with an increased training-related concussion risk. In other studies, concussion history has been associated with higher HAE magnitudes⁵ and odds of a second concussion.^{4 6} There could be multiple reasons for the present finding. Only 20% of players in the current study recorded a concussion in training over eight seasons, and only 4% of player-training-week observations involved a player with a concussion history (a diagnosed concussion in the 12 months prior to the week start date). This modest sample proportion might have reduced model power to detect subtle risk factor effects. Additionally, concussion history calculation limitations should be recognised. A concussion history was defined as having a diagnosed concussion in the previous 12 months, which was identified using surveillance data from the preceding season. Players might not have been in surveillance in the previous season (eg, academy graduates, overseas players joining the league) and as such their concussion history is unknown.

Policy implications

This longitudinal analysis of eight seasons describes the training-related concussion landscape and risk factors associated with injury in elite men’s rugby union in England. Concussion prevention strategies should primarily focus on matches, where the incidence of concussions is 80 times higher. However, training is a relatively controllable and modifiable environment compared with matches, potentially making concussion prevention strategies more feasible. Given preseason and full-contact training were associated with higher levels of training concussion risk, clubs should consider strategies, and governing bodies should consider policies, that encourage a graduated return to contact during preseason, delaying the introduction of full-contact training until after players have been exposed to training involving modified and controlled versions of contact. Given the inter-player variability in HAE accumulation²⁵ and baseline concussion risk in the present study, individual player management remains an important adjunct to population-based policy. This approach should still allow coaches and players sufficient opportunity to develop skill execution and contact proficiency while also reducing the risk of concussion during this period of higher risk.

Limitations

Assumptions made regarding training exposure are unlikely to be exact.⁹ Training exposure was analysed at the week-level, and so it was assumed that exposure was distributed evenly across each training week. In practice, however, training load is typically concentrated in the middle of the week before tapering toward match day.¹¹ This mismatch between true within-week load patterns and the assumption of uniform exposure may introduce exposure misclassification. Specifically, players who sustained a concussion may have accumulated a disproportionately high share of their weekly exposure immediately before injury. By spreading exposure evenly across the week, the analysis may underestimate the intensity or timing of the exposure that preceded concussion, reducing the precision of estimated injury rates and potentially obscuring true within-week risk patterns. It is difficult to validate contextual information associated with training concussions, such as the training activity and inciting injury event. However, concussion data were cross-checked with the SCRM app to ensure that all diagnosed concussions were recorded on Rugby Squad. Finally, while overall model diagnostics indicated adequate fit, the model observed underdispersion. Therefore, results might be more conservative, and fixed effects might have greater influence than the model suggests (online supplemental figure 1).

Conclusion

Overall, training-related concussion incidence was low compared with matches, and concussions in training are rare. There was a significant increase in the incidence of training-related concussions in the period 2021–2024 compared with earlier seasons. Training concussion incidence was ∼2 times higher in the preseason compared with in-season periods, and ∼4.5 times higher during full-contact training compared with controlled-contact. However, playing position, match minutes played in the previous 28 days, and concussion history did not predict training concussion risk, with total training exposure accounted for. Future work should aim to understand player-level training exposure to understand how and why particular players are more susceptible to concussions in training. While it is already very low compared with matches, practitioners might wish to consider strategic planning of full-contact and preseason training to mitigate concussion risk.

Data availability statement

All data relevant to the study are included in the article or uploaded as supplementary information.