Over the six-month meet at Santa Anita racetrack, in Arcadia, California, 30 Thoroughbred racehorses died or were euthanized due to injuries sustained while training or racing. When the precise reasons for catastrophic racehorse injuries aren’t clear, as is the case at Santa Anita, science—specifically, collections of data—can help. But for that data to be most useful for preventing future injuries, it needs to include comprehensive, accurate medication and training history data, says Prof. Tim Parkin, Sc, BVSc, PhD, DECVPH, MRCVS, veterinarian and epidemiologist at the University of Glasgow and consultant to The Jockey Club’s Equine Injury Database (EID).
An epidemiologist is someone who seeks to find the cause of health outcomes and disease in populations. In his EID role Parkin analyzes information on all horses at participating tracks that die or are euthanized as a direct result of injuries sustained while participating in a race and within 72 hours of a race (this includes musculoskeletal injuries, nonmusculoskeletal injuries, and sudden deaths). The database also contains information on training and nonracing fatalities, though those are not included in the EID annual statistics.
We at The Horse were curious about the data the EID is collecting and how the industry could make it even stronger and more useful, so we interviewed Prof. Parkin.
The Horse: Could you describe the gamut of data you collect surrounding each catastrophic injury, please? Surface type, distance, and age are in the reports, but what further data do you collect, if it’s more expansive?
Parkin: There is plenty of other information available to us from The Jockey Club that relates to the racing history of the horse, the conditions on the day of the race and, to a certain extent, prior workout data. Perhaps more importantly we create a lot of new potential risk factors from the data provided.
For example, we calculate for every start the number of prior starts in the last 30 days, 30 to 60 days, and 60 to 90 days, etc. In the next iteration of the models (which assess how successfully a potential risk factor does its job in predicting whether a horse sustains an injury) we are developing, we will also include measures of changes in intensity of racing or working at speed that are likely to help improve the predictive ability of the models. Those models will now include on initial assessment close to 150 different potential risk factors. The two peer-reviewed publications so far … include details of all significant variables we have investigated.
The Horse: Do you also collect information on injuries that are not career-ending or fatal? If so, how do you put that data to use?
Parkin: The EID does collect this information, but it is more difficult to verify and classify. It is easy to tell if a horse has died during racing, but much more difficult to definitively identify horses with particular types of injury. Having said that, nonfatal injuries during racing have been included in previous models predicting fatal injury and have been shown to be significant. IF we were able to collect (with some consistency and certainty) nonfatal injuries from training, I am sure these would be significant in terms of predicting the risk of fatal injury during racing.
The Horse: We learned about the importance of “big data” in an interview in March and said that epidemiologists have evaluated at least 300 potential risk factors in this area of study. Does the EID produce the amount of big data needed to realistically pinpoint the combination of factors causing the breakdowns at Santa Anita? If not, what more would it need to collect/would you like to see?
Parkin: Big data has two aspects:
- The number of observations collected (in this case, starts by individual horses on the track). In this regard we have plenty of data – multi-millions of rows of data with each row representing a different start, so there is no problem with statistical power of the studies to identify risk factors if they exist.
- The scope of the data – in other words how many different variables, covering different aspects of the horse’s career,,are collected. This is the area that should be the focus for the next generation of models. If we knew exactly what every horse experienced on every day of its life (nutrition, training, veterinary history, treatments etc.), I am sure our models would be much more predictive.
The work I presented at the Welfare and Safety Summit in 2016 shows that we could explain about 35% of the drop in fatal injury rate from 2015. That left 65% unexplained, which I, at least in part, attribute to information and data associated with training and veterinary histories. The goal of this work is to predict 100% of what is predictable. There will always be some “bad luck” involved in (racing) fatal injuries and we will never be able to predict and prevent all injuries, but with additional data scope, covering as many areas of the horses’ “lifetime experience” as possible, I am convinced we can do better.
The Horse: In a release in March you noted a need for focus on the medications present in horses during racing and training, transparency of veterinary records for all starters. What challenges do you face in increasing the amount of data you can collect on this, especially regarding testing challenges for some drugs? And with the current focus on safety after the Santa Anita deaths, do you expect transparency to improve? What will it take to get information on medication that you need?
Parkin: This is really very simple. We need those involved in racing who care for and are responsible for horses that race for our pleasure to understand the importance and potential impact of medications on risk during racing. It needs people like me to be better at discussing and explaining the importance of these factors to stakeholders. We need to be better at providing the evidence-base showing the link between, in particular, the use of NSAIDs and corticosteroids, etc., and the risk of breakdown. Horses that are able to race (or train) faster than they otherwise would, simply because pain is being masked, are bound to be at greater risk of injury.
Reliable medication information will only be made available with stakeholder buy-in. Compelling declaration of medical records without ensuring that trainers and others recognize the importance of this information will compromise the quality of that information and compliance with its supply. Part of the work we do to encourage supply of this information has to be a clear recognition, on our part, of the pressures that trainers face in terms of needing to get horses to the track and the accompanying financial considerations. We need to work with them, hand-in-hand, so that we all have the opportunity to contribute to improved welfare of the racehorse.
The Horse: You’d also mentioned the need for information from morning training hours. It sounds like horses sustaining fatal catastrophic injuries during morning training hours also undergo necropsies, but is that information less complete than what’s collected from racing? What about non-career-ending injuries?
Parkin: As above, the more information we have from any source the more predictive our models will be.
The Horse: For nonfatal catastrophic injuries that undergo surgery, is there useful additional information to be gained by how the horse’s bones heal? Or would that dataset be too small to give credence?
Parkin: Probably of use, but as you suggest it would be such a small subset of the data that it would not influence the overall picture.
The Horse: Sky’s the limit, if you were to curate the perfect information set to dial down the specific constellation of factors that leads to breakdown, in order to prevent it, what would it include?
Parkin: On top of what we already have, these would be my priorities:
- Complete and accurate medication and treatment records with accompanying diagnostic reasons for the particular veterinary
- Complete and accurate training information (primarily relating to work, but also slower work and, if appropriate, work).
- Anything else! But in reality Ithink the potential impact of the first two would significantly outweigh additional benefit provided by all other types of data.