Defining subsequent injury risk in relation to recovery from an initial or index injury

To date, subsequent injuries have generally been considered from a clinical management and return-to-play (or time away from sport to recover from injury) perspective.1–3 10–12 However, there are many issues with defining and recording subsequent injuries in the sporting context. Sports injury surveillance guidelines and several conceptual papers describe the complex issues associated with properly classifying injuries as recurrent, reinjury, exacerbations or overuse.4 ,8 ,13 ,14 There has also been considerable attention given to consensus statements for sports injury surveillance, with the vast majority considering recurrent injury as one type of subsequent injury.13–16 Most of these follow the suggestions of Fuller et al,7 though a new conceptualisation of overuse injury has been proposed by Bahr,8 and a refined definition of subsequent injury was given more recently by Hamilton et al.4 Reported definitions of subsequent injury have been both sport-specific and context-specific and whether or not the person has fully or partially recovered (or continues to play injured) from a previous injury has been considered to varying degrees.4 ,7 ,10

The most recent taxonomy of subsequent sports injury4 recommends that injuries be coded as: (1) new injury=different location, (2) local injury=same location but different type and (3) recurrent injury=same location and same type. In this taxonomy, the last two categories represent injuries that could be correlated with an initial injury. It has been accepted that the nature (at a tissue level), rather than just the general location of the injury (eg, knee or thigh), is also important when categorising repeat injuries.7 Others have argued that a recurrent injury should only be defined as one that is of the same type occurring at the same site but only after full recovery from an earlier injury.7 ,13 ,14 None of this prior work, however, discusses subsequent injuries from a statistical viewpoint, and therefore adequate recognition of the various dependencies (or possible relationships) between injuries within players is lacking in the sports medicine literature.

A new model of subsequent injury categorisation (the SIC model)

The SIC model is one that takes into account the need to include both acute and overuse injuries and ten different dependency structures between injury types, and is outlined in table 1. Compared to the existing sports injury definitions, it adopts a more statistically oriented approach towards categorising injury types that makes provision for all potential within-person injury dependencies.

This approach is consistent with statistical modelling principles.17 The model was constructed using Bayesian data analysis principles that first require that a full probability model of all the possible outcomes be constructed.18 The starting point was therefore to construct a probabilistic model that included all combinations of the possible relationships between two injuries, irrespective of how likely they were. These categories were then checked to ensure that they had clinical meaning and that they were relevant to most sports injury data collections. In Bayesian statistical parlance, this is the prior predictive distribution of subsequent injury outcomes.

Table 1 also maps this new SIC model against the recent Hamilton et al4 subsequent injury definition and the Fuller et al7 recurrent/reinjury/exacerbation definitions. The overuse definitions are consistent with the approach advocated by Clarsen et al.19

It is recognised that for some real-world datasets some of the dependency categories within the SIC model may not apply (with the potential for some zero count cells), but it is still important to include them in an overall categorisation model, especially from a statistical model point of view. The categories within the SIC model are therefore holistic and intended to capture all types of subsequent injuries, while also being mutually exclusive. For example, to make a clinical case for category 5, a back-related hamstring strain could have an unclear clinical diagnosis; at times, it could present only like a hamstring strain and at other times, only like a low back issue. A player who presents twice with this injury will fit category 5. Category 6 is easier to define and would include cases such as a player who sustains a fracture from a direct blow in one season and then fully recovers but sustains a similar fracture in the following season; this clearly fits category 6 rather than categories 7–10. Similarly, the category ‘no injury’ is included for model completeness from a statistical viewpoint, but is expected to have zero counts.

Using the SIC model to categorise injuries: incorporating multiple index injuries

Consistent with the Bayesian statistical modelling approach,18 the next stage was to calculate the posterior probability distribution by applying the model to an existing dataset and, in this case, to count the number of injury cases assigned to each category.

Comparison in a pairwise fashion to an earlier injury is required to assess the recurrent nature of an injury. Access to reliable information before the first injury recorded or onset of the problem is difficult, and so the first index injury needs to be taken as the first recorded injury for a given athlete in a case series or new study. Using data from existing prior sources could potentially bias the data in relation to the duration and quality of the information contained in those records. For overuse injuries, this first index injury can only be taken to be the first (recorded) time presentation of symptoms as there is no way to determine loss of structural and functional integrity of body tissue that remains asymptomatic, and asymptomatic pathology in the sporting population is high.

Depending on the nature of the first index injury, it may not be reasonable to assume that all subsequent injuries will be related to that one. If an athlete sustains several injuries, and these are recorded in sequence, then it is possible that there will be different index injuries. This is analogous to the competing risks issue that arises in survival analysis where people can be at risk of several different health problems at the same time.20 To allow for competing injury risks, the model needs to allow different and multiple index injuries.

Application of the SIC model to sports injury data

The suitability of the SIC model was assessed with sports injury data from the Preventing Australian Football Injuries through eXercise (PAFIX) study. Full details of the study methods and data collection processes have been published elsewhere,21 ,22 as has a specific description of the nature, body region and cause of these injuries.23 All injuries were assigned a diagnosis code according to the internationally recognised Orchard Sports Injury Coding System (OSICS-10).24

The OSICS-10 codes assigned to each injury were used to categorise them, and decisions were supplemented with the body region and nature of injury codes, as well as text narrative descriptions of the injury events. All injuries for a given person were ordered by date of occurrence, to provide a temporal sequence of injuries for each player. The first injury was taken to be the first index injury against which the SIC model was applied, but it was also recognised that there could be more than one index injury within a given series of injuries in some athletes.

Over the 2007 and 2008 playing seasons, the PAFIX study prospectively recorded 1082 injuries sustained by 1564 community Australian football players.21 A total of 36% of the players sustained at least one injury (range 1–8).

All injuries were ordered sequentially according to the date of their occurrence to determine the first index injury and all other subsequent injuries following it (either another index injury or related injuries). A subsequent injury was defined to have occurred in the subset of players with two or more reported injuries. Overall, 469 subsequent injuries were sustained by 282 players. According to the Hamilton et al4 sports injury taxonomy, 12% of these subsequent injuries were recurrent and hence likely to be dependent on previous injuries (table 2).

Table 3 classifies the 469 subsequent injuries in the PAFIX dataset according to the SIC model categories, relative to the first injury recorded in the dataset (ie, the first index injury) and other identified subsequent index injuries. This table shows what the model looks like when applied to a sports injury dataset and provides the Bayesian posterior distribution18 of the model categories. The most likely reason for there being no cases coded to category 6 and this dataset recorded injuries to players over one playing season only. Collectively, 15.6% of all subsequent injuries in the PAFIX community Australian football injury dataset would be coded to categories representing injuries that are directly related to previous index injuries. This can be contrasted to the 12% shown in table 2, where injuries were categorised according to the Hamilton et al4 scheme.