Article Text
PDF
Abstract
Background Internal workload (ie, from training and matches) is considered one of the most important injury risk factors for elite European football teams, however there is little published evidence to support this belief.
Objective We examined the association and predictive power of internal workload and non-contact injuries.
Methods Five elite European teams, 171 players (age: 25.1±4.9 years; height: 181.6±6.7 cm; body mass: 77.5±7.2 kg) participated over one full competitive season. Using the session-rating of perceived exertion (s-RPE) method player’s internal workloads were calculated for acute week, week-to-week changes, cumulated weeks, chronic weeks and acute:chronic ratios and analysed for association with non-contact injury (using generalised estimating equations (GEE)). Associated variables from GEE analysis were categorised into very low to very high workload zones and checked for increased relative risks (RRs). Associated workload variables were also analysed for predictive power (receiver operating characteristics).
Results Acute:chronic workload ratios at 1:3 and 1:4 weeks were associated with non-contact injury (P<0.05). Specifically, a greater risk of injury was found for players with an acute:chronic workload at 1:4 weeks of 0.97 to 1.38 (RR 1.68; 95% CI 1.02 to 2.78, likely harmful) and >1.38 (RR 2.13; 95% CI 1.21 to 3.77, very likely harmful) compared with players whose acute:chronic workload was 0.60 to 0.97. An acute:chronic workload 1:3 of >1.42 compared with 0.59 to 0.97 displayed a 1.94 times higher risk of injury (RR 1.90; 95% CI 1.08 to 3.36, very likely harmful). Importantly, acute:chronic workload at both 1:4 and 1:3 showed poor predictive power (area under the curve 0.53 to 0.58) despite previous reports and beliefs that it can predict injury.
Conclusions This study provides evidence for the acute:chronic internal workload (measured using s-RPE) as a risk factor for non-contact injury in elite European footballers. However the acute:chronic workload, in isolation, should not be used to predict non-contact injury.
- injury
- soccer
- training load
- prevention
- risk factor
Statistics from Altmetric.com
Introduction
The UEFA Elite Club Injury Study (ECIS) aims to answer problems that will impact elite teams’ daily practice by providing evidence-based recommendations. To do so, the first step is to ask practitioners what is relevant to them and how sports science/medicine research can help. According to ECIS teams,1 internal training and match workloads are one of the most important risk factors and they desire scientific investigations into these to provide practical recommendations.
The majority of injury workload research to date, deals with football codes such as rugby league2–4 and Australian Rules Football.5–8 The International Olympic Committee’s 2016 consensus statement9 recommends that large absolute workloads (eg, acute weekly workloads, week-to-week changes, cumulated weeks) and high relative acute:chronic workloads (in particular >1.5) increase the risk of injury, while high chronic workloads (eg, 2, 3 and 4 weeks) may offer protection. Although sparse, there is an increased focus investigating workload and injury in elite football (soccer).
Studies so far in elite football have identified potential links of internal workload with injury. Using the session-rating of perceived exertion (s-RPE: duration* perceived intensity) method of quantifying internal workload,10 Malone et al 11 found that professional players exerting acute 1-weekly workloads ≥1500 arbitrary units (AU) to ≤2120 AU)were at higher risk of injury than those <1500 AU (OR 1.95, 95% CI 0.98 to 3.95). Additionally, 2-weekly and 3-weekly cumulated workloads ≥5980 AU and ≥9154 AU also showed significantly higher injury risk during preseason (OR 4.74, 95% CI 2.74 to 5.66 and OR 5.11, 95% CI 4.26 to 5.14, respectively). When considering week-to-week changes of ≤350 AU to ≥500 AU versus ≤200 AU an increased risk of injury was found (OR 1.66, 95% CI 1.30 to 2.21) while s-RPE acute:chronic workload, between >1.00 AU to <1.25 AU significantly reduced injury risk (OR 0.68, 95% CI 0.08 to 1.66). Finally, in another study of professional players12 it was shown that, compared with seasonal averages (using a repeated measures one-way analysis of variance to determine differences), s-RPE weekly workloads were significantly higher in all of the 3 weeks leading to injury. However, in that same study,12 there were no excessively inflated s-RPE acute:chronic workloads prior to injury.
An important aspect and current hot topic is the debate on the ability of variables/risk factors to predict injury. It has previously been suggested that workload predicts injury.13 However, this was not actually statistically analysed in the aforementioned study. While the ability to predict injury is attractive for football science and medicine practitioners, due to the multifactorial nature of injury it is unlikely that one variable alone will have sufficient predictive power to accurately identify players that will definitely go on to incur an injury.14 Indeed, a study on elite-level football (soccer) players, by Fanchini et al 15 has provided the first answer to this question. It was revealed that while associated with non-contact injury, those workload variables (including the acute:chronic workloads at 1:2, 1:3 and 1:4) all showed poor predictive power for non-contact injury, as assessed by a low area under the curve (AUC) between 0.54 to 0.60. To put this into perspective, an AUC of 0.5 is equal to flipping a coin.14 It is therefore important to correctly distinguish between a variable that is ‘associated’ with and one that is ‘predictive’ of an injury. Identifying a predictive workload variable would allow practitioners to identify individuals that will go on to incur an injury, while association allows practitioners to identify a player with an increased risk and justifies implementing risk mitigation strategies.16
Despite a growing body of literature in elite football, results are not conclusive and while the aforementioned studies are an important first step in understanding workload and injury in football, there are several methodological limitations including; the definition of injury, type of injury and statistical power. Large-scale, multiteam, multicentre studies have been highlighted as an integral next step in further understanding injury risk and enhancing generalisability of results and practical recommendations.17–19
The aims of the present study were to (1) investigate the association of workload variables commonly studied in research and used in practice with non-contact injury and (2) analyse the predictive power of any workload variables found to be significantly associated with non-contact injury. We hypothesised that (1) acute week, week-to-week changes, cumulated and acute:chronic workloads would significantly increase non-contact injury risk while high chronic workloads would be protective. (2) Workload variables used in isolation would have poor predictive power.
Methods
Participants
During one competitive season (2015/2016), six elite teams belonging to the UEFA ECIS Study participated. The UEFA ECIS Study teams represent teams competing in the highest level of European competition; the UEFA Champions League. At the end of each season UEFA holds a postseason team meeting for the 36 Champions League team medical doctors. At the 2014/2015 postseason meeting, the rationale and initial outline of the present study protocol was proposed and teams were asked to express their interest to participate. Six teams registered interest and agreed to participate in this pilot study of one season. One of the six teams was excluded due to inconsistent match workload data being recorded. All teams were prospectively followed (age: 25.1±4.9 years; height: 181.6±6.7 cm; body mass: 77.5±7.2 kg). Each season was split into two distinct periods: preseason (early July to mid-end August depending on the team schedule and league start date), and inseason (mid-end August to end May). Preseason period lasted 40.4±5.4 days, the inseason (ie, competitive) period was 276.0±6.7 days, respectively. The study was approved by the UEFA Football Department Division and the UEFA Medical Committee.
Quantification of internal workload
The internal workload was assessed using a rating of perceived exertion (RPE). Players were instructed to rate the global intensity of all sessions and matches using the category ratio scale developed by Borg20 by answering the question: ‘How was your workout?’. The RPE was collected 30 min after completion of the session/match by the same science/medicine staff member. Workload was calculated using the method proposed by Foster,10 whereby the intensity rating of the session is multiplied by the duration for each player for each training session or match. Total training workloads were calculated from all training sessions (ie, field sessions, gym sessions and recovery sessions) and matches (friendly and competitive matches). The weekly workload corresponded to the sum of the workload for all training sessions and matches for each week. The absolute week-to-week change in workload was calculated as the difference between current weekly workload and previous weekly workload. Cumulative workloads were calculated for 2 weeks, 3 weeks and 4 weeks. Chronic workloads were calculated as rolling averages for 2 weeks, 3 weeks and 4 weeks. An acute (ie, weekly workload) to chronic (ie, rolling average of previous weeks as described previously by Hulin et al) 21 ratio was calculated for 2 weeks, 3 weeks and 4 chronic weeks (acute:chronic_1:2, acute:chronic_1:3, acute:chronic_1:4, respectively). As the five teams consisted of international-level players, they would be called up for national team duty throughout the season. During these periods, teams were asked to collect the quantity of training sessions and matches in addition to RPE and duration of sessions/matches. If this was not possible, the missing data set was averaged based on player’s seasonal match or training values throughout the season.
Injury data collection
An injury was considered an injury when a player was unable to fully participate in future football training or match due to physical complaints.22 23 Illnesses, disease and mental complaints were not considered as physical complaints, but were taken into account to calculate match and training exposure. Training missed due illness or for psychological reasons was not considered as training missed due to physical injuries. However, any workloads prescribed to players during periods of absence due to illness or psychological reasons were included in the calculation of overall training exposure. Since non-contact-injury is the most common injury in elite football and the workload will mainly influence this type of injury, only information pertaining to non-contact injuries was collected.5 7 An injured player was considered injured until he was cleared by the club doctor for participation in full training or matches.
Standardised protocols among teams
Detailed instruction forms were sent to teams in order to standardise the data collection. Players were excluded based on the advice of the main contact person in each team, for example, if it was deemed that the data were untrustworthy or inconsistently reported by the player/s. In instances of missing data, for example, during international breaks, the player’s season average for training or match loads were used. This was decided by a round table discussion that included the current authors and two other applied practitioners/researchers. There is currently no best practice for dealing with missing data, and averages were chosen as they are commonly used in practice by teams and thus reflect the real world setting. Future research should explore how to best deal with missing data.
Statistical analysis
Variance inflation factors (VIF) were determined to detect multicollinearity between markers; if a VIF ≥10 was found the variable was excluded from the analysis.24
Determining association: generalised estimating equations
A generalised estimating equations (GEE) analysis GEE (SPSS, V.21. IBM Company, New York, USA) was used to determine the association between workload measures and non-contact injury in the subsequent week.25 GEE analysis was chosen due to its ability take into account the correlated nature of repeated measures data, making it superior to traditional logistic regression methods.25 Specifically, a Poisson loglinear (link function) model was used, with a robust estimator with an autoregressive working correlations matrix (based on quasi-likelihood under the independence model criterion ie, the lower quasi-likelihood under the independence model criterion (QIC) value.26 If GEE analysis was statistically significant (P<0.05) the exponential (Exp(B)), as well as the 95% CI were presented. The exponential coefficient corresponds to the multiplicative term to be used to estimate injuries when the workload variable increases by one unit and represents the relative risk (RR). Significant variables were then split into four groups based on the 15th, 50th and 85th centiles to compare injury risk between zones of different workloads: extremely low (<15th centile), moderately low (>15th to 50th centiles), moderately high (form >50th to 85th centiles) and extremely high (>85th centile).15
For comparison between the risk of non-contact injury in different workload zones, RR and 95% CIs were calculated (MedCalc Software, Ostend, Belgium). Magnitude-based inferences were used to interpret change in risk between zones.27 The smallest beneficial and harmful effect for a risk ratio was considered as <0.90 and >1.11, respectively. The effect was considered unclear if the chance of the true values was beneficial and was >25% with an RR <66. If the effect was considered clear, thresholds for assigning qualitative terms of beneficial, trivial, harmful were as follows: <0.5%, most likely; 0.5%–5%, very unlikely; 5%–25%, unlikely; 25%–75%, possible; 75%–95%, likely; 95%–99,5% very likely; >99.5% most likely.28
Determining predictive power—receiver operating characteristics curves and diagnostic accuracy testing
Workload variables that exhibited a significant association following GEE analysis, were tested for predictive power using receiver operating characteristics (ROC) curves. The ROC curve examines the discriminant ability of a marker used to classify players in two groups and plots the true positive rate (sensitivity) against the true negative rate (specificity) to produce AUC. Following the analysis method by Crowcroft et al,24 an AUC of 1.00 (100%) represents perfect discriminant power, where 0.50 (50%) would represent no discriminatory power. An AUC >0.70 and the lower CI >0.50 was classified as a ‘good’ benchmark.29 All ROC curve results were presented as AUC ±95% CI. All ROC analysis was performed using SPSS V.21 (IBM Company, New York, USA).
To examine the predictive power with diagnostic accuracy assessment, the workload zones (<15th, 15–50th, 50–85th, >85th centiles) were assessed for their sensitivity, specificity and positive predictive values (MedCalc Software, Ostend, Belgium). In addition, the Youden Index was calculated (Youden index=sensitivity+specificity−1) from all ROC curve plots to determine the point where the sensitivity and specificity were optimised (ie, high Youden index) and considered the score at which a ’cut-off’ value from each workload marker might be acceptable to discriminate a player at risk of injury. The maximum Youden Index of 1 would suggest perfect discriminatory ability, while a score of 0 would reflect no diagnostic value.30
Results
Internal workload and non-contact injury
A total of 171 players from five teams were included in the analysis. Twenty-nine players were excluded due to inconsistent data. One hundred and twenty-three non-contact time-loss injuries were recorded and included. Cumulated 2, 3 and 4 weeks' in addition to chronic 4 weeks' workloads showed substantial multicollinearity (ie, VIF >10) and therefore were excluded from analysis. One team collected all data during international duty. Of the other four teams, 1.7% of the overall data were averaged (team 2%–3.3%, team 3%–1%, team 4%–1.5% and team 5%–1.7%).
Association between internal workload and non-contact injury
The results of the GEE analysis are presented in table 1. Acute:chronic_1:3 and acute:chronic_1:4 showed significant associations with non-contact injuries in the subsequent week (all p<0.05). No significant associations (p>0.05) for acute weekly, week-to-week changes, acute:chronic_1:2, or chronic 2-weekly or 3-weekly workloads were found.
Association and predictive power of acute:chronic workload markers with non-contact injury in elite European Football players
The RR and 95% CI calculated for comparison between risks in different workload zones and corresponding magnitude-based inferences are presented in table 2. Instances where the 95% CI of the RR did not cross 1 have been specifically reported and discussed. Using this criteria, players exhibiting an acute:chronic_1:4 of 0.97 to 1.38 and >1.38 compared with 0.60 to 0.97 showed an increased risk of non-contact injury (RR 1.68; 95% CI 1.02 to 2.78, likely harmful and RR 2.13; 95% CI 1.20 to 3.77, very likely harmful, respectively). Additionally, an acute:chronic_1:3 of >1.42 compared with 0.59 to 0.97 displayed a 1.94 times higher risk of injury (RR 1.90; 95% CI 1.08 to 3.36, very likely harmful). In all other instances, the 95% CIs for remaining RR comparisons all crossed 1 (see table 2 for all workload zone comparisons).
Injury risk comparisons between different zones of workload (<15th, 15–50th, 50–85th, >85th centiles)
Predictive power of perceived workload variables and non-contact injury
The values for AUC (95% CI) and the Youden Index for acute:chronic_1:3 and acute:chronic_1:4 are presented in table 1. Although significantly associated in the GEE analysis, they showed poor predictive power for non-contact injury (AUC 0.56 to 0.58). Sensitivity, specificity and positive predictive values for all acute:chronic_1:3 and acute:chronic_1:4 workload zones are shown in table 3.
Diagnostic accuracy assessment for acute:chronic 1:3 and 1:4 for each ratio zone (<15th, 15–50th, 50–85th, >85th centiles)
Discussion
Using the largest sample size to date in any football code, we aimed to identify whether workload was a risk factor for injury in elite footballers and whether workload variables predicted non-contact injuries. The results demonstrated that only acute:chronic_1:3 and acute:chronic_1:4 were associated with non-contact injury occurrence. However, neither of the acute:chronic workloads had any predictive power to identify players who went on to incur a non-contact injury.
s-RPE and non-contact injury risk in elite football
Acute:chronic workload
Acute:chronic workload, that is, changes relative to what a player has been prepared for is suggested as the most pertinent measure of workload to identify injury risk in athletes.9 While this workload measure is well established in other football codes10 it is less clear in football (soccer). Malone et al 11 found an acute:chronic workload of >1.00 AU and <1.25 AU to be protective using the traditional 1:4 weeks, however CIs crossed 1 and therefore the results are not clear. Lu et al 12 found no excessively inflated acute:chronic workloads prior to injury. The traditional acute:chronic workload at 1:4 may not apply to elite football due to congested fixture scheduling.31 Therefore, the current study investigated three acute:chronic ratios (1:2, 1:3 and 1:4 weeks) and found that the traditional acute:chronic_1:4 between 0.97 to 1.38 and >1.38 increased the risk of non-conact injury (RR=1.68 and 2.13, respectively, when compared with reference acute:chronic workload of 0.60 to 0.97). Additionally, acute:chronic_1:3 >1.42 also showed an increased risk of non-contact injury (RR=1.90) compared with the reference acute:chronic workload of 0.59 to 0.97. No significant association was found for acute:chronic_1:2 in the GEE analysis. Interestingly, Stares et al 32 also investigated varying acute:chronic workloads and while they found no improvement of various acute:chronic time frames over and above the traditional 1:4 weeks, they recommended that according to the practical needs of the particular sport or team, practitioners can choose the ratio best suited to them. The present results support the use of acute:chronic_1:3 and 1:4 in the elite football setting depending on whichever is deemed most suitable to that team’s practical setting.
Acute week and week-to-week changes
The current study revealed no association between acute weekly workloads or week-to-week changes with non-contact injury. This is in contrast to another study of two elite-level European football teams11 where injury risk was increased in players exerting a 1-weekly acute workload ≥1500 AU to ≤2120 AU (OR 1.95; 95% CI 0.98 to 3.95) and large week-to-week changes (>350 AU to 550 AU) (OR 1.66, 95% CI 1.30 to 2.21). However, it should be noted that all injuries (contact and non-contact) were taken into account in the Malone study.11 In the present study, only non-contact injuries were included in the analysis, with contact injuries excluded, which may explain the difference in results. While there is a contention that contact injuries may also be affected by workload and ensuing fatigue, the current authors argue that while this may be so, in order to include contact injuries in the analysis, each injury case should be reviewed carefully and closely to determine if it is plausible that workload could have been implicated in that injury. For example, a contact injury could be recorded as being stood on by an opponent’s boot, or a player clashing into the back of another without any plausible way that could have been affected by the workload of the player incurring injuries. Another issue relating to injury definition that should be addressed in future studies is the lack of recording for ongoing management of overuse/chronic injuries.
Chronic workloads
No association or increased/decreased injury risk was found for any chronic workload. This is in contrast to the current literature, that is, high chronic workloads may be protective of injury while low chronic workloads could increase the risk. This could be explained by the fact that the current study analysed chronic workloads in isolation. It has in fact, been shown in elite football33 that, 3-week chronic workloads ≥2584 AU were protective, but when players also covered a 1-weekly high-speed running (HSR) of 701 m to 750 m. The same 1-weekly HSR distance with low chronic workload (<2584 AU) increased injury risk. This highlights the importance of future studies and practitioners to analyse chronic workloads in conjunction with other parameters such as external running workloads. Additionally, although not performed in the present study, it may be important to consider chronic workloads alongside the acute:chronic workload as it has been shown that acute:chronic workload becomes a higher risk factor in either low or very high chronic workload conditions.32
s-RPE as a tool to monitor workload
Session-RPE is particularly useful to monitor internal workload of footballers, with several advantages. First, there is no cost, no hardware is needed, it is easy to implement and is valid and reliable.34 Additionally, s-RPE can be used to collect workload from players during all types of training sessions (field, gym, rehabilitation, recovery) both indoor and outdoor. A useful advantage, particularly in elite football, where players are often required to join their national teams for multiple, prolonged periods, is that s-RPE can still be collected, minimising missing data points.
Despite these advantages, as with all monitoring tools, there are some potential limitations to s-RPE; it cannot differentiate between short high-intensity and long low-intensity sessions.9 The example given in the workload consensus statement9 is a 30 min session with an RPE of 8 and a 120 min session with an RPE session of 2 will yield an s-RPE of 240. While the authors acknowledge this limitation, the example given in the consensus statement is somewhat unrealistic; however the concept remains. It is for this reason that consideration of the context and the art of the practitioner (practical experience) is crucial to interpret s-RPE values. It is among these reasons why monitoring both internal (eg, s-RPE) and external (eg, Global Positioning Satellite Systems GPS) workloads and considering within the context of the player and situation is likely best practice when implementing an effective workload monitoring programme. Another limitation could be that athletes/players may potentially misuse the technique by providing a false perception of effort in order to influence subsequent training sessions.35 However, in the current study, the study protocol set out in advance to staff and players that the data collected should not influence the decision-making process. Additionally, any instances where a players’ data were deemed untrustworthy it was excluded from the analysis.
What is the predictive power of acute:chronic workload and non-contact injury?
A popular debate in the sport science/medicine discipline is the prediction of injury using screening/monitoring. A study of a single elite football team15 has revealed poor predictive power of workload variables including acute:chronic 1:3 and 1:4 weeks with non-contact injury. The current study supports the results of the study by Fanchini et al.15 The predictive power, sensitivity/specificity and positive predictive value of acute:chronic 1:3 and 1:4 in this study are shown in tables 1 and 3, respectively. Both acute:chronic workloads, independently, showed poor predictive power (all AUCs <0.60, Youden Indices close to 0, poor sensitivity, ie, ability to detect when a non-contact injury will occur and low positive predictive values). It should be noted that multivariate approaches to the analysis of workload measures have been shown to improve predictive power36 and should be considered in future workload research in elite football. The current authors have deliberately kept this section of the discussion brief, as the aim of injury risk mitigation is not necessarily about predicting injury, but rather understanding and establishing risk factors to target for risk mitigation strategies.37
Limitations
In the present study, the statistical power was not calculated prospectively. As it is not appropriate to calculate statistical power retrospectively38 no power analysis has been included. To the authors' knowledge, the present study is the first study of its type in football to include >100 time-loss, non-contact only injury cases in the analysis. As recommended by Bahr and Holme,39 20 to 50 injury cases are needed to detect moderate-to-strong associations in a risk factor study and small-to-moderate associations would need about 200 injured participants. Nevertheless, we cannot be sure of the statistical power in the present study and future studies should be required to calculate this prospectively.
Conclusion
With the largest number of non-contact injuries analysed in an elite football cohort to date, this study provides additional evidence for practitioners to consider acute:chronic workload as one risk factor for non-contact injury in their players. Specifically, elevated acute:chronic workloads of 1:3 and 1:4 weeks using the s-RPE method were identified. Practitioners should be cautious, however, if using any single workload variable in isolation to predict if a player will incur a non-contact injury.
What are the findings?
Internal acute:chronic workloads (using the session-rating of perceived exertion (RPE) method) of 1:3 and 1:4 weeks are associated with non-contact injury in elite football players.
While internal acute:chronic workloads were associated with non-contact injury occurrence, these markers showed poor predictive ability in isolation to identify individual players who will actually go on to incur a non-contact injury.
How might it impact on clinical practice in the future?
Our findings support the use of a simple, free and validated method of monitoring internal workload (session-RPE) in the practical elite football setting in order to monitor injury risk in players.
Differentiating correctly between association and predictive ability will allow practitioners to know the difference and improve their communication to key stakeholders on players’ injury risk.
Acknowledgments
The authors thank the teams involved for their important contribution to achieving this unique data set. In particular (in alphabetical order of team); Darcy Norman (AS Roma), Dr Mikel Aramberri (CF Real Madrid), Dr Ricard Pruna and Eduard Pons (FC Barcelona), Roberto Sassi and Antonio Gualtieri (FC Juventus), Dr Nelson Puga (FC Porto) and Barthelemy Delecroix (Olympique Lille Sporting Club). The authors also thank the players for their participation. The authors also thank Dr Maurizio Fanchini for the invaluable discussions on methods of statistical analysis.
References
Footnotes
Contributors All three authors came up with the idea for the study. JE recruited the teams. AM and GD collected the data and performed the statistical analysis. All three authors contributed to interpretation of the results and writing of the manuscript, and to the revision and second version of the manuscript.
Funding This study was supported by grants from UEFA and the Swedish National Centre for Research in Sports.
Competing interests None declared.
Patient consent Detail has been removed from this case description/these case descriptions to ensure anonymity. The editors and reviewers have seen the detailed information available and are satisfied that the information backs up the case the authors are making.
Ethics approval UEFA Football Department Division and the UEFA Medical Committee.
Provenance and peer review Not commissioned; externally peer reviewed.