Risk factors for race-associated sudden death in Thoroughbred racehorses in the UK (2000–2007)

Authors

  • C. H. LYLE,

    Corresponding author
    1. Royal (Dick) School of Veterinary Studies and Roslin Institute, University of Edinburgh, Easter Bush Veterinary Centre, Roslin, Midlothian, UK
    • email: Catriona.Lyle@up.ac.za. Present address: Department of Companion Animal Clinical Studies, Onderstepoort Veterinary Academic Hospital, University of Pretoria, Private Bag X04, Onderstepoort, Pretoria, 0110, South Africa.

    Search for more papers by this author
  • K. J. BLISSITT,

    1. Royal (Dick) School of Veterinary Studies and Roslin Institute, University of Edinburgh, Easter Bush Veterinary Centre, Roslin, Midlothian, UK
    Search for more papers by this author
  • R. N. KENNEDY,

    1. Royal (Dick) School of Veterinary Studies and Roslin Institute, University of Edinburgh, Easter Bush Veterinary Centre, Roslin, Midlothian, UK
    Search for more papers by this author
  • B. C. MC GORUM,

    1. Royal (Dick) School of Veterinary Studies and Roslin Institute, University of Edinburgh, Easter Bush Veterinary Centre, Roslin, Midlothian, UK
    Search for more papers by this author
  • J. R. NEWTON,

    1. Animal Health Trust, Lanwades Park, Kentford, Newmarket, Suffolk, UK
    Search for more papers by this author
  • T. D. H. PARKIN,

    1. School of Veterinary Medicine, College of Medical, Veterinary and Life Sciences, University of Glasgow, Glasgow, UK
    Search for more papers by this author
  • A. STIRK,

    1. Equine Science and Welfare Department, British Horseracing Authority, London, UK
    Search for more papers by this author
  • L. A. BODEN

    1. Institute of Biodiversity, Animal Health and Comparative Medicine, College of Medical, Veterinary & Life Sciences, University of Glasgow, Glasgow, UK.
    Search for more papers by this author

Summary

Reasons for performing study: Sudden death adversely affects racehorse welfare, jockey safety and the public perception of horseracing.

Objective: To describe the risk of racing-associated sudden death in Thoroughbred racehorses in the UK from 2000 to 2007, to identify whether there were risk factors uniquely associated with sudden death and to improve the understanding of the pathogenesis of racing-associated sudden death by identification of risk factors for such cases.

Methods: A sudden death was defined as an acute collapse and death, in an apparently healthy Thoroughbred racehorse, during or immediately after racing, in the absence of clinical data indicative of a catastrophic orthopaedic injury. The retrospective study included 201 case race starts and 705,712 control race starts. Univariable and multivariable logistic regression were used to identify risk factors for sudden death at any one start.

Results: In the multivariable model, age, distance, race type, season and number of starts in the 60 days prior to the race were associated with sudden death.

Conclusions: The risk factors identified in this study are not uniquely associated with sudden death and have been also been identified in studies using all causes of fatality as the outcome. These data suggest that a generic approach to reduce fatal musculoskeletal injury and sudden death may be possible.

Potential relevance: The identification of risk factors allows speculation on the underlying mechanisms of sudden death in racing. This may stimulate hypothesis-led investigations into the pathogenesis of exercise-related arrhythmias, exercise-induced pulmonary haemorrhage and blood vessel rupture.

See also correspondence by Cook

Abbreviations
CI

Confidence interval

EIPH

Exercise-induced pulmonary haemorrhage

OR

Odds ratio.

Introduction

Sudden death associated with exercise in the apparently healthy Thoroughbred racehorse appears to be a rare occurrence; however, the risk of such events has only been quantified in racehorses in Victoria, Australia [1]. The risk of sudden death in that population was 0.08 per 1000 starts in flat races and 0.29 per 1000 starts in jump races, and the proportion of racing fatalities that were classified as sudden death (proportional mortality rate of sudden death) was 19% in flat races and 3.5% in jump races. In other Thoroughbred populations where proportional mortality rates have been recorded, similar proportions of racing fatalities were attributed to sudden death: 12% (256/1981) in the UK (2000–2009; data supplied by the British Horseracing Authority, reproduced with permission) and 9% (58/659) in California, USA [2].

Risk factors for catastrophic musculoskeletal injury associated with racing have been investigated [3–17]. Other risk factor studies have used fatality as an outcome [18–21] and have therefore included catastrophic musculoskeletal injuries as well as sudden death cases. These studies using fatality as an outcome have identified associations between fatality and age, sex, prior racing history, race length, racecourse location, surface conditions, race type, running the first race of a new type, season and year [18–21]; however, there are no published studies investigating risk factors for racing-associated sudden death alone. This is probably due to the rarity of sudden death and the difficulty in clarifying case definition without post mortem examination. In most racing jurisdictions, post mortem examination is not routinely done owing to logistical difficulties and expense.

The aims of this study were to describe the risk of sudden death in the UK, to identify whether there were factors uniquely associated with sudden death and to improve the understanding of the pathogenesis of racing-related sudden death through the identification of risk factors. A sudden death was defined as an acute collapse and death, in an apparently healthy Thoroughbred racehorse, during or immediately after racing, in the absence of clinical data indicative of a catastrophic orthopaedic injury.

Materials and methods

Potential risk factors for sudden death in flat or jump races in the UK from 1 January 2000 to 31 December 2007 were investigated using a retrospective case–control study with 201 case starts and 705,712 control starts. The study was conducted at the level of the start (where a start represented a horse starting the race).

Selection of cases and controls

This study was specifically designed as a whole-population study to make use of systematically collected data from all starts in the given study period. Data on all racing starts from 1 January 2000 to 31 December 2007 inclusive were obtained from the Equine Science and Welfare database compiled by the British Horseracing Authority.

A case start was defined as a start in a race that resulted in sudden death. A sudden death was defined as an acute collapse and death, in an apparently healthy Thoroughbred racehorse, during or immediately after racing, in the absence of clinical data indicative of catastrophic orthopaedic injury.

Three cases classified as sudden death by the racecourse veterinarians were removed from the database because they occurred before the start. All noncase starts in the database were used as control starts, except one start where data were incomplete. A control start could therefore be a case horse in a start prior to its death or a start for a noncase horse. This resulted in 3511 controls per case start.

Risk factors

A total of 25 variables for each start were used in the analysis. These variables comprised 4 horse-related variables (age, animal type, use of eye equipment and use of a tongue-tie), 7 prior racing history-related variables (starts ever, starts in the last 30 days, starts in the last 60 days, starts in the last 90 days, starts in the last 180 days, starts in the last 365 days and total money won), 2 jockey- and handicap-related variables (weight carried and type of handicap), 10 race-related variables (distance, change in run distance, race class, race time, race type, number of runners, running sequence, season, selling/claiming and the value of the race) and 2 track-related variables (surface and going).

Details of the 705,913 starts and potential explanatory variables were downloaded into an Access database (Microsoft Access 2003)a.

Data analysis

A power calculation indicated that the study would provide at least 80% power to detect odds ratios of 1.7 or more, with 95% confidence, given exposure prevalence in the control population of between 0.13 and 0.73.

Descriptive statistics (mean, median, minimum and maximum) were generated for each continuous variable [18]. The ‘best fit’ of the variable was determined by graphical assessment of the relationship between the log odds of the outcome by categories of independent variable [18]. If the relationship was nonlinear, binary or polytomous categorical terms were investigated at the univariable level and multivariable level [22].

Potential risk factors were screened using univariable logistic regression. Variables with a likelihood ratio test P≤0.25 were available for inclusion in subsequent single-level, multivariable logistic regression model building.

A causal web diagram (supplementary information Fig S1) was also developed based on biologically plausible hypotheses. The multivariable model was created using a manual method based on the causal web. Variables were retained in the multivariable model if the likelihood ratio test P values were <0.05 when establishing the model. The Wald test P value was used when comparing categories with the reference category. The multivariable model therefore identified the effects of the retained variables, having controlled for the effects of other significant variables within the model. Biologically plausible interactions between variables in the final model were identified, and interaction terms were generated and assessed [18]. All starts were investigated for clustering within horse, jockey, trainer, course and sire (data on case dam were not available). Intraclass correlation coefficients, variance inflation factors and rhos were calculated for each of these levels.

The fit of the final single-level multivariable model was assessed using the Hosmer-Lemeshow goodness-of-fit test [23]. Regression diagnostics were performed, and covariate patterns with the greatest Pearson residual, standardised Pearson residual, leverage, delta beta, delta χ2 and delta deviance values were identified. Individual observations within these covariate patterns were then removed from the model and changes in the direction and value of the coefficients examined [23,24]. The predictive ability of the model was determined by generating a receiver operating characteristic curve.

StataSE 10.0b was used to perform all statistical analyses.

Results

A total of 705,913 starts were represented in the study population, with 201 case starts and 705,712 control starts. Sudden death occurred during the race in 114 of the case starts and immediately after the race in 70 of the case starts. The timing of the sudden death was not specified in the other 17 cases.

The overall risk of sudden death was 0.3 per 1000 starts (201/705,913). The risk of sudden death in the different race types was 0.07 per 1000 starts (22/319,872) in turf flat races, 0.09 per 1000 starts (12/128,699) in all-weather track flat races, 0.4 per 1000 starts (9/21,518) in National Hunt flat races, 0.5 per 1000 starts (71/146,543) in hurdle races and 1 per 1000 starts (87/89,281) in steeplechases.

Descriptive statistics for the continuous variables entered into the final model are presented in Table 1. Univariable analysis for all the variables entered into the model-building process is presented in Table 2. A total of 15 variables of the 25 variables screened at univariable level were used in the building of the final multivariable model. Univariable analysis of variables that were not entered into the model-building process is presented as supplementary information online (Table S1).

Table 1. Descriptive statistics for continuous variables submitted to the multivariable model for sudden death in Thoroughbred racehorses in 705,913 race starts in the UK from 2000 to 2007
VariableNumber of observationsMinimal value25th percentileMedian valueMean value75th percentileMaximal valueLinearBest fit
Age (years)705,9132344.98617YesContinuous
Distance (km)705,91311.42.02.463.47.2YesContinuous
Starts in the last 30 days705,9130010.8619YesContinuous
Starts in the last 60 days705,9130111.68315YesContinuous
Starts in the last 90 days705,9130122.31318NoCategorical
Starts in the last 180 days705,9130133.60535NoCategorical
Starts in the last 365 days705,9130256.14960NoCategorical
Weight carried (kg)705,91344.4555.7958.5161.0567.5982.10NoCategorical
Table 2. Univariable analysis of variables (P≤0.25) submitted to the multivariable model for sudden death in Thoroughbred racehorses in 705,913 race starts in the UK from 2000 to 2007
VariableTotal (n = 705,913)Cases (n = 201)Controls (n = 705,712)CoefficientStandard errorP valueOdds ratio95% confidence interval
  • *

    P values were based on likelihood ratio tests; other P values are Wald test P values. Abbreviations: F = firm, Fa = fast, G = good, ref. = reference category, S = slow, St = standard, Sl = slow.

Age (years)705,913201705,7120.370.024<0.0011.441.38–1.51
Animal type     <0.001*  
 Neither maiden nor novice481,777140481,637   1 (ref.) 
 Maiden121,59513121,582-1.000.2900.0010.370.21–0.65
 Novice102,54148102,4930.480.1670.0041.611.16–2.24
Change in run distance     0.014*  
 None402,74894402,654   1 (ref.) 
 Increase167,21760167,1570.430.1650.0091.541.11–2.13
 Decrease135,94847135,9010.390.1790.0281.481.04–2.1
Distance705,913201705,7120.740.053<0.0012.091.89–2.32
Going     0.100*  
 G, G to F, G to S, St, St to Fa, St to Sl568,773150568,623   1 (ref.) 
 Heavy, soft, slow111,11243111,0690.380.1730.0271.471.05–2.06
 Hard, firm, fast26,028826,0200.150.3630.6731.170.57–2.37
Race time     0.199*  
 Evening94,8452194,824   1 (ref.) 
 Afternoon or morning611,068180610,8880.290.2310.2161.330.85–2.09
Race type     <0.001*  
 Flat448,57134448,537   1 (ref.) 
 Hurdle146,54371146,4721.860.209<0.0016.394.25–9.62
 Steeplechase89,2818789,1942.550.202<0.00112.878.66–19.13
 National Hunt flat21,518921,5091.710.375<0.0015.522.65–11.51
Season type     0.096*  
 Spring/autumn/winter496,731152496,579   1 (ref.) 
 Summer209,18249209,133-0.270.1640.1040.770.55–1.06
Starts in the last 30 days705,913201705,712-0.440.098<0.0010.640.53–0.78
Starts in the last 60 days705,913201705,712-7.780.098<0.0010.760.68–0.85
Starts in the last 90 days     0.014*  
≤3starts532,509166532,343   1 (ref.) 
4 or more starts173,40435173,369-0.110.0470.0190.900.82–0.98
Starts in the last 180 days     <0.001*  
 0–1223,82693223,733   1 (ref.) 
 2–3181,88042181,838-0.590.1860.0020.560.39–0.8
 4–5129,91734129,883-0.460.2000.0210.630.43–0.93
 6+170,29032170,258-0.790.205<0.0010.450.3–0.68
Starts in the last 365 days     <0.001*  
 0–2205,80879205,729   1 (ref.) 
 3–5170,76955170,714-0.180.1760.3180.840.59–1.18
 6–9168,97243168,929-0.410.1900.0300.660.46–0.96
 10+160,36424160,340-0.940.233<0.0010.390.25–0.62
Surface     <0.001*  
 All-weather track128,69912128,687   1 (ref.) 
 Turf577,214189577,0251.260.298<0.0013.511.96–6.3
Weight carried     <0.001*  
 Less than 63.5 kg449,45945449,414   1 (ref.) 
 63.5 kg or greater256,454156256,2981.800.169<0.0016.084.36–8.47

The final multivariable model is shown in Table 3. This model represents the effects of the retained variables, having controlled for other significant variables identified in the univariable analysis. Sudden death was associated with age, distance of race, type of race, season and the number of starts in the 60 days prior to the race. Increasing age was associated with increased odds of sudden death (odds ratio [OR] per extra year 1.3, 95% confidence interval [CI] 1.2–1.4). Increasing distance was associated with increased odds of sudden death (OR per extra kilometre 1.3, 95% CI 1.1–1.6). Increasing number of starts in the last 60 days was associated with reduced odds of sudden death (OR per extra start 0.8, 95% CI 0.7–0.9). Compared with racing on the flat, sudden death was more likely to occur in starts in hurdle races (OR 2.2, 95% CI 1.2–4.0), steeplechase races (OR 2.3, 95% CI 1.1–4.6) and National Hunt flat races (OR 3.1, 95% CI 1.4–7.1). Racing in the summer was associated with increased odds of sudden death when compared with racing in spring, winter or autumn (OR 1.8, 95% CI 1.3–2.5).

Table 3. Final multivariable model for sudden death in Thoroughbred racehorses in 705,913 race starts in the UK from 2000 to 2007
Risk factors for fatalityCoefficientStandard errorP valueOdds ratio95% confidence interval
Race type     
 Flat   1 (ref.) 
 Hurdle0.780.310.0122.181.19–4.01
 Steeplechase0.810.360.0242.261.12–4.57
 National Hunt flat1.140.420.0063.131.39–7.07
Season type     
 Spring/autumn/winter   1 (ref.) 
 Summer0.570.170.0011.771.26–2.48
Age (years)0.250.03<0.0011.291.21–1.37
Distance (km)0.250.100.0161.281.05–1.57
Starts in the last 60 days−0.180.060.0030.830.74–0.94

None of the interaction terms was statistically significant. Although there was no evidence of clustering when assessing rho or the intraclass correlation coefficients, the variance inflation factors for course, horse, jockey, sire and trainer were greater than one; however, the large variance inflation factors were driven by the large number of observations per level (because the intraclass correlation coefficient values were small). To investigate this further, the model was fitted with course, horse, jockey, sire and trainer as random effects. There was no significant change in coefficients, standard errors or odds ratios, so the fixed effects model was retained.

Removal of covariate patterns with the greatest Pearson residual, standardised Pearson residual, leverage, delta beta, delta χ2 and delta deviance values did not alter the direction of any of the odds ratios. No particular covariate patterns were considered to be influential.

There was no evidence of lack of fit for the model (Hosmer-Lemeshow statistic 0.93, P = 0.82). The area under the receiver operating characteristic curve was 0.81.

Discussion

This is the first study to investigate risk factors for racing-related sudden death alone.

In the final model, one horse-level variable (age), one prior racing history-related variable (starts in the 60 days prior to the race) and 3 race-level variables (distance, type and season) were associated with sudden death.

The area under the receiver operating characteristic curve was 0.81. A model with an area under the receiver operating characteristic curve of greater than 0.8 is considered to provide excellent discrimination between cases and controls [23]; however, owing to the extremely low incidence of sudden death, the positive predictive value of the model is very low, and as such, the model should not be regarded as a predictive model. Rather, it is intended to help generate hypotheses regarding the pathogenesis of sudden death in racehorses.

The model produced in this study represents the effects of the retained variables, having controlled for other variables that were significant at univariable level. For example, the odds of sudden death associated with a steeplechase race are reduced from 12.87 in the univariable analysis to 2.26 in the multivariable analysis. This reflects confounding by other variables, such as distance (steeplechase races are longer than other race types), but shows that even once distance is controlled for there is an increased risk of sudden death in a steeplechase race. The precise mechanisms by which these retained variables contribute to death are unclear, and the following discussion provides a speculative rather than proven explanation for the findings. It is hoped that this speculation may lead to development of hypothesis-led studies to further our understanding of the mechanisms of sudden death on UK racecourses.

There are no large published post mortem studies of sudden death cases in the UK, and post mortem data were not available for horses in this study. In other racing populations around the world, the following 2 conditions have been identified as the major definitive causes of sudden death: pulmonary haemorrhage (19–82% of cases) [2,25–27] and idiopathic blood vessel rupture (9–24% of cases) [27–29]. (It is important to note that pulmonary haemorrhage is commonly identified during post mortem examination of sudden death cases, but death is only attributed to this finding in some cases [27].) Importantly, though, the cause of death often remains unknown in a significant number of cases (20–68%), despite thorough post mortem examination [2,25,27–30]. The most likely explanation for cases in which a definitive diagnosis is not reached is a fatal cardiac arrhythmia, although this remains unproved. Assuming similar pathology in cases of sudden death in the UK from 2000 to 2007 compared with racing populations in these previous post mortem studies, the risk factors identified in this study are likely to be risk factors for cases with negative post mortem findings (probably fatal cardiac arrhythmia) or cases in which death is attributed to pulmonary haemorrhage or vascular rupture.

The odds of sudden death increased 1.3-fold with every 1 km increase in race length. In this study, 35% (70/201) of sudden deaths occurred after the race, and in a previous study 43% (114/268) of sudden deaths occurred in the post exercise period [27]. In addition, preliminary data from another study involving observation of race videos reveal that sudden deaths that occur during races occur towards the end of the race (C. H. Lyle, unpublished observation). These observations illustrate that most sudden deaths occur after completing or almost completing the full race distance. Increased exposure time at risk during longer races may therefore contribute to this finding.

Inconsistent with exercise-induced pulmonary haemorrhage (EIPH) as a cause of death is that epistaxis/EIPH have been more commonly identified after shorter races (<1400 m) [31] than longer races (<1600 m) [32]. It should be noted, however, that investigations of EIPH have had variable study design, with some studies using the presence of epistaxis as a diagnosis of EIPH, while other studies have used tracheobronchoscopy for diagnosis of EIPH, and this will have affected the results of these studies. In discussing these studies, we have used the term ‘epistaxis’ if the diagnosis of EIPH was made on the observation of epistaxis and the term ‘EIPH’ if the diagnosis of EIPH was made using tracheobronschoscopy.

Increased race length may cause more severe metabolic derangements, which may contribute to the development of fatal ventricular arrhythmias. Exercise-related arrhythmias are poorly understood in horses, but ventricular premature complexes are commonly identified immediately post exercise in apparently healthy horses [33,34]. Such ventricular premature complexes could initiate fatal ventricular fibrillation, and the immediate post exercise period is well recognised as a risk period of sudden death in people [35]. During intense exercise in horses, there is a decrease in blood pH, increase in plasma potassium [36] and increase in circulating catecholamines [37]. If any of these metabolic derangements were to occur alone there would be an increased risk of fatal cardiac arrhythmia [35]. However, these metabolic changes are well tolerated during exercise because the combination of these metabolic derangements results in mutual antagonism; the catecholamines offset the harmful effects of acidosis and hyperkalaemia and vice versa[35]. Immediately after intense exercise, however, potassium concentrations fall, while catecholamine concentrations remain elevated, leaving the heart vulnerable to catecholamine-induced arrhythmogenesis [35]. In addition, development of exercise-associated arrhythmias has been associated with a tendency for a higher blood lactate value [38], which is likely to occur in longer races.

The odds of sudden death increased 1.3-fold with every 1 year increase in age. Unfortunately, it is impossible to separate the effect of age from time in training, because there were no untrained controls and training data were not available for investigation. Cardiac electrophysiological changes are well recognised as a result of cardiac hypertrophy secondary to athletic training in humans, with frequent and complex ventricular ectopy being common in human athletes [39]. Cardiac remodelling (hypertrophy) has been identified in horses in athletic training [40–42]. The increased risk of sudden death with age may therefore be partly explained by increased time in training, as it is possible that cardiac hypertrophy may also lead to increased risk of fatal cardiac arrhythmia in horses; however, age was not associated with development of exercise-related arrhythmias in one treadmill study [38].

Exercise-associated epistaxis has been reported to be more common in older horses in some studies [32,43], and increased risk of EIPH may therefore contribute to the increased risk of sudden death associated with increased age. Conversely, though, some other studies have not associated epistaxis/EIPH with increased age [31,44]. The recurrence rate of epistaxis has been estimated at 13.2% [32], and it has been suggested that some cases of sudden death may be fulminant cases of EIPH [26]. Recently, regional pulmonary veno-occlusive remodelling has been identified in EIPH, although it remains unclear whether this pathology relates to cause or effect [45]. Accumulation of pathology, such as veno-occlusive remodelling, with age could explain the increase in incidence of EIPH associated with age in some studies.

Likewise, accumulation of microscopic cardiac pathology, such as cardiac fibrosis, with age may increase the risk of development of fatal cardiac arrhythmia. Ventricular fibrillation is more likely to develop when acute cardiac ischaemia is superimposed on a healed infarct. This has been used to produce animal models to study the pathophysiology of cardiac arrhythmias [35]. In addition, the efficacy of mutual antagonism is reduced when hyperkalaemia, acidosis and increased noradrenalin are superimposed upon a heart with regional ischaemia or a small infarct [35]. Gross and microscopic myocardial fibrosis have been reported as an incidental finding in horses at post mortem[46–48], but no association with age was detected in these studies. Currently, neither the role of myocardial fibrosis in arrhythmogenesis in the horse nor the true prevalence of this lesion in any age group is known.

Arterial wall fibrosis has been associated with peripartum uterine artery rupture in mares, and this condition has been associated with increased age [49]. Similar age-related degenerative changes may occur in other blood vessels and predispose to arterial rupture, described as the cause of sudden death in some cases [27]. Calcification of the tunica media of the large arteries has been reported as a common finding in young Thoroughbred racehorses (mean age 3.84 ± 0.30 years), particularly affecting the pulmonary arterial branches [50]. It has been suggested that this may lead to rupture of these vessels due to reduced vascular compliance. Although pulmonary haemorrhage is commonly described in post mortem examination of sudden death cases, rupture of the pulmonary arteries is rarely reported. In cases of sudden death attributed to vascular rupture, abdominal vessels have been reported to be more common, and there is often no pathology associated with the site of rupture [27].

The number of career starts was not associated with increased risk of sudden death, either at univariable level or when forced into the final multivariable model. This suggests that increased exposure time at risk did not contribute to the increased risk associated with age. A limitation of the present study was the lack of data on exercise during training. It is possible that exposure time during racing is too crude a measure of overall exercise exposure to demonstrate an association with sudden death. Further investigation of total exercise exposure time (in both racing and training) would be justified in future studies.

Increased odds of sudden death were observed in hurdle and steeplechase races compared with flat races. Horses competing in steeplechase races have previously been identified to be at greater risk of epistaxis [32,44], suggesting an increased prevalence of EIPH in steeplechasers. This has been attributed to increased locomotory impact-induced trauma [44]. Jump races are likely to be more strenuous than flat races, so hurdlers and steeplechasers may develop more marked metabolic derangements. This may contribute to development of arrhythmias as described earlier. In addition, heart rate is more likely to fluctuate during jump races than flat races. Changes in heart rate may be associated with increased propensity for ventricular premature complexes due to changes in autonomic tone [33]. Larger heart size in this group compared with horses racing in flat races [51] may also predispose to ventricular fibrillation, although type of racing was not associated with development of cardiac arrhythmias in one treadmill study [38].

Misclassification of traumatic catastrophic injuries, such as cervical vertebral fractures, as sudden death cases may also have contributed to the increased odds in this group, because cases were not defined at post mortem level. Given the unexpected and rapid nature of some sudden deaths and some traumatic accidents, it can be difficult to determine accurately whether the collapse was traumatic or nontraumatic.

Increased odds of sudden death were observed in National Hunt flat races compared with conventional flat races. Horses that have not raced on the flat are allowed to compete in up to 3 National Hunt flat races before racing over jumps. These young horses may have increased circulating levels of catecholamines secondary to their inexperience and possible psychological stress. Psychological stress has been proposed as a risk for development of post exercise arrhythmia in horses [33], and increased catecholamine levels may predispose horses to fatal arrhythmias [35]. Larger heart size in this group compared with horses racing in conventional flat races [51] may also predispose to fibrillation. In addition, these horses may be less fit and therefore predisposed to fatal arrhythmias. High vagal tone secondary to training confers protection against ventricular fibrillation [52–54]. The mechanism of vagal protection against the proarrhythmic effects of high sympathetic tone and hypokalaemia is unknown but may occur through antagonism of calcium channels [35]. Genetic ion channelopathies (e.g. long Q-T syndrome) and accessory pathways that cause sudden death in young human athletes [55] have not been identified in horses. However, if such conditions exist it is likely that they would manifest early in the horse's career, and this may explain the increased risk for sudden death in this group of horses. Conversely, there was no evidence of clustering within sire, which might be expected if there was a genetic predisposition.

Racing in the summer was associated with increased odds of sudden death. This may reflect more severe acid-base and electrolyte derangements associated with sweating in warmer weather and so increased risk for fatal cardiac arrhythmia. Conversely, racing in the spring has been associated with increased risk of epistaxis [44], but that study did not take into account occult cases of EIPH that had no epistaxis.

Increased frequency of starts prior to the index start was protective at univariable and multivariable level for all time periods investigated (30, 60, 90, 180 and 365 days prior to the start). ‘Starts in the last 60 days’ was used in the final model. Horses racing more frequently prior to the race start are likely to be fitter, and these variables may therefore be proxies for athletic fitness and health (i.e. absence of illnesses that would have resulted in time off). As described earlier, less fit horses may be at increased risk of fatal cardiac arrhythmia. This hypothesis may contradict the previous suggestion that exercise-induced hypertrophy may predispose horses to arrhythmias, but it is possible that both long-term training (as measured by age) and short-term fitness (as measured by starts in the previous 30–365 days) have an effect, given the inclusion of both these variables (age and starts in the last 60 days) in the final model.

Previous risk factor studies for racecourse incidents that have used fatality as an outcome are heavily biased by cases of catastrophic musculoskeletal injury [1,18–21] but have identified similar risk factors to those identified in this risk factor study for fatality due to sudden death alone.

Similar to the findings in this study, increased race length [18,20] and increased age [20,21] have been associated with increased risk for fatality. Previous studies have also identified increased risk of sudden death in jump races compared with flat races [1,21], in steeplechases compared with hurdle races [19] and in National Hunt flat races compared with conventional flat races [21]. Contrary to our study, Henley et al. [20] identified slightly increased risk in hurdle races compared with steeplechase races. Racing in February to May was identified as a risk factor for fatality in one study [20], whereas racing in summer (June–August) was identified as a risk factor in the present study. In previous studies, racing on firmer going was associated with increased odds of fatality compared with racing on softer ground in all race types [20,21] and in flat starts [18], but was not statistically significant in jump starts [19]. Going was not significant at multivariable level in the present study.

Previous studies investigating fatality have identified more starts in the 60 days prior to the race as contributory in flat starts [18] but protective in jump races [19]. Henley et al. [20] reported that the risk of fatality was decreased with increased number of starts in the previous 12 months. In studies investigating fatal and nonfatal musculoskeletal injury, increased frequency of high-speed exercise in the 30 or 60 days prior to the race has been identified as detrimental in some studies [6,56], protective in others [3,57] and nonsignificant in others [13,15]. Studies in which more frequent high-speed exercise prior to the race has been protective have attributed this to the ‘healthy horse effect’ or survival bias, in which healthier horses are more likely to be racing [3,19,57]. Studies in which more frequent high-speed exercise prior to the race has been detrimental have attributed this to accumulation of pathology (bone and soft tissue damage) that is incompletely repaired before the next start [5,6,18].

Increased age, racing in hurdle or steeplechase races, hard going, female sex, racing over shorter distances and colder air temperatures have all been associated with increased risk of epistaxis or EIPH [31,32,43,44]; however, the study design used in these investigations has been very variable and some of the data are conflicting. Further investigation into EIPH events and sudden death within the same database is warranted given the high prevalence of pulmonary haemorrhage in sudden death cases shown in post mortem studies [2,25–28,58].

As with most risk factor studies, the value of this model is limited, because it has not been validated on another data set. Ideally, future studies would be implemented to validate this model on new data. In any case, the number of sudden deaths that could be prevented by removal of risk factors and implementation of protective factors is likely to be minimal due to the low incidence. However, as the risk factors for sudden death and fatality as a whole appear similar, such measures may be beneficial in reduction of overall fatality. There appears to be conflicting evidence as to whether racing intensity is protective or detrimental regarding overall fatality risk, as well as both fatal and nonfatal musculoskeletal injury risk, so further investigation is warranted. As the frequency of fatal and nonfatal musculoskeletal injuries is much greater than the frequency of sudden death [1,20,21], implementation of a common factor that reduces the odds for sudden death but increases the odds for catastrophic musculoskeletal injury would be counterproductive.

In conclusion, the risk factors identified in this study were not uniquely associated with sudden death and have also been identified in studies using fatality as the outcome. These data suggest that a generic approach to reduce fatal musculoskeletal injury and sudden death may be possible. The identification of risk factors allows speculation on the underlying mechanisms of sudden death in racing. It is hoped that this will stimulate hypothesis-led investigations into the pathogenesis of exercise-related arrhythmias, EIPH and blood vessel rupture.

Authors' declaration of interests

No conflicts of interest have been declared.

Sources of funding

The Horserace Betting Levy Board funded C.H.L.'s senior clinical training scholarship at the University of Edinburgh.

Acknowledgements

Data were supplied by the British Horseracing Authority.

Manufacturers' addresses

a Microsoft Corp., Redmond, Washington, USA.

b StataCorp, College Station, Texas, USA.

Ancillary