Cover Story | Sports Medicine

Can you predict which horses will come up lame after a race?

By Marie Rosenthal MS

Perhaps the most difficult job for veterinarians working at the racetrack is predicting which horse is going to suffer a catastrophic injury or lameness while racing. Tim Parkin, BSc, BVSc, PhD, DECVPH, FHEA, FRCVS, a veterinary epidemiologist at the Bristol Veterinary School, University of Bristol, in England, and his colleagues are working on a predictive model to make that job easier.

It has taken years to collect and analyze the data for this undertaking, but Dr. Parkin said a model that track and regulatory veterinarians can use is just a few years away. Soon, veterinarians will be able to review a horse’s history and current status with software on their tablets and, along with their clinical acumen, decide whether a horse should start in that race. The goal is to prevent a catastrophic event, he explained at the 71st AAEP Annual Convention, held in Denver.

“I can report that we have almost reached the horizon,” he said, but the journey has been long. They began collecting digital records in 2009 and have used data from more than 6.5 million starts, more than 350,000 horses and 10 million training events to develop this risk-factor analysis that employs complex statistical modelling to crunch the data.

He explained that they wanted to find “the subtle nuance” in the data that would allow the models to predict which horse could race without injury. “And overall, with all that information—we deal with several billion data points when we run those models—I think the amount of data and the computing power that we have now mean that we can do this quickly,” he said.

The industry already is working to improve the safety of horse racing, most notably with new tracks that conform to the safety and integrity standards set by the Horseracing Integrity and Safety Authority (HISA) and more stringent regulations. Data from the 2025 Equine Injury Database saw a 47% reduction in the fatal injury rate in North American
Thoroughbred racing since 2009, he explained. The rate of fatal injury in Thoroughbred racing was 1.07 per 1,000 starts, which is the lowest rate since the data­base began tracking in 2009, according to the EID annual report.

“That’s approximately 200 fewer horses dying on the racetrack now than there would’ve been in 2009,” Dr. Parkin said. “I think a lot of that has been driven by attitudinal change,” which emphasizes animal welfare.

Risks according to the model
“There have also been some important regulatory changes that we would like to think have been causally associated with reducing that risk as well,”  Dr. Parkin said. For example, claim horses with the strictest type of void claim rule were about 32% less likely to sustain a fatal injury compared with horses at a track where there is no void claim rule.

“Claiming is an important part of the industry over here [in North America]. We know that it’s probably much more prominent here than anywhere else in the world, and it continues to be something that comes out in the model as significant [as a risk factor], particularly claim price,“ Dr. Parkin said. “Those horses in the lowest claiming prices are about 31% more likely to sustain a fatal injury compared with non-claimers. And even the highest claiming prices are at increased risk compared with non-claimers. So, it is certainly the case that being a claiming horse, as many of you are aware, automatically elevates your risk of fatal injury on the racetrack.”

As the claiming price decreases, the risk elevates, he said. With a lower price, the horse was between 10% and 15% more likely to sustain a fatal injury.

Training also comes into play, he explained. Hors­es without pathology that can train before a race have a lower risk of suffering an injury on the track compared with horses that are unable to work out.

Changing trainers also increases the risk of injury, he said. Regardless of how much time the horse spends with the new trainer before the race, its risk for injury increases by 15% to 20% versus horses that keep the same trainer, and that risk remains for up to 4 to 6 races.

A horse placed on the veterinarians’ list is at greater risk, but that risk decreases over time. Horses that race within 30 days of coming off the veterinarians’ list have about a 45% to 55% increased risk for a fatal injury on the track, but that risk decreases when horses get within 6 months to a year off the veterinarians’ list, to about 30%. However, they still have a “10% greater risk of fatal injury compared with horses that have never been put on the list,”
Dr. Parkin said.

Real-world data
Although models are good at predicting risk, there are so many factors that affect an individual horse’s risk that models will not be accurate 100% of the time, Dr. Parkin warned. To improve the predictability of the model, he added, they need more data from trainers and veterinarians.

The team worked with the Hong Kong Jockey Club to demonstrate that increasing the trainer and veterinary data would improve the predictions. In that study, they focused on more common outcomes, which improved the predictive value. Rather than try to predict a fatal injury in this pilot study, they tried to predict post-race lameness or unacceptable performance, he explained.

“You automatically start improving your positive predictive value when you start dealing with outcomes that are more prevalent,” he said, and by default, you also predict less prevalent injuries because some of those with post-race lameness will suffer fatal injuries in future starts.

This type of risk assessment will also enable veterinarians, owners and trainers to decide where to put their resources.
Veterinary histories and training data are critical to these assessments, he said. “The important thing is they must be comprehensive and complete,” which means every horse needs a complete history.

Hong Kong is a good place to study racing because it is a more controlled environment. The Hong Kong Jockey Club stables between 1,300 and 1,400 horses at 1 of their racetracks. All veterinary care is provided by the club. Every horse has a pre-race veterinary inspection by regulatory veterinarians once it’s entered or declared for a race, and has clinical veterinarians who work with the horse.

“They work quite closely together, but they have to stay a little bit apart for obvious reasons, but they all walk around with iPads.” Whenever anything is done for the horse, such as ordering diagnostics, medications or supplements, the information is uploaded automatically to the central databases.

With this type of automatic data entry, the new models achieve 85% predictive accuracy, whereas previously the maximum was 70%, according to Dr. Parkin. This means the predictions are right about 85% of the time, rather than 70%.

“I’ve never ever said to any racing jurisdiction that prior to this with 70% predictability that I thought we should use this in any regulatory or real-time work-on-the-ground because I don’t think that that’s a sufficient level of predictive ability. However, with predictive ability now approaching 85% we have actually started using this in real time,” he said.

“Of the 22 race meetings for which we’ve run these predictions, with 2,387 eligible starts…3 out of 10 horses that were predicted to have some form of deleterious lameness-related issue within 4 days of the race are genuinely going to end up with that.

“The vets are now looking at 10 horses predicted to be high risk, and they’re going to pick up 3 that genuinely are at greater risk and actually maybe wouldn’t have been picked up before.”

At present, “the model is trained to identify post-race lameness, but it’s also good at identifying other related outcomes,” Dr. Parkin said. MeV