In the modern era of sports, teams and teams sport clinicians have searched high and low for incremental improvements that affect individual or team performances to achieve optimal results. One variable which has come under the limelight in recent years has been predicting outcomes in sport and specifically, the ability to predict, and therefore prevent injury. (1)
In the context of team sports, clinicians should be mindful of how they hedge their bets as there is a large grey area when it comes to interpretation of risk factors and their association to injury, and the predictive accuracy of tests or evaluations often using during screening. In 2016 Professor Roald Bahr published “Why screening tests to predict injury do not work – and probably never will” (2) which created a stir in the sports medical journals prompting several, published response letters. The take home message in this critical review is that, to date, there is no known test that can predict occurrence of an injury no matter how strong the association is. In the same paper, Bahr draws parallels between injury screening and medical screening past and present with the famous examples of phenylketonuria and cancer. Professor Bahr also offers insight into the validation process of a screening tool in a three-step model.
Association vs. Prediction
McCall et al. (1), appropriately informs us that the misunderstanding of this concept, may explain much of the confusion regarding our ability to predict injury. This point is illustrated by the authors who performed database searches for ‘performance AND prediction’; ‘talent AND prediction’ and ‘injury AND prediction’ respectively – they found that only 23%, 10% and 35% of the respective studies found, reported using statistical modelling demonstrating predictive analysis, while the remaining used statistics that were linked to association. If there is a misunderstanding in the literature, we can understand where the grey area on the ground regarding clinicians stems from.
Association or risk factors are generally studied in large populations to ascertain if a marker is associated with an outcome. A misinterpretation comes in when the association is seen in a large population, then generalized to individuals. While there is an element of prediction in classification of athletes by clinicians when screening, we should be aware that their appropriateness both relies on the sensitivity and specificity of the screening tool as well as accuracy in the cut-off metrics to separate risk catagories. True-Positives (sensitivity), and False-Positives (specificity) can be used to aid our decision-making process (3). In the spectrum of Periodic health examinations for example, when screening for cardiovascular disease, the test should have a low number of False-Positives observed, to avoid flagging too many individuals, as the result of the screening would require specialist referrals and cardiovascular investigations. With regards to musculoskeletal screening on an individual basis, the same thinking applies – in the example of isokinetic testing of quadriceps and hamstring strength deficits in predicting hamstring strains in elite soccer by Van Dyk et al. (4) – the authors reveal that through a 4-year cohort study, they identify a statistically significant association between the deficit and hamstring strain, but when they apply more appropriate predictive statistical analysis, receiver operative curve shows that the predictive value of isokinetic testing in hamstring strains is no more accurate than a coin toss. They conclude that strength deficits measured by isokinetic testing to predict hamstring injury is unsupported.
No matter how strong the association, the association does not become predictive for the individual. Association generally describes the relationship of one factor to an outcome, while prediction shows the likelihood of the outcome occurring (sensitivity, specificity, positive and negative predictive values). Screening tests are usually measured on continuous scales, while predictive values are dichotomous – this gap can explain why even strong association, does not equivocate to prediction of outcomes. If we look at the hamstring example through a cost-benefit model, in terms of the validity of a test like Isokinetic dynamometry as a predictor of muscle strain (4) there is no indication that this test would be of use to prevent muscle strains if all members of the team were assessed. In this example, a validated blanket prevention program like the “Nordic Hamstring Protocol” may be more useful to the team due to being cost effective and a more constructive use of clinician’s time and effort.
Ethical considerations when screening and classifying athletes:
The ability to flag athletes into injury risk categories may be aimed at making medical management easier, but with the variation in association between certain measures and injuries, clinicians must be aware of accuracy of the screening tool as well as its sensitivity to cut off metrics before grouping or excluding athletes from selection, or participation due to injury risk. The financial implications for the professional athlete needs to be considered as well as the relationship between the sports medical staff and coaching staff when putting forward suggestions of exclusion or classification derived from screening. McCall et al. give the example of an injury flagged athlete being selected by the coach and then not sustaining injury, and the potential damage that it can cause in the athlete, coach and medical team relationship.
It is also important to remember the medico-legal principles. The athletic population is vulnerable to this predicament as their inclusion or exclusion is often in the hands of the medical and or coaching staff. Is it ethical to exclude an athlete that displays a valgus knee during landing due to ACL injury risk? Or should an individual with a limb symmetry index deficit on isokinetic evaluation not return to training until the deficit is corrected? If we are to advocate the evidence, no test to date is predictive of injury – if we shift to correct potential mechanisms of injury with these test findings, there may be ways to address these issues without the ethical dilemmas of exclusion or “crying wolf” regarding injury.
What should the team clinician do?
While understanding the difference between associative and prediction is one aspect of this misunderstanding, the team clinician may be left feeling that they’re damned if they do or don’t in this situation.
Screening the individual athlete does not predict injury – but it is useful if timed well, and subsequent risk factors are then addressed. Consistently, the strongest non-modifiable risk factor for new injury is previous injury. History, assessing fallout from previous injury and addressing these individually may be of use to the team clinician. During the post-season period, athletes can be grouped into prehab teams and address certain deficits together as a blanket intervention – provided that there is evidence that the blanket intervention has benefit for all at risk, as well as the individual with strong markers like previous history to the same area.
At this point in time, the likelihood of attaining accurate prediction models may seem nearly impossible, but what does this mean to the sports clinician in a team setting? It is important not to throw the baby out with the bathwater – so to speak. The aim of predicting injury is to ultimately prevent them or minimize their impact on the individual and the team – why not shift our focus towards injury prevention then? Van Mechelen et al proposed a 4 step implementation strategy for injury prevention, identifying burden of injury, recognizing mechanisms of injury, applying an intervention, then reassessing the injury epidemiology (5). Adding to that, Bahr has proposed the role of understanding the injury mechanism diagrammatically as the most important modifiable risk factor for mitigating injury. The benefits of low cost screening in the team setting is surely not the prediction of injury but allowing an early entry point for the athlete into the medical system for multiple interventions or monitoring strategies to occur. The International Olympic Committee consensus on Periodic Health Examinations relating to musculoskeletal screening echoes this sentiment commenting that the while the screening process builds rapport within the clinician athlete relationship, its main purpose is to detect current injury and should be timed at the end of the season as to address existing recent injury deficits (6).
1. McCall A, Fanchini M, Coutts A. Prediction: The Modern Day Sports Science/Medicine ‘Quest for the Holy Grail’2017. 1-11 p.
2. Bahr R. Why screening tests to predict injury do not work—and probably never will…: a critical review. British Journal of Sports Medicine. 2016;50(13):776-80.
3. Pepe MS, Janes H, Longton G, Leisenring W, Newcomb P. Limitations of the Odds Ratio in Gauging the Performance of a Diagnostic, Prognostic, or Screening Marker. American Journal of Epidemiology. 2004;159(9):882-90.
4. Dyk Nv, Bahr R, Whiteley R, Tol JL, Kumar BD, Hamilton B, et al. Hamstring and Quadriceps Isokinetic Strength Deficits Are Weak Risk Factors for Hamstring Strain Injuries. The American Journal of Sports Medicine. 2016;44(7):1789-95.
5. van Mechelen W, Hlobil H, Kemper H. Incidence, Severity, Aetiology and Prevention of Sports Injuries: A Review of Concepts1992. 82-99 p.
6. Ljungqvist A, J Jenoure P, Engebretsen L, Alonso J, Bahr R, F Clough A, et al. The International Olympic Committee (IOC) Consensus Statement on Periodic Health Evaluation of Elite Athletes, March 20092009. 347-65 p.
*Nick is the founder and physio at Enhanced Physio – this is a clinical commentary in relation to current literature and conversation around injury prediction. Nick is currently completing his thesis in: “Incidence of injuries and associations in elite field hockey, South Africa”.