The odds of suspensory apparatus failure, for horses with rim shoes, was about a third that of horses without [67]. Interestingly, a follow‐up study conducted approximately 10 years later failed to identify the same toe‐grab‐related associations [45]. The authors suggest that this may have been due to the confounding effect of other exercise‐related variables. It is also likely that publicity surrounding the initial work and the subsequent banning of high toe grabs on front shoes is a demonstration of intervention and messaging that has had the desired effect. In other words, 10 years on, the effect of this particular risk factor has somewhat diminished.
The value of being able to include treatment records in predictive models was demonstrated in a study examining the hazard of musculoskeletal injury in the days and weeks following local corticosteroid injection [62]. This compared the incidence in untreated horses or in horses prior to treatment with local corticosteroid injection (1.22 per 100 horse months) with that post treatment. On average, the hazard increased by 4.8 times and only returned to a level that was indistinguishable from that seen in untreated horses or horses prior to treatment after 49 days. Work from the USA was less precise in the estimates of risk, but nevertheless also showed that once a horse has been placed on a ‘vets list’ the risk of fracture for that horse was 80% greater in all future starts, regardless of how long after being placed on the vet list those starts occurred [19]. In certain races in South America, declared use of phenylbutazone is permitted, and horses competing in these races were at approximately 80% greater odds of musculoskeletal injury compared with horses competing in races in which phenylbutazone was not permitted [68, 69]. Similarly, in a case series of almost 1500 horses receiving intrasynovial medication, those horses that had received more than three previous ‘treatments’ were more than twice as likely to sustain a fracture within 56 days, compared with those that had been medicated fewer than four times [70]. Finally, a questionnaire survey of attending veterinarians in California demonstrated that horses that had been subject to euthanasia due to catastrophic musculoskeletal injury were more likely to have been lame in the period three months prior to the date of death [71]. The study also attempted to identify associations with recent medication usage, but largely due to the high proportion of horses (both cases and controls) on medication and the relatively low level of statistical power, no such associations were identified.
In order to better determine the true relationship between veterinary and medication history and the risk of fracture or other injury, it is essential that racing jurisdictions make further attempts to encourage the accurate recording of and willingness to share medical records. Once issues surrounding confidentiality are addressed and this becomes normal practice, the inclusion of such data in future predictive models will undoubtedly improve our ability to better identify horses at increased risk of fracture.
Predictability and Potential for Effective Screening
The majority of work in the field of prediction and screening for injury or fracture has focused on joint injury and potential biomarkers in blood or urine [72–74]. The link between joint injury and subsequent fracture is often referred to in these publications, suggesting that the ultimate goal is to predict and prevent catastrophic fracture (as well as less dramatic lesions such as osteoarthritis). A recent review highlights that there is some way to go before blood or urine biomarkers will be useful, and highlights the need to establish standardized methods of sample collection, reproducible marker measurement and well‐documented biobanks [72].
Two papers looking at biomarkers for fracture or musculoskeletal disease in two‐ and three‐year‐old Thoroughbreds demonstrate both the difficulties and also their potential for future use [73, 74]. The first failed to identify any significant associations between ‘start of season’ bone biomarker levels and subsequent fracture [73]. This is perhaps unsurprising as fractures often occurred some months after blood samples were taken. The second study addressed this deficiency by acquiring monthly blood samples during training [74]. When longitudinal samples were investigated, there were some significant associations between changes in a range of biomarkers and subsequent musculoskeletal disease (one of which was stress fracture). The authors quote a 73.8% ability to correctly classify horses as injured or not. However, further calculations from their data show negative and positive predictive values for all injury types of 81 and 68%, respectively. In other words, 68% of subsequently injured horses would be identified as at risk before the event. This may be seen as some degree of success, but as in the imaging studies referred to below, it is important to remember that positive predictive value is strongly influenced by the prevalence of the outcome under investigation. So, when trying to predict catastrophic fractures, which are relatively rare, without a test specificity of very close to 100% the positive predictive value of a test will always drop off very quickly meaning that the number of false positives generated by that test will be high [75].
More recently, magnetic resonance imaging (MRI), computed tomography and microradiography have been assessed with regard to their ability to identify pre‐fracture changes in Thoroughbreds. As with the blood biomarker study [74], some encouraging findings were identified in a study of MRI and the risk of lateral condylar fracture in Mc3 [75, 76]. In particular, the authors identified an optimal cut‐off in the depth of dense subchondral and adjacent trabecular bone in the palmar half of the lateral parasagittal groove, detectable using MRI, that could best discriminate between bones of horses that had and had not sustained a lateral condylar fracture. The authors went as far as calculating the positive and negative predictive values of this cut‐off in horses with fractures, and the wider population demonstrating a clear difference when the prevalence of the outcome of interest is so different. In the study population, because the work was designed as a case–control study, the prevalence of lateral condylar fracture was 47%, thus producing a positive predictive value of 84%. However, when proposed as a screening tool for, at the extreme, all Thoroughbreds in training (for which the prevalence of lateral condylar fracture was estimated to be 0.5%), the positive predictive value drops significantly to 3%. In other words, only 3% of ‘test‐positive’ horses would be truly positive and therefore at genuine risk of lateral condylar fracture [75]. Nevertheless, the authors do point out that routine MRI assessment of the depth of dense subchondral/trabecular bone within the palmar half of the lateral sagittal groove of distal Mc3 can be a useful ancillary test when investigating lameness in racehorses. In essence, the clinical examination in this situation is acting as the pre‐screening test, selecting for MRI examination those horses that are more likely to be at risk of fracture.
Work with computed tomographic images [77, 78] and microradiography [79] has shown some potential to predict fractures of the lateral condyle of Mc3 [77] or proximal sesamoid bones [78, 79]. Bone density at the distal articular surface of Mc3 was significantly greater in fractured bones and contralateral bones from the same horse, compared with bones from horses that had sustained a non‐limb‐related death or euthanasia while racing. The heterogeneity of articular surface bone density was also greater in fractured bones [77]. In proximal sesamoid bones, morphometric differences were detected between fractured and non‐fractured bones such that the abaxial margin of the medial base of fractured bones was up to 3.5 mm more prominent than in non‐fractured bones [78]. Both fractured and non‐fractured bones from horses that had sustained a proximal sesamoid bone fracture had more compact trabecular bone than bones from horses that had died for other reasons [79].
If identification of markers for pre‐fracture change can be detected, using either existing or future imaging modalities, then reliable screening methods for fracture risk can be developed. A key advance will be our ability to correctly identify and select horses for whom such ‘intensive’ screening would be most useful. Identification of a reliable pre‐imaging screening test that effectively rules out a large proportion of horses from being at risk of fracture should therefore be the priority. In other words, a simple, quick and cheap pre‐screening test that has a very high sensitivity (resulting in very few false negatives and a high negative predictive value) is required so that only those horses that are at greatest risk are evaluated. This would effectively increase the prevalence of fracture in the population subjected to the secondary test with resultant
