The effect of exercise regimens on racing performance in National Hunt racehorses

Authors

  • E. R. ELY,

    Corresponding author
      Present address: Veterinary Laboratories Agency, Centre for Epidemiology and Risk Analysis, Weybridge, Surrey KT15 3NB, UK. Email: e.ely@vla.defra.gsi.gov.uk
    Search for more papers by this author
  • J. S. PRICE,

    1. Royal Veterinary College, Veterinary Clinical Sciences, Hatfield, UK; University of Bristol, Department of Clinical Veterinary Science, Bristol, UK
    Search for more papers by this author
  • R. K. SMITH,

    1. Royal Veterinary College, Veterinary Clinical Sciences, Hatfield, UK; University of Bristol, Department of Clinical Veterinary Science, Bristol, UK
    Search for more papers by this author
  • J. L. N. WOOD,

    1. University of Cambridge, Department of Veterinary Medicine, Cambridge, UK.
    Search for more papers by this author
  • K. L. P. VERHEYEN

    1. Royal Veterinary College, Veterinary Clinical Sciences, Hatfield, UK; University of Bristol, Department of Clinical Veterinary Science, Bristol, UK
    Search for more papers by this author

Present address: Veterinary Laboratories Agency, Centre for Epidemiology and Risk Analysis, Weybridge, Surrey KT15 3NB, UK. Email: e.ely@vla.defra.gsi.gov.uk

Summary

Reasons for performing study: A previous study has identified exercise undertaken during training to be associated with racing performance in flat racehorses. However, no such studies have been conducted in National Hunt (NH) horses.

Aim: To determine whether exercise undertaken during training is associated with racing performance in NH racehorses.

Methods: Data were collected as part of a larger study investigating injury occurrence in NH racehorses. Race records and daily exercise data were obtained from NH racehorses at 14 training yards. Canter, gallop and race distances accumulated in the 30 days preceding a ‘case race’ were calculated. Associations between exercise-, horse- and race-level variables and the odds of winning, winning prize money, being pulled up and falling were identified using mixed effects logistic regression.

Results: Data from 4444 races run by 858 horses were included in analyses. Horses accumulating longer canter distances in the preceding 30 days were more likely to win or win prize money and less likely to be pulled-up. Horses accumulating longer race distances in a 30 day period were more likely to win, whilst those accumulating longer gallop distances in a 30 day period were more likely to win prize money. Horses that had jump-schooled in the preceding 30 days were more likely to fall during the race than those that had not. Trainer and horse were associated with racing performance after adjusting for exercise.

Conclusions: Results from this study suggest that NH race performance may be improved through modification of exercise regimens. After controlling for the abilities of individual trainers and horses and conditions of the case race, horses accumulating longer exercise distances in the 30 days preceding a race were more likely to be successful. However, horses that had jump-schooled in the 30 days preceding a race were more likely to fall.

Introduction

Whilst there have been numerous studies on the effects of exercise regimens on exercise performance in man (Mujika et al. 2000, 2002; Burgomaster et al. 2005; Gibala et al. 2006; Rakobowchuk et al. 2008), there are few equivalent studies in horses. One study in UK flat racehorses found that accumulation of high-speed exercise was associated with the odds of winning a race and that average prize money was higher in horses doing less canter and more high-speed exercise or less high-speed and more canter exercise in the 30 day period prior to the race (Verheyen et al. 2009). To our knowledge, no similar studies have been reported in National Hunt (NH) horses (those that race over jumps). Flat race lengths range from 5–20 furlongs (f) whereas NH races are typically longer, ranging from 16–36 f. It is, therefore, likely that the exercise requirements for optimal performance in flat and NH races will differ; it has been suggested that a 5–6 f race is dependent upon around 33% aerobic energy supply (McMiken 1983), whereas races in excess of 10 f may rely on 90% aerobic energy supply (Evans 1994). Aerobic capacity is increased by long distances of low intensity exercise such as canter (Evans 1994) and, therefore, high cumulative canter distances are likely to be important for optimal performance in NH races.

Unlike in flat races, performance in hurdle and steeplechase races is also dependent upon successful jumping and consequently, NH horses are ‘schooled’ over jumps during training. Despite the importance of falls, both in terms of failure to win and their strong association with injuries both in jockeys and horses (McKee 1995; Pinchbeck et al. 2002; Parkin et al. 2006; Hitchens et al. 2009), only one study appears to have investigated the effect of the number of preceding jump-schooling days on the risk of falling and found no significant association (Pinchbeck et al. 2004). However, much of the horse and exercise-level data in this study were gathered retrospectively and, therefore, may have been subject to recall bias. Given the strong positive association between the deliberate practice of a task and subsequent improved performance of that task (Starkes and Ericsson 2003) it seems likely that horses accumulating more jump-schooling days will be less likely to make a jumping error.

The aim of the work reported here was to determine whether exercise undertaken during training is associated with racing performance in NH racehorses. We hypothesised that horses accumulating longer canter distances in the 30 days preceding the case race would be 1) more likely to win, 2) more likely to win prize money and 3) less likely to be pulled-up during a race. We also hypothesised that horses accumulating more jump-schooling sessions in the 30 days preceding a race would be less likely to fall and/or unseat the jockey.

Materials and methods

Data were collected as part of a larger study investigating the associations between exercise regimens and fractures and tendon/ligament injuries in NH racehorses. Details of the study design, data collection and data management methods were reported elsewhere (Ely et al. 2009). Briefly, the study population comprised NH or dual-purpose racehorses in training at 14 training yards in England. Data were collected between October 2003 and April 2005, incorporating 2 NH racing seasons. Daily exercise activities performed by each horse were recorded by the trainer or designated member of staff. For canter, gallop and race exercise, the distance exercised over in furlongs (one furlong is approximately 200 m) was recorded. Other activities recorded included jump schooling, ‘reduced’ exercise (walk and trot) or ‘rest’ (days where horses were not exercised).

Details of all races run in 2003, 2004 and 2005 by all horses in training at participating yards were obtained from Raceform Ltd, Newbury and were edited manually to remove horses not part of the study sample and to exclude dates outside of the study period. Race details included: race date, race type (i.e. flat, NH flat [NHF], hurdle or steeplechase), finishing position, amount of prize money won and the number of runners in the race.

Horse details, including date of birth, sex and background (ex-flat or ex-store), were obtained from the trainer, the Racing Post online database (http://www.racingpost.co.uk) and/or Horses in Training (Bell 2003; Turner 2004, 2005). Horse background was classified as ‘ex-flat’ if the horse had been in training for flat racing prior to entering NH training and ‘ex-store’ if specifically bred for NH racing and had, therefore, not previously been in training for flat racing.

Data processing and statistical analysis

Total canter, gallop and race distances and the number of jump schooling, reduced or rest days accumulated in the 30 days preceding each ‘case race’ were calculated. Case races where more than 3 days of prior exercise data were not recorded were excluded from analyses. Other exposure variables investigated included the horse-level variables; age, sex, background and trainer and also race-level variables; case race distance, case race type and the number of runners in the case race.

Four binary outcomes of performance were investigated: ‘win’ (whether the horse won the race), ‘won prize money’ (whether the horse won prize money in the race), ‘pulled-up’ (whether the horse was stopped by the jockey during a hurdle, steeplechase or NHF race for reasons such as excessive fatigue or injury) and ‘fell’ (whether the horse fell or unseated the jockey in a hurdle or steeplechase race). Screening of all exposure variables was performed using mixed effects logistic regression including ‘horse’ as a random effect to account for multiple races by individual horses. Continuous exposure variables were initially categorised using quartiles of the distribution of that variable and where there was no evidence of departure from a linear trend, the continuous (rather than categorical) form was used. For variables whose association with the outcome was not linear, fractional polynomial transformations were made using powers -3, -2, -1, 0 (natural logarithm), 1, 2 or 3. Variables with a P<0.30 were considered for inclusion in multivariable models. The models were built using a forward selection procedure whereby variables with a Wald-test P≤0.05 or variables that improved the fit (likelihood ratio statistic [LRS] P≤0.05) were retained in the model. Two-way interaction terms were tested between all exposure variables in the final models. All analyses were conducted using Intercooled STATA 91 and the level of statistical significance set at P≤0.05.

Results

Descriptive results

A total of 5228 races were run by 915 individual horses (75% of the main study population) and ranged from 1–22 races per horse. The remaining 308 horses from the main study population did not race during the study period and were, therefore, excluded from analyses. Hurdle races accounted for 51% of races; 39% were steeplechase races, 7% were NHF races and 3% were flat races. A total of 807 (15%) races were won by 421 horses (46% of horses that raced). Prize money was won in 2700 (52%) races by 695 individuals (76% of horses that raced). Horses failed to finish in 869 (17%) races, out of which horses were pulled-up in 531 (10%) races and horses fell or unseated the jockey at a fence or hurdle in 324 (6%) races. The remaining 14 horses (1%) either refused to jump (n = 10), refused to start (n = 2) or slipped up on the flat (n = 2). Of the 915 horses that raced, 488 (53%) failed to finish in one or more races.

Of the 5228 races, 4444 (85%) run by 858 individuals had at least 27 days of preceding exercise information recorded and were therefore included in analyses. Descriptive statistics of the exercise accumulated in the 30 days preceding the case race are shown in Table 1.

Table 1. Descriptive statistics of exercise accumulated by 915 National Hunt horses in a 30 day period preceding 5228 races
 MeanMedianRange
Canter distance (f)215.81950–862
Gallop distance (f)26.622.50–99
Race distance (f)13.3160–115
Jump schooling days1.410–13
Reduced exercise days3.730–26
Rest days3.740–23

Variables associated with winning the case-race

Following univariable screening, all exposure variables except background, cumulative rest days and case race distance were eligible for inclusion in the multivariable model, which is summarised in Table 2. An initial sharp increase in the likelihood of winning a race with increasing cumulative canter distance up to a distance of around 150 f was followed by a slower rise in odds with longer canter distances (Fig 1): horses accumulating 100, 200 or 300 f of canter in the 30 days preceding a race were 1.4, 1.6 and 1.7 times (respectively) more likely to win than horses that had not cantered. For each furlong increase in race distance accumulated in the 30 days preceding the case race, the odds of winning increased by 1.01 times; such that horses accumulating 60 f of race distance in the preceding 30 days were 1.5 times more likely to win than horses that had accumulated no race distance. There was no significant association between gallop distances or the number of jump schooling days accumulated in the 30 days preceding the case race and the odds of winning after trainer and the number of runners were taken into account.

Table 2. Results of multivariable mixed effects logistic regression analyses of variables associated with the likelihood of winning a race or winning prize money in a race
Outcome: winβ Coeffs.e.Odds ratioWald-test P value95% confidence intervalLRS1
P value
  • 1

    Likelihood ratio statistic.

  • 2

    2 Canter, gallop or race distances accumulated in the 30 days preceding the case race.

  • 3

    3 Horse ID included as a random effect.

Canter distance2 (f)(distance)^-1−0.910.320.400.010.22–0.760.002
Race distance2 (f)Continuous0.010.0031.010.031.00–1.010.03
Age (years)Continuous−0.130.030.88<0.0010.84–0.93<0.001
No. runnersContinuous−0.130.010.88<0.0010.86–0.90<0.001
Trainer (12 d.f.)      <0.001
Horse3      0.02
Outcome: prize money      
Canter distance2 (f)(distance)^-1−0.800.250.450.0010.28–0.73<0.001
Gallop distance2 (f)0–13  Ref  0.04
13.1–230.250.111.280.031.02–1.60 
23.1–360.330.131.390.011.09–1.78 
36.1–990.170.151.190.240.89–1.58 
Age (years)Continuous−0.100.030.90<0.0010.86–0.95<0.001
No. runnersln (runners)−2.410.120.09<0.0010.07–0.11<0.001
Trainer (12 d.f.)      <0.001
Horse3      <0.001
Figure 1.

Graphical representation of the associations between canter distance accumulated in a 30 day period preceding a race and the odds of winning (-—) or winning prize money (- - -). Estimated odds of winning adjusted for race distance accumulated in the preceding 30 days, horse age, the number of runners in the race and trainer. Estimated odds of winning prize money adjusted for gallop distance accumulated in the preceding 30 days, horse age, the number of runners in the race and trainer.

Whilst in univariable analyses older horses had appeared more likely to win a race than younger horses, after controlling for the number of runners in the case race in multivariable analyses, the odds of winning decreased linearly with increasing age. This suggests that younger horses were more likely to compete in races with more runners as horses were less likely to win races with more runners. After accounting for different exercise regimens, both horse and trainer were significantly associated with the odds of winning a race, suggesting that some horses and trainers were more likely to win/train winners than others. Neither sex nor case race type was associated with the odds of winning.

Variables associated with winning prize money in the case race

Following univariable screening, all exposure variables except sex, cumulative rest days and case race distance were eligible for inclusion in the multivariable model, which is summarised in Table 2. The associations between age, cumulative canter distance, trainer, horse, the number of runners in the case race and the odds of winning prize money were similar to that described for the outcome ‘win’.

The odds of winning prize money initially increased with increasing gallop distance, with horses accumulating 13–23 f and 23–36 f of gallop in the 30 days preceding the race being 1.3 and 1.4 times (respectively) more likely to win prize money than horses accumulating less than 13 f. Whilst the odds of winning prize money appeared to decrease as gallop distances exceeded 36 f, they were not significantly different from horses accumulating less than 13 f. Cumulative race distance and jump-schooling days and case race type were not associated with the odds of winning prize money.

Variables associated with being pulled up in a hurdle, steeplechase or National Hunt flat race

Following univariable screening, all variables except cumulative gallop distance and jump schooling days were considered for inclusion in the multivariable model, which is summarised in Table 3. The odds of being pulled up decreased linearly with increasing canter distance accumulated in the 30 days preceding the case race: horses accumulating 300 f of canter were approximately half (OR = 0.55) as likely to be pulled up as those that had not cantered in the preceding 30 days. Race distance accumulated in the 30 days preceding the case race was not significantly associated with the odds of being pulled-up after accounting for the effects of race-level variables.

Table 3. Results of multivariable mixed effects logistic regression analyses of variables associated with the likelihood of horses being pulled-up in a hurdle, steeplechase or NH flat race
  B Coeffs.e.Odds ratioWald-test
P value
95% confidence intervalLRS 1
P value
  • 1

    Likelihood ratio statistic.

  • 2

    2 Canter distances accumulated in the 30 days preceding the case race.

  • 3

    3 Horse ID included as a random effect.

Canter (furlongs)2Continuous−0.0020.0010.9980.011.00–1.000.01
No. runnersContinuous0.090.021.10<0.0011.06–1.13<0.001
(Runners)^2−0.010.0020.99<0.0010.99–1.00<0.001
Case race distance (furlongs)Continuous0.160.021.17<0.0011.12–1.21<0.001
Case race typeHurdle  Ref   
Chase0.400.151.490.011.12–1.99<0.001
NHF−3.371.020.030.0010.01–0.25 
Trainer (12 d.f.)      0.002
Horse3      <0.001

Both horse and trainer were significantly associated with the odds of being pulled-up after accounting for different exercise regimens, indicating that some horses and trainers are more likely to be pulled-up/train horses that are more likely to be pulled-up than others for reasons besides exercise accumulated in the preceding 30 days. Age and background were not significantly associated with the odds of being pulled-up after accounting for the effects of race-level variables.

Horses were more likely to be pulled-up in races with more runners, up to a maximum of 18 runners, after which the odds levelled off. There was around a 2-fold increase in the odds of being pulled for each 5f increase in the distance of the case race and study horses 1.5 times more likely to be pulled up in a steeplechase race than a hurdle race, but about 30 times less likely to be pulled-up in a NHF race than a hurdle race.

Variables associated with falling in a hurdle or steeplechase race

Following univariable screening, all exposure variables except sex, cumulative canter and cumulative race distance were eligible for inclusion in the multivariable model, which is summarised in Table 4. Horses that had jump-schooled in the 30 days preceding the case race were 1.5 times more likely to fall than those that had not schooled. However, this effect was only significant when trainer was included in the model (with trainer excluded: OR = 1.31 CI: 0.98–1.76, P = 0.07). Whilst not significant, it was considered important to include trainer in the multivariable model to control for unmeasured confounders at the trainer-level, such as nonexercise-related management methods. Gallop distances accumulated in the 30 days preceding the case race were not significantly associated with the odds of falling or unseating the jockey during the race.

Table 4. Results of multivariable mixed effects logistic regression analyses of variables associated with the likelihood of horses falling or unseating in a hurdle or steeplechase race
 B Coeffs.e.Odds
ratio
Wald-test
P value
95% confidence intervalLRS1
P value
  • 1

    Likelihood ratio statistic.

  • 2

    2 Horse ID included as a random effect.

Jump-schoolingNo  Ref  0.01
Yes0.430.161.540.011.12–2.12 
Case race typeHurdle  Ref   
Chase1.220.163.39<0.0012.48–4.64<0.001
Trainer (12 d.f.)      0.10
Horse2      0.01

Some horses were significantly more likely to fall/unseat than others. Neither age nor background was associated with the odds of falling after adjusting for case race type.

After taking account of case race type, horses appeared more likely to fall/unseat in races with more runners, but this effect disappeared after excluding seven observations where the case race was the ‘Grand National’. The Grand National is a challenging 36 f race over difficult fences in which falling is common. The number of runners is typically around 40, whereas 99% of races in the dataset had fewer than 24 runners. These observations were therefore excluded from the multivariable model and the number of runners in the race was then not associated with the likelihood of falling or unseating the jockey.

Discussion

Unlike in most human sports, tradition tends to govern racehorse training regimens rather than science. None of the participating trainers formally used techniques such as interval training or tapering, which have been found to improve performance in both human and equine athletes (Shearman et al. 2002; Mujika and Padilla 2003; Burgomaster et al. 2006; Bosquet et al. 2007). Instead, all trainers used a system of daily cantering and twice-weekly high-speed workouts.

Canter distance accumulated in the 30 days preceding the case race was the most consistent predictor of race performance, being associated with 3 performance outcomes. In agreement with our hypotheses, accumulating longer canter distances tended to increase the odds of both winning and winning prize money and decreased the odds of being pulled up during a race. Given that NH races comprise only a small proportion of anaerobic activity such as sprinting or jumping, this agrees with suggestions that aerobic capacity is increased by long distances of low intensity exercise such as canter (Evans 1994). Studies in human athletes have also shown that long distances at submaximal intensities are important for fatigue resistance in endurance events such as rowing, cycling or marathon running (Ingham et al. 2008; Seiler and Tonnessen 2009). Conversely, in a study of flat racehorses that race and train over shorter distances than NH horses (Lonnell 2003), there was no effect of preceding cumulative canter distance on winning or winning prize money, but accumulating greater high-speed distances (gallop and race combined) increased the odds of winning/winning prize money (Verheyen et al. 2009). Interestingly, when looking at gallop and race separately, this study also found that gallop distance was associated with winning prize money but not with winning a race; a finding consistent with that of the present study.

Whilst the odds of winning or winning prize money tended to increase with longer preceding exercise distances, the relationship was not linear. Instead, an initial sharp increase in the likelihood of winning or winning prize money, associated with increasing cumulative canter distance up to a distance of around 150 f, was followed by a slower rise in odds with longer canter distances. Similarly, the odds of winning prize money decreased in horses accumulating more than 36 f of gallop distance following an initial increase. Other studies have suggested that there is not a simple dose-response relationship between exercise and performance (Busso 2003). Using a systems model, Busso (2003) demonstrated an inverted ‘u-shaped’ relationship between exercise and performance when controlling for the fatiguing effects of each exercise bout. Principally, this suggests that there is an optimal level of training for peak performance and deviations from this will lead either to inadequate adaptation or ‘over training’.

In addition to the effects of exercise on performance, increasing age was associated with decreasing odds of winning or winning prize money. As case race distance increased with increasing age and older horses were more likely to run in steeplechases, this decrease in performance with age may have been due in part to the higher risk of failing to finish in older horses. In univariable analyses, the odds of being pulled-up or falling during a race increased with increasing age, but this effect disappeared after controlling for the effects of race type and/or distance.

‘Trainer’ was a significant predictor of race performance after adjusting for the effects of horse-, race- and exercise-level variables. This suggests that unmeasured factors at the trainer-level may influence performance. Management factors may play a role, including variables that are difficult to quantify, such as the trainer's ability to select the correct race for the horse; their ability to tailor the regimen to the individual or the choice of jockey. Trainer was not significantly associated with the odds of a horse falling in race but was included in the final multivariable model to account for its confounding effect upon jump-schooling (there was significant trainer variation in the proportion of training days that horses spent jump-schooling; data not shown). Jump-schooling was the only exercise-level variable associated with falling or unseating the jockey in a race, but contrary to our hypotheses, horses were more likely to fall if they had jump-schooled in the preceding 30 days. It is unlikely that practising jumping makes the horse more likely to fall, but rather horses perceived to be at a high risk of falling are more likely to be schooled before a race than more competent jumpers. The significant horse-level clustering also indicates that some horses are more likely to fall than others and this may relate to innate abilities or jumping experiences gained throughout life. It is likely that jumping skill is acquired over longer periods than 30 days and therefore a wider time-frame may have identified different effects of jump-schooling.

Significant horse-level clustering was also present in the ‘won’, ‘prize’ and ‘pulled-up’ models suggestive of individual predisposition to good or poor performance. Reasons for this may include the heritability of performance (Willham and Wilson 1991) or conformation variation (Weller et al. 2006).

In agreement with Pinchbeck et al. (2004), the risk of falling was higher in steeplechases than hurdle races. However, overall, a greater proportion of starts resulted in a fall than previously reported. We identified a risk of 6.2 per 100 starts compared with 4.3 per 100 starts found by Pinchbeck et al. (2004) at 6 UK racetracks. This difference will be due in part to the inclusion of horses that ‘unseated’ their jockey in the present study. These were included to capture the causes of ‘jumping errors’ that resulted in a failure to finish rather than falls per se.

Whilst the results presented here provide a useful indication of the effects of different exercise regimens on racing performance, they are not without their limitations. ‘Winning’ and ‘winning prize money’ as measures of performance do not account for the abilities of other horses in the race. Furthermore, prize money is given up to sixth place in some races and as 10% of races had fewer than 7 runners it is, therefore, possible to both win money and to come last. However, the latter was controlled for by adjusting for the number of runners in the race. An alternative measure of performance is the amount of prize money won, but this too is an imperfect measure of performance, being equally affected by the abilities of other horses in the race. Furthermore, the presence of large numbers of horses earning zero prize money leads to non-normal distributions that generates problems with model-fit and can lead to nongeneralisable results (Verheyen et al. 2009).

There may be important and different effects of longer or shorter term exercise on race performance that were not detected here, as we chose to limit our investigation to the effects of exercise occurring in the 30 day period preceding the case race. This time period was selected to avoid the inclusion of low intensity exercise performed by horses in the early stages of their racing season, as most horses would undertake at least 4 weeks of ‘full’ training prior to a race, therefore making all horses more comparable. Inclusion of calendar month as a random effect would have partly accounted for this, although not all horses begin training at the same time each year. The start of training may vary by as much as 6 months according to the preferred ground conditions of individual horses and will also be affected by injury (author's personal observations).

In conclusion, results from this study suggest that NH race performance may be improved through modification of exercise regimens. After controlling for the abilities of individual trainers and horses and conditions of the case race, horses accumulating longer exercise distances in the 30 days preceding a race were more likely to be successful. However, horses that had jump-schooled in the 30 days preceding a race were more likely to fall.

Acknowledgements

This project was funded by the Horserace Betting Levy Board and the first author was supported by a PhD scholarship from the Royal Veterinary College. Special thanks go to the trainers, yard staff and veterinarians who participated in this study. Thanks also to Rodney Petinga and Barry Gunning from Raceform Ltd. for their generous provision of the racing data.

Conflicts of interest

None declared.

Manufacturer's address

1 StataCorp LP, College Station, Texas, USA.

Ancillary