Reporting Multiple Individual Injuries in Studies of Team Ball Sports: A Systematic Review of Current Practice

A search of the PubMed, Web of Science and Embase databases was performed, on the basis that the leading sports medicine journals are indexed within these databases. The full search strategy is described in electronic supplementary material Appendix S1. Studies that reported injury incidence in team ball sports for more than 1 calendar year or at least two consecutive seasons were included. As not all studies include keywords that consider duration of the study (i.e. we sought more than one season), the search strategy was initially kept broad and a relatively large number of publications screened to allow a specific check of inclusion dates in the full text. All articles were screened by two authors independently, using the criteria described below. Differences were discussed and when no consensus was obtained upon study inclusion, a third author adjudicated. No date restriction was placed for the beginning of the search, with studies included up to an end date of 31 December 2015. Where we could not locate an original paper from our resources, authors were directly contacted.

Articles were included if (1) they reported injury incidence (i.e. not specific to one injury type) over the time period (e.g. not game injuries only) because we were interested in multiple injuries, studies focusing on single injury types or those sustained in games only were deemed overly restrictive; (2) the study was conducted with participants of team ball sports; (3) data were collected prospectively; (4) the follow-up was more than 1 calendar year or over at least two consecutive seasons; (5) data were collected in a defined cohort (e.g. club, division, school, team, league); and (6) an individual identifier was clear in the collection of data (i.e. an ID or name that would allow the research team to potentially link more than one injury within individuals). The latter criterion was deemed crucial if the original authors had been able to report multiple injuries within an individual.

Non-English language articles, conference abstracts, commentaries and reviews were excluded. Studies based on hospital data were excluded because generally people do not always attend hospital with a sports injury, nor do they exclusively attend the same hospital, therefore such data cannot be used to confidently identify multiple injuries within individuals. Studies conducted during (multiple or consecutive) tournament-style competitions were also excluded as there are often long time frames between such competitions and athletes are likely to participate in other competitions/events over the period during which their injury risk exposures and outcomes would be unaccounted. Where more than one publication was presented from the same source study/dataset, we initially included all publications identified (that met all inclusion criteria) as different approaches may have been used to report/analyse data across papers; if the same data and analysis were reported (albeit for a different aim), only the earliest study published was kept and is presented with a note indicating subsequent publications. The exact number of publications excluded by different reasons is not provided as they were excluded on the first identified reason only (therein potentially presenting an inaccurate picture, as additional reasons, other than the first identified, could also apply).

The following data were extracted from the articles, by two authors, into an Excel database. Relevant results were tabulated and presented, along with their descriptive information across five tables.

Descriptive information provided in Table 1 included:

From publications that presented pooled (grouped) injury results, the number of injuries and number of injured participants were summarised, demonstrating that there was more than one injury occurrence by individuals in these studies (Table 2). Where injury data were presented on a per-person basis (i.e. not only presenting pooled injury results for an entire group), additional information was extracted on how individual injury data were presented, the number of injuries, and the number of individuals injured (Tables 3, 4, 5).

The most common presentation of data was pooled injury counts summarised across teams/seasons. This is problematic if there is no indication of the number of athletes sustaining more than one injury, particularly in studies aiming to infer injury risk. Although many of the included studies did not have a primary aim to look at more than one injury in individuals, the issue still needs to be considered. For example, a common study aim might be to follow specific teams/squads to determine changes in team outcomes based on new training practices. Similarly, school-level sport competitions may be interested in how overall safety levels change over time, not individual risks. However, when reporting injury incidence for an understanding of injury risk, the individual distribution of the number of injuries is important because of its impact on the correct statistical reporting of incidence rates [17]. As an example, one study reported 134 injuries in 268 athletes over 3 years “yielding a risk of injury of 50 %” [38] (p148). Such a statement is only true if there were no athletes with more than one injury within the study population. As the authors did not report the distribution of injuries by participants, it is not possible to verify this but, in general, findings reported in this manner will miss details that are vital for interpreting results. In its most simplistic form, this information is depicted in Fig. 2, reporting scenario A. It can be seen that the incidence outcomes within this scenario (10 athletes included, with 20 injuries in total) are similar for both seasons, although the actual situation is very different (refer to the numbers in black circles).

Where the potential for an individual having more than one injury in the results was recognised, this was sometimes addressed by reporting the range of injuries sustained or the maximum number of injuries sustained by an individual, e.g. ‘between one and n injuries occurred’. Other authors acknowledged the problem of multiple injuries by reporting the mean and standard deviation for injuries per athlete/team per season (e.g. “On average, an athlete incurred 3.6 ± 0.7 injuries” [39]). Figure 2, reporting scenario B, shows that these approaches still fail to adequately address how an individual with more than one injury impacts the group result as it can again be seen that incidence outcomes using this scenario are similar for both seasons (range 0–9 injuries; maximum: 9; mean: 2 injuries per athlete per year), although the actual situation is very different (refer again to the numbers in black circles).

It was clear that many cohort participants within the reviewed studies had likely experienced more than one injury (i.e. where the number of injuries recorded exceeded the number of injured participants). This is common in studies of sports injury. Similarly, the number of injured athletes is often less than the number of athletes included, as can be seen in Table 2. In other words, injury data in team sports are commonly skewed, with a large number of zero counts, and this distribution needs to be considered [17]. One, or just a few athletes, might sustain the majority of team injuries [15, 28], as shown in Fig. 2, scenario C. For any team, at any time point, the proportion of athletes contributing to the injury count will differ. Therefore, even those studies reporting team-level data and overall changes in annual injury counts (and not necessarily concerned by any individual outcomes) still need to consider individual bias in the results.

Irrespective of whether a simple or complex study design and analysis method is chosen to meet a study’s aim, authors must clearly identify the relevant details of included variable characteristics as they relate to multiple injuries. As a minimum, the frequency distribution of the number of injuries is recommended for inclusion in future work (see example in Finch and Cook [15]). This information would at least enable readers to gain a sense of the individual burden and scope for potential dependency between multiple injuries sustained, as well as providing crucial support for the choice of statistical modelling applied (e.g. if there is overdispersion of data or a high number of zero counts) [17, 40]. This more detailed presentation of the data is similar to recommendations of analyses where time to injury data is considered, with improved reporting of assumptions and detailing of the event being modelled [7, 41].

A key finding was that the majority of studies presented broad annual injury rates across teams/seasons. Few gave specific reasoning as to why the results were presented in this way, with the exception of an aim being to compare year-to-year outcomes. One study team reported that results in two of their publications were both pooled as there was no statistically significant change over time [29, 37]. Reasons why pooled data may be favoured by researchers could include the avoidance of difficulties with ethical approval if data are reported at a group level not an individual level; a need for confidentiality, as pooling the data hides small or distinctive values; the analysis is less complicated if data are grouped; pooling data will give larger numbers, allowing for more sophisticated analyses; or to avoid the difficulties in differentiating between index/recurrent/subsequent injuries. While the results across studies were presented largely as pooled data at the team level, the collection of data itself was at an individual level, with some form of identifier in the data collection process a requirement of our study inclusion process. Based on these studies, it would seem that it is easier to record, rather than adequately report, more than one injury within study cohort members.

The overall number of publications extending beyond one season/year has increased over the last 2–3 years. This possibly reflects a growing understanding of the need to look not only beyond one season when considering injury risk but also new validated technology to facilitate long-term data collection. For example, recent studies document the value of online applications and SMS text messaging in enabling a relative ease of access and follow-up of participants [42, 43]. Online and automated forms of data collection will be further enhanced in coming years if analytical methods and reporting algorithms that make the most of individual-level data also become widely used.

A limitation of the current understanding about multiple injuries in the reviewed studies is the focus being largely limited to recurrent injuries—injuries that are exactly the same site and same type as previously incurred by an individual. Clinical experience and new classification models substantiate the need to look at the relationships between injuries more broadly [15]. Another barrier to reporting multiple injuries is how to document and measure recovery. Recovery is a key element in determining whether a subsequent injury is new or potentially related to a previous occurrence. Unlike other injury contexts, the risk of more than one injury in the sports setting is high and there can be a short time lag between the injury events. Within this review, we have not identified and reported how recovery was defined or addressed, although the importance of this topic and its relation to reporting multiple injuries is acknowledged. We initially attempted to include the information but found the message confused our primary aim of reporting methods of multiple injuries. What was clear in terms of recovery was that authors mostly used a return-to-sport definition, with or without clearance from a medical specialist and at differing levels (i.e. training, partial return to competition, or full return to competition). Operationalising ‘return to sport’ as an outcome measure is challenging [44, 45] and, indeed, there is an entire consensus statement developed for this specific issue [46]. With the exception of the date of return, recovery from injury as a clearly defined outcome currently lacks universal objective measures and is influenced by a range of potential factors, including access to medical care and previously sustained injuries [44‐46]. A second challenge in defining recovery arises with the recent expansion of more inclusive injury definitions that are no longer consistent with earlier recovery definitions. For example, if a holistic approach is used in a definition, such as ‘recordable incidents inclusive of psychological factors’[47], then it is not clear how concepts of readiness, fear and confidence contribute to recovery and return-to-play definitions. The ongoing discussion among researchers, and proposed methods for addressing this topic, are promising and will only serve to improve not only an understanding of recovery but also, in turn, multiple injuries.

We included all studies where it was clearly stated that an individual unique code had been assigned to identify each included athlete, in which case an individual injury history could theoretically have been traced over the duration of the study. Despite a conservative approach to excluding papers, referring to the full text for information in order to be certain of exclusions, there remains a possibility of having missed papers owing to the challenges of identifying and interpreting authors’ methods for inclusion and recording of injury data. In particular, it was often unclear from the identified studies whether individual data were available based on the described methods of data collection. Nevertheless, it is likely we have captured a highly representative sample of the majority of publications that have adopted the most commonly used methods in the general team ball sports injury research literature. Future work could consider multiple injuries in individual sports and determine whether reporting in this setting is different. It would also be worth assessing if there is better or worse capture of overuse injuries than the more traumatic, time-loss injuries, which tend to feature more heavily in team ball sport research, and determining whether this is due to data collection methods or true prevalence. Although the inclusion criteria specified that all injury types were to be captured by the studies (i.e. excluding studies of one injury type only), the delineation between injury types (traumatic, overuse) was not the focus of our results. As our conclusions are in relation to the reporting of injury counts only, they will apply equally to different injury types and also to non-team ball sports.

We chose to limit our studies to those of more than one season/year duration as an increased number of injuries is likely over increased time and we were interested in how researchers handled this methodologically. However, with this length of time it was not always clear how many individuals were followed. It is possible that in some studies, none, or very few individuals were followed up, as was the case with statements such as “only 45 players (7 %) were present in both years” [37] (p. 277). In some instances, authors reported the number of people who played each of the seasons whether or not they were injured, information that would be valuable to include in future studies. As examples, one study reported the number of athletes for each subsequent season as “thirty six players were followed for one season, 163 for two seasons, 701 for three seasons, 412 for four seasons and 171 for five or six seasons” [48] (p. 464). Similarly, in other papers, authors reported “the average player was in the database for 3.7 ± 3.2 seasons” [26] (p. 286) and 56 of 472 (12 %) athletes were included for the full duration of the study [49]. Our study findings are thus limited by the data reported in the original studies as we cannot be sure how many individuals were followed-up for a given length of time.