A comparison of survival models for assessing risk of racehorse fatality.

This research work used survival analysis to study risk factors associated with fatal injuries of racehorses in the UK, comparing different survival models including racing intensity and types of races. Further, the study utilizes two different measurements: number of days from the first race and number of races until injury. The two proposed models connected to the number of races to injury provide efficient computational strategy and enables testing of various risk associations over time.

Study Model and Data

• The research utilized a Cox regression model and a discrete-time logit model. The Cox model was fitted partially with likelihood using an Efron approximation for handling ties, while the discrete-time logit model was fitted with maximum likelihood.
• The logit model provided improved computational efficiency and enabled checking various functional forms concerning hazard timings.
• Data used in the study were retrospective and derived from all race starts on the 59 courses in Britain from 1990-1999. The data acquisition didn’t include horses that started their racing career abroad.
• The data included a total of 47,424 horses that started their racing career in the UK, accounting for 538,895 starts and 1,228 fatal injuries.

Findings and Observations

• Comparative results from the chosen models corroborated each other and agreed with previously published studies, thereby strengthening the findings.
• It was found that hurdle races and steeplechase presented a higher risk of fatal injury when compared to flat races with relative risks of 1.7 and 1.5, respectively.
• Other factors that showed increased risk included the hardness of the ground, age of the horse, and race distance, with the risk levelling out at 20 furlongs.
• Conversely, the risk decreased with greater racing intensity, reaching a plateau after seven races run within 12 months.
• Horses participating in a new type of race for the first time also exhibited an increased risk of fatal injury (relative hazard of 1.5).

Differences between the Models

• The main divergence between the time models (number of days and number of races) was noted for the factor of age.
• Where the model using number of races identified the age of the horse at the race as more significant, the model using number of days highlighted the age of the horse at the first race.