The growing popularity of sports metrics and statistics

Over recent years, there has been a great evolution in the science of sports, and therefore also even more interest given to the level and rigour of the statistics or analysis approaches applied. The application of statistics to sports interests everyone from sports managers to decision-makers, trainers, players, coaches, journalists, bookmakers, scouts and video analysts, academics, fans, sports scientists, sports medicine experts, physicians, physiotherapists, psychologists, epidemiologists, researchers (in other fields) and statisticians (to design and develop new statistical analysis models).

Although scouting, sabermetrics and the book Moneyball had already previously given importance to statistics in this field, the Hollywood film of Moneyball was the trigger to help awaken the interest in analytics in sports science.5 In sports science, for example, the combination of statistics with the passion and knowledge of baseball has led to a new profession called sabermetricians. Similarly, the combination of strong statistical skills with the passion and knowledge of sports science and with economics and computer science would enable more sports data scientists to become like the stars of Moneyball (box 2).

Currently, most of the discussion about data in sports science fundamentally either concerns the performance analysis of players or teams, or is indirectly related to injury prevention in players. The first of the specialisations (known in the UK as a performance analyst, or as a sports analyst in the USA) is a great combination of skills such as statistics and sports science, and is growing exponentially thanks to the aforementioned new technologies. Many professional sporting teams have started to invest in analytics departments in recent years, and they all now employ statisticians.6 In his book, Alamar6 has already stated many opportunities for sports analytics, including that these fields are in continuous development.

Potential for Big Data in sports science and injury prevention

New technologies and the phenomenon of Big Data have given more value to data in recent years, especially in analytics (used often already in business). With so much information able to be obtained on and about athletes, the sports area is flooded with Big Data7 that opens up new analytical possibilities to investigate aspects of data resources such as variability, reliability and complexity. At this point, however, it is worth remembering that we do not only need more data, but we need the right data to answer the right questions.

This paper focuses on the increasing opportunities and roles that statisticians can make to sports injury research or what is now recognised as a specialisation in injury prevention, performance enhancement and injury surveillance.8 For example, the combination of statistics and epidemiology and public health or medicine and sports science gives a profession that is not very well known as Sports Biostatistician (box 2). It is opportune to take advantage of the current trending topic of new terms of professions within applied statistics, such as data scientist or smart scientist,9–15 to argue strongly for recognition also of the Sports Biostatistician professional. In the case of injury research, it is already common to find an injury data professional that is very similar to that of Sports Biostatistician, but focused solely on applying routine approaches (epidemiological or statistical) for better modelling to improve estimation of injury rates and risks. Even though there have been great advances in the profession of statistics and biostatistics in broader public health and medical research areas,16–19 this profession of Sports Biostatistics is rarely spoken about or promoted.

Need for training of new researchers/practitioners in this field

Currently, researchers who like statistics and sports science have the possibility of combining both through being trained via Massive Online Open Courses (MOOC) and other courses, but these are focused primarily on sports analytics20–26 with few opportunities to be exposed to injury data.27 To progress sports injury prevention, it would also be beneficial to incorporate Sports Biostatistics training specifically for injury data in the career development of future Sports Biostatisticians. There is no doubt that some new training programmes (eg, coursework, targeted workshops, etc) are needed. People coming to the field from a statistical or mathematical background can be trained in specialisations such as biostatistics through university-based Master's degrees in departments with strong affiliations to other specialisations such as bioinformatics, business and social statistics, operations research, etc. The development of new Master’s and doctoral programmes that include a formal internship with injury prevention research teams and other professionals could be a good start in the career development of future Sports Biostatisticians.

Consolidated departments of biostatistics, epidemiology and public health have traditionally been successful through their focus on data relating a specific disease or set of health conditions. It would also be of interest to do the same with Sports Biostatisticians and injury data, with such departments promoting and developing core teaching and research capacity in this area.

Attending conferences and other scientific meetings from different (but related) fields, such as sports science and medicine, epidemiology and biostatistics, can be important for updating one's skills and being exposed to new ideas, approaches and contacts. Some particularly relevant conferences are the following: International Olympic Committee (IOC) World Conference on Prevention of Injury and Illness in Sport (http://www.ioc-preventionconference.org/), the Safety World Conference (https://www.thl.fi/fi/web/injury-prevention/safety-2016), New England Symposium on Statistics in Sports (http://www.nessis.org), MIT Sloan Sports Analytics Conference (http://www.sloansportsconference.com/), MathSport International (http://www.mathsportinternational.com/), American College of Sports Medicine Conference (http://www.acsm.org/attend-a-meeting/annual-meeting), European College of Sport Science Congress (http://ecss-congress.eu/2016/16/), World Congress of Epidemiology (http://wce2017.umin.jp/welcome/index.html), International Society for Clinical Biostatistics (http://www.iscb.info/), International Biometrics Conference (http://www.biometricsociety.org/meetings-events/ibcs/). Apart from these conferences, there are also opportunities to attend statistics or biometrics conferences, but many of the presentations at these refer to clinical trials, genetics, infectious diseases or cancer. Very few are related to sports science and medicine, let alone injury data. Similarly, general epidemiology conferences most commonly are comprised of talks related to other health themes such as tobacco use, social inequality, diabetes, infectious diseases and even road traffic injuries, but not injury prevention in sports science. Unfortunately, this limits their ability to introduce their attendees to Sports Biostatistics as an important application area.

To train future Sports Biostatisticians, universities and relevant professional bodies could encourage and develop skills in the following. Primarily, the injury data specialist needs to know the scope of injuries and how they relate to sports science/medicine very well. Second, such professionals need to have a strong background in statistics, analytics, epidemiology and/or programming. Specific statistics/epidemiological software skills, such as R, SAS, STATA, Epi Info or similar, are indispensable, and some knowledge of data-management software such as Phyton, SQL, GitHub would be helpful too. Third, good communication skills are needed in order to work with sports medicine clinicians and sports scientists who do not have such well-developed quantitative and information technology (IT) skills, and may be wary of them. Finally, in any injury research team, the work of the Sports Biostatistician should start at the beginning of the study design process, well before data have been collected. It is not just the case that statistics needs to be properly applied to sports science or that sport scientists should apply statistical methods. Importantly, both are needed, and the Sports Biostatistician is the professional for both.

The Sports Biostatistician reporting gap

From the field of public health, it is a reality that the control and surveillance of sports-related injuries is a priority, even though there has been a data gap in recent years due to a lack of availability of high-quality information about the types of injuries sustained by these populations. There remains a number of unanswered reasons for this disconnect, but one could be the lack of records of success and impact in this field by Sports Biostatisticians, especially if people are not promoting this profession in their affiliations. It is very likely that while success exists, there is an overall difficulty in accessing literature about relevant data used, and data applications undertaken, by Sports Biostatistics. This is probably compounded by little interest or knowledge in the science of the topic or lack of political interest to date.

A notable example of success in this area has been Australian research conducted over the past 20+ years,28–35 which has advanced the design of injury surveillance systems, contributed to establishing and developing international efforts to improve the coding/classification of sports injuries, set standards for methodological rigour in the design and conduct of injury prevention evaluation and applied some recent statistical developments to the analysis of sports injury data. This programme of work has fostered the development, testing and application of new and/or enhanced approaches for injury data collection, classification and analysis that have since been used to track and assess specific sports injury risk factors. In the USA, other important injury statisticians such as Shrikant Bangidwala (he does not undertake sports injury research, but injury research more widely) and Steve Marshall (a recognised Sports Biostatistician) have also encouraged new and improved statistical analysis approaches in injury epidemiology and prevention.36–40

The need for more development in the methodology of sports injury prevention

There is no doubt that injury prevention now needs specialists who are Sports Biostatisticians to solve the significant challenges that arise with sports injury data. Recent papers have indirectly talked about the importance of this specialty to tackle the problem.41–46 In the sports science/medicine applications, it is important to use data appropriately to quantify injury occurrences, identify control strategies and prevent them. All of these require high-quality data presented with robust statistical approaches. In the face of multiple factors possibly associated with injuries, it is necessary to have control of confounding and multiple influences and knowledge by means of statistical tools. Better data that are generated and analysed by with emerging data technologies and more robust statistical methods, could influence the public health recognition of this sports injury problem, and especially the long-term economic costs they place on athletes, clubs and society.

As noted above, the growth of quantitative methods in the biomedical sciences has made biostatistics a key component in many research areas. It is worth considering why two major fields of medical research, namely public health and clinical trials, depend on statistics as a fundamental tool in the achievement of their goals. There may be lessons in this that could help further develop the Sports Biostatistics discipline.

In both sports and health sciences, advanced statistical models or techniques arise as an important methodology to predict outcomes (whether injury risk or performance success) and for assessing the associations between outcomes and risk/protective factors. Several applications of statistical modelling in injury epidemiology have been published, and it would be interesting to study their impact in terms of understanding and managing sport injuries.47–50

Epidemiological and statistical methods applied to injury data could certainly improve, change or innovate to address sports science and injury prevention. For example, regarding quality of data reporting and its analysis, a recent paper informs about the gap of the reporting and accuracy incidence estimates in multiple injuries within individuals which is much more limited.51

Second, rather than only examining the impact of a factor on injury outcome, injury prediction is one of the most challenging issues for injury prevention. The multifactorial complex and the dynamic nature of sports injuries highlight the importance of using different strategies of statistical modelling to reflect what happened. As several works point out,52–56 statistical approaches may be more accurate in predictions than linear methods or classical models such as logistic regression. There are some limitations of a non-dynamic model for predicting a sports injury. Meeuwisse et al55 highlighted the dynamic nature of risk factors, as the pattern of change in one variable could influence injury risk more than its absolute value in one point in time. Typical assumptions may be violated in some situations (non-linearity, dependence structure, etc), such as longitudinal studies, where there are repeated measures and, hence, correlated data. Ignoring correlation of data when fitting the model may lead to biased estimates and misinterpretation of results.50 For instance, injury data analysts should take into account situations when the outcomes are counts (small sample size, large overdispersion, small marginal mean or a high number of zero counts) to use other methods or alternatives (eg, negative binomial distribution, Poisson–lognormal distribution, quasi-Poisson distribution, generalised estimating equation (GEE), zero-inflated models, etc).48 ,50 ,57 When the outcome is time-to-event (survival), it is also important to take into account the existence of extensions in survival analysis (eg, joint modelling, competing risk, multistate models, recurrent model with time-dependent covariate, frailty models, etc) that take into consideration the dynamic nature of sports injuries.33 ,58–60 In contrast, the methods commonly used in sports analytics and data science include classical and Bayesian statistics, regression and classification, machine learning (ML), neural networks, data visualisation or time-series analysis.61 Nowadays, some of them are also used in injury prevention such as ML.62 Statistics (generally regarded as a ‘mathematical’ discipline) and ML (known as ‘engineering’ discipline) share the same goals—they focus on data modelling—but have philosophical differences. Statistics is more focused on statistical inference and identifying risk/associated factors from an aetiological point of view, whereas ML is more concerned with making predictions, taking into consideration higher order interactions even if the prediction cannot be explained very well. It would be interesting to start using these ML techniques more widely, despite some criticism about injury data being contaminated (eg, through error, selectivity or bias).63

Third, in recent years, the best scientific journals have alerted researchers to the need for increasing transparency and reproducibility in the publication of their scientific results. For example, some very recent commentaries by important statisticians and the American Statistical Association outline, in detail, the correct use of p value and transparency.64 ,65 Many of these important discussions may not be on the radar of sports science/medicine researchers who do not read widely outside of their discipline, and Sports Biostatisticians have the potential to link the two together. Moreover, Sports Biostatisticians have an important role in efficiently testing investigational hypotheses by avoiding biases and accounting for all the sources of variability present in injury data sets. This usually leads to complex study designs, and Sports Biostatisticians are needed to contribute their skills here.

Communicating complex sports injury study findings from studies to the public

From a disciplinary perspective, the success of a statistician is often measured by their ability to innovate as well as how well they adopt new and original ways with other statistical procedures. However, Sports Biostatisticians must also make an effort to explain and translate these essentially mathematical ideas in such a way that the sports community (coaches, players, sports medicine clinicians, trainers, physiotherapists, sports scientists, other epidemiologists and decision-makers in the clubs) can intuitively follow them and understand their application in broader efforts to prevent and predict injuries. Just as important as mastery of the technical tools is the essential skill of Sports Biostatisticians also being able to effectively communicate their analysis approaches and findings to non-technical audiences.