Reasons for performing study: Fitness assessment can be challenging. The use of global positioning systems (GPS) with heart rate (HR) monitors has been promising; however, evaluation of speed parameters during training has not been reported.
Objectives: To evaluate speed indices during training in Thoroughbreds using a GPS-HR monitor.
Methods: Thoroughbreds (n = 102) were assessed during training with data collected each work day (WD; sprinting). Speed indices evaluated included maximal velocity (Vmax), duration at Vmax (Vmaxt), acceleration rate (m/s2) from 800 m to Vmax (Acc800-Vmax), the distance (m) 6 (VmaxD6) and 12 (VmaxD12) s before (acceleration [a]) and after (deceleration [d]) Vmax and the deceleration rate from Vmax to the finish (VmaxDFd). Blood for plasma lactate ([LA]) and creatine kinase ([CK]) measurements were taken before (T0), 5 mins (T1) and 6 h after exercise (T2). WD accumulation, jockey, gallop condition, horse gender, age, total distance covered (DistT), maximum HR (HRmax), velocity at 200 beats/min (V200) and velocity at maximum HR (VHRmax) for each WD were evaluated for associations with [LA], [CK], speed indices and racing performance. Data were analysed by repeated measures ANOVA with P<0.05 significant.
Results: No speed parameter clearly changed with training. Gallop condition affected Vmax, Vmaxt and all distances covered with Vmax and distances increasing and Vmaxt decreasing as gallop surface became firmer. Jockey influenced Vmax, VmaxD6a and all decelerations, while DistT was inversely associated with Acc800-Vmax, HRmax and V200 and positively associated with Vmax, all accelerations and decelerations. [LA] at T1 was positively associated with DistT and VmaxDFd.
Conclusions: Speed parameters did not change with training but were affected by jockey, gallop condition and exercise distance. This information may help to modify training to maximise fitness, minimise injury and choose distances best suited for individuals.
The goal of any training programme is to obtain and maintain peak fitness without injury while maximising performance. Fitness and performance assessment can be challenging for racehorse trainers and typically focuses on the horse's speed measured either subjectively by the jockey or more objectively by use of a stopwatch. However, the introduction of global positioning systems (GPS) has allowed much more specific and sensitive analysis of speed responses to training programmes. The use of GPS with heart rate (HR) monitors is commonly used in man (Larsson 2003) during training and competition to measure velocity, acceleration rate and distance covered. In equine studies the use of the GPS has been used to estimate exercise measuring speed (Curtis et al. 1997; Kusunose and Takahashi 2002; Spence et al. 2008) and to evaluate locomotion and stride parameters (Witte et al. 2006; Parsons et al. 2008). When used in conjunction with HR monitors, GPS technology has been shown to be promising for the assessment of fitness (Kingston et al. 2006; Vermeulen and Evans 2006) and performance (Gramkow and Evans 2006).
Identifying physiological factors by which to assess equine athletic potential has previously been attempted by a number of researchers and includes measurements of muscle fibre type and enzyme activity (Roneus et al. 1999), blood lactate concentrations, heart rate and oxygen consumption (Bayly et al. 1987; Harkins et al. 1993). Among these, the association of blood lactate concentrations with performance has particularly been investigated (Evans et al. 1993; Harkins et al. 1993) with some researchers reporting a correlation between speed and blood lactate (i.e. the fastest horses have the highest blood lactate concentrations) (Bayly et al. 1987; Harkins et al. 1993) and others finding no correlation (Roneus et al. 1999). The definition of performance evaluation itself has also varied between researchers and is often defined based on a single factor such as total race earnings, earnings per start or timeform ratings.
To date, the majority of equine exercise studies utilising GPS-HR technology appear to have focused on HR-to-speed relationships, exploring the effect of training on HR parameters including V200 (Kobayashi et al. 1999) and VHRmax (Gramkow andEvans 2006; Vermeulen and Evans 2006) as well as Vmax (Gramkow and Evans 2006); however, no studies have specifically evaluated different speed indices during training nor have any of the studies investigated the association of these speed parameters with physiological parameters and performance ratings. Therefore, the aims of the present study were to identify and evaluate how various speed indices vary during the training of a group of Thoroughbred flat racehorses through the use of a GPS-HR monitoring system and to investigate the relationship of these speed indices with plasma lactate and creatine kinase concentrations and performance outcomes.
Materials and methods
The study was approved by University College Dublin Animal Research Ethics Committee. A total of 102 Thoroughbred flat racehorses (65 two-year-olds [21 males and 44 females], 37 three-year-olds [12 males and 25 females]) from a single training stable were assessed during training (March–November) between 2007 and 2008. All horses were maintained and managed in the same fashion with regard to nutrition, housing and health care, with veterinary records for each animal maintained while in the yard documenting any illness and subsequent treatment. All horses were judged daily to be healthy and free of any musculoskeletal problems by physical examination prior to training.
All horses were trained 6 days per week. The training programme consisted of 4 progressive stages. Horses were initially placed onto a walker and worked on long reins; then when broken to be ridden, they were taken out for ‘slow’ workouts consisting of trotting and cantering with ‘fast’ workouts (i.e. work days [WD], a simulation of a race where horses run against each other) gradually introduced as training progressed. For most horses, the initial ‘slow’ workout distance was 600 m, gradually increasing to a distance of 1000 m. Work days generally consisted of distances between 800 and 1000 m. ‘Fast’ and ‘slow’ workout days were alternated throughout the training period with training modified and adapted to an individual animal as required based on that animal's soundness, fitness and aptitude. Following the onset of WD, horses were entered into competitive races dependent on their perceived fitness and performance. All decisions on the training and racing schedule were made by a single trainer.
All of the horses in the study trained on an outdoor all-weather gallop with a woodchip surface. The gallop was 1500 m in length and 6 m wide with both sides surrounded by natural vegetation approximately 3 m high providing wind protection. The gallop was located on a hill and for the final 800 m consisted of a straight line with a 2.7% incline. Every morning the gallop surface was levelled with a tractor.
Data were only recorded for horses undergoing a WD (Fig 1) which consisted of the following:
1Horses initially warmed-up on a horse walker for 30–60 min (walking and trotting).
2Horses were then walked under saddle for 5–10 min in the parade ring of the main yard.
3On the way to the bottom of the gallop the horses walked for 300 m, trotted for 700 m and then walked for 100–300 m (depending on the distance of the sprint test).
4Horses then turned around and galloped with velocities reaching maximal intensity for 800–1000 m.
Prior to going onto the horse walker, each horse was instrumented with a HR telemetry system (Polar Equine S810i)1. Each jockey was given a hand-sized GPS unit (SPI10)2 which was either maintained in their left pocket or secured to the left side of their belt. After data collection and at the end of each day the GPS and HR data were downloaded to an equine-specific software program (Race watch Software). Both the GPS and HR units recorded data on a per second basis, with a distance accuracy of 0.5% for distances greater than 1000 m (as per GPS manufacturer guidelines); a speed accuracy of 0.4 m/s had also been previously determined for a similar GPS unit (Kurzawa 2008). The GPS unit recorded variables including speed, time, distance and altitude as well as the exact map of each horse's exercise (Fig 2). Prior to onset of the study, the entire gallop had been prerecorded using one of the GPS units via driving the entire length, with speeds of the truck correlating well with speeds recorded by the GPS unit. The prerecorded map of the gallops also allowed specific location of the distances that the horses were exercising and were the basis from which a multitude of speed parameters were selected (Fig 3).
Blood was collected from the jugular vein prior to each horse going onto the walker (T0) and 5 min (T1) and 6 h (T2) after exercise and placed into fluoride oxalate and lithium heparin tubes for the measurement of plasma lactate ([LA]) and creatine kinase ([CK]) concentrations, respectively. Once collected, the tubes were immediately placed onto ice until centrifuged with the plasma separated and frozen at −80°C until samples could be analysed as a batch3.
Data were only included from horses that had at least 2 satisfactory GPS recordings of WD. [LA] and [CK] were only included when paired with a satisfactory GPS recording of WD. The total number of WD and races for each horse were recorded during the year. For each WD, the distance exercised, the jockey and gallop condition were recorded. Horses were ridden by one of 5 jockeys selected by the trainer; depending on the jockey's weight, a laid bag was added in order to equalise the weight carried between horses. Gallop condition was subjectively graded by one observer (RGF) on each WD from 1–4, with grade 1 representing the firmest and grade 4 the softest gallop condition. All speed measurements were recorded from a distance of 800 m from the finish line as the total distance exercised on a WD differed slightly for each horse. Speed indices evaluated included maximal velocity (Vmax), duration at Vmax (Vmaxt), acceleration rate (m/s2) from 800 m to Vmax (Acc800-Vmax), the distance (m) 6 (VmaxD6) and 12 (VmaxD12) seconds before (acceleration [a]) and after (deceleration [d]) Vmax and the deceleration rate from Vmax to the finish (VmaxDFd). Acceleration and deceleration rates were calculated using the formula (Newton's first law):
Vi represented the initial velocity (m/s), Vf the final velocity (m/s) and T the total time (s) from start to finish.
In addition to jockey, gallop condition and WD, the horse gender, age, total sprint distance exercised on the day (DistT), maximum HR (HRmax), velocity at 200 beats/min (V200) and velocity at maximum HR (VHRmax) were evaluated for their influence on speed indices. To better evaluate Vmaxt, Vmax zones were created and included: 14–14.4 m/s (zone 1 [Z1]), 14.4–15.1 m/s (Z2), 15.1–15.6 m/s (Z3), 15.6–16.1 m/s (Z4), 16.1–16.6 m/s (Z5) and 16.6–17.4 m/s (Z6). These zones represented ranges of speeds since the Vmax tended to waver 1–2 m/s for each horse and allowed a clearer idea of the time spent at Vmax. V200 and VHRmax were calculated from HR vs. speed linear regression equations (GraphPad Prism)4 using criteria previously described (Vermeulen and Evans 2006).
To more clearly evaluate the effect of training (i.e. the association between WD and the speed indices), the training time was divided into 2 month intervals (period [P]) with the number of WD and races recorded for each period. Period one (P1) began when each horse actually started training. A variable denoted as work days nested within period was created to assess the effect of the number of workdays within a period on outcome variables of interest. Therefore, P1(WD) was the summation of WD/races for the first 2 months of training and P2(WD), P3(WD) and P4(WD) the summation of WD/races for the second, third and fourth 2 months training intervals, respectively.
Performance was characterised for each horse using the system described by Hill et al. (2010) and was based on retrospective racecourse performance records (http://www.racingpost.com/ or http://www.racingpost.co.uk). Horses were categorised as ‘Thoroughbred-elite’ (TBE) if they had won at least one Group/Stakes and/or listed race and as ‘Thoroughbred-other’ (TBO) if they had never won a Flat race or had a handicap rating (Racing Post Rating [RPR]) of ≤89. According to the distance of the best race won by horses categorised as TBE, horses were subsequently subdivided into ‘sprint’ (≤8 furlongs) or ‘long distance’ (>8 furlongs) groups.
Data processing and analysis
All data were analysed using Statistical Analysis Systems software (SAS)5. Data were examined for adherence to normality using UNIVARIATE procedure of SAS. Data not normal were transformed by raising the variable to the power of lambda with the appropriate lambda value obtained by conducting a Box-Cox transformation analysis in the TRANSREG procedure in SAS. The transformed data were used to calculate P values; however, the corresponding least squares means and standard errors of the nontransformed data are presented in the results for clarity. A mixed model ANOVA (PROC MIXED) was conducted to determine the effect of fixed effects (such as gallop condition, age, horse gender) and corrected for random effects (such as jockey) on outcome variables of interest. In all analyses the individual animal was denoted as the experimental unit. The statistical models included terms for fixed effects and covariates and their interaction where appropriate. The effect of level of training was accounted for by the inclusion of the nested effect of WD within period which was indicative of training intensity within period. Interaction terms, if not statistically significant (P>0.05), were subsequently excluded from the final model. The Tukey critical difference test was used to determine statistical difference between mean comparative group values for each outcome variables. Orthogonal contrasts were used to examine the trajectory of the effect of a number of scaled variables (i.e. gallop condition) on outcome variables of interest. For binary variables such as TBE or TBO classification or sprinting ability (yes/no) logistic regression was conducted using the LOGISTIC procedure of SAS. Statistical differences were denoted at P<0.05, with trends towards statistical significance denoted as P<0.10. All values are expressed as means ± standard error of the mean (s.e.m.).
Performance prior to achieving Vmax
Acc800-Vmax was not affected by training but was affected by DistT (P<0.0001; Acc800-Vmax decreased as DistT increased), gallop condition (P = 0.0002; Acc800-Vmax increased as gallop condition became firmer), jockey (P = 0.04) and age (P<0.0001; Acc800-Vmax was greater for 2-year-olds than 3-year-olds). VmaxD12a was affected by training (P = 0.002; VmaxD12a was similar for periods 1 and 2, increased in period 3 and remained constant across periods 3–5 in a cubic fashion) and was also affected by DistT (P = 0.04; VmaxD12a increased when DistT increased) and gallop condition (Table 1, P<0.0001; VmaxD12a increased as gallop condition became firmer). VmaxD6a was influenced by training (P = 0.009) although there was no clear pattern evident with both linear (P = 0.03) and quadratic (P = 0.01) components detected for the response. VmaxD6a was also affected by the gallop condition (Table 1, P<0.0001; VmaxD6a increased when gallop condition became firmer) and jockey (P = 0.04).
Table 1. Effect of gallop condition on mean values (± s.e.) for maximal velocity (Vmax), duration at Vmax (Vmaxt), distance from 800 m to Vmax (Dist800-Vmax), distance 12 s (VmaxD12a) and 6 s before Vmax (VmaxD6a) and distance 12 s (VmaxD12b) and 6 s after Vmax (VmaxD6d). Gallop conditions (G) are shown with G1 the firmest and G4 the softest gallop surface
16.5 ± 0.7
16.4 ± 0.7
16.2 ± 0.5
15.4 ± 0.1
2.6 ± 0.1
2.5 ± 0.2
2.6 ± 0.2
3.3 ± 0.2
186.8 ± 0.7
186.7 ± 0.8
183.9 ± 0.8
177.4 ± 0.7
96.6 ± 0.5
96.3 ± 0.6
95.9 ± 0.6
89.9 ± 1.0
98.4 ± 0.3
97.6 ± 0.4
96.5 ± 0.4
93.3 ± 0.7
192.2 ± 0.8
190.1 ± 0.8
187.7 ± 0.8
183.2 ± 1.6
Performance at Vmax
Vmax was not affected by training but was affected by DistT (P = 0.03; Vmax increased as DistT increased), jockey (P = 0.02) and gallop condition (Table 1, P<0.0001; Vmax increased as gallop condition became firmer). Vmaxt was not affected by any variable measured.
Performance following achieving Vmax
VmaxD6d did not change following training but was associated with DistT (P = 0.003; VmaxD6d increased as DistT increased), gallop condition (Table 1, P<0.0001; VmaxD6d increased as gallop condition became firmer) and jockey (P = 0.0001). VmaxD12d was not affected by training but was affected by DistT (P<0.0001; VmaxD12d increased as DistT increased), gallop condition (Table 1, P = 0.005; VmaxD12d increased as gallop condition became more firm) and jockey (P<0.0001). VmaxDF did not change following training but was affected by gallop condition (Table 1, P = 0.03; VmaxDF decreased as gallop condition became firmer), age (P = 0.0003; VmaxDF was greater for 2-year-olds), jockey (P<0.0001) and DistT (P<0.0001; VmaxDF increased as DistT increased).
Measurements associated with HR
Of the 102 horses measured, accurate HRmax recordings were obtained for 80 horses (46 two-year-olds [17 males and 29 females] and 34 three-year-olds [10 males and 24 females]) with 3.5 ± 2.2 valid HRmax readings for each horse. Accurate V200 and VHRmax readings were obtained for 55 horses (25 two-year-olds [9 males and 16 females] and 30 three-year-olds [7 males and 23 females]) with 3.9 ± 2.2 valid calculations for each horse. HRmax decreased as DistT increased in a cubic fashion (P = 0.002) while V200 was greater for fillies (10.6 ± 0.2 m/s) than for colts (10.1 ± 0.3 m/s; P = 0.03).
Measurements associated with [LA] and [CK]
[LA] at T1 was positively affected by DistT (P = 0.02; [LA] increased as DistT increased) with mean [LA] 26.1 ± 1.6 mmol/l, 30.6 ± 0.7 mmol/l and 30.7 ± 0.5 mmol/l for 800, 900 and 1000 m, respectively. [LA] at T1 was also positively associated with VmaxDFd (P<0.0001; [LA] increased when VmaxDFd increased). [CK] at T2 was greater for fillies (227.4 ± 49.4 µmol/l) than colts (126.1 ± 75.5 µmol/l; P = 0.03). However, when the differences between the pre- and post exercise concentrations for [LA] and [CK] were evaluated, no associations were found with any of the speed indices.
Measurements associated with performance outcome
Apart from the relationship between Acc800-Vmax and sprint ability (P = 0.03) no other speed indices were significantly associated with performance evaluation. There were no associations with [LA] or [CK] with performance evaluation.
Desired performance characteristics for a flat racehorse include acceleration rate, speed and the ability to maintain maximum speed over a certain distance. We hypothesised that speed parameters reflecting these characteristics may change with training, so the primary objective of the present study was to investigate the effects of training on various speed indices in a group of flat racehorses, as well as to evaluate for any associations these speed indices may have had with various factors normally present under typical training conditions. With the advancement of field technologies such as GPS systems, the ability to investigate these types of parameters in an environment similar to what the horse will be asked to perform under have become increasingly feasible.
In the current study, both Vmax and Vmaxt did not change with training which is in agreement with another study evaluating National Hunt racehorses where Vmax was also found not to change following training (Ferrari et al. 2009). Although the horses included in the study were all placed into appropriate training periods (i.e. P1, P2, etc) for analysis and the majority of horses had had few WD prior to the first GPS reading, it is possible that the amount of submaximal training prior to recordings did influence Vmax resulting in no apparent training associations. In the present study some acceleration rates (expressed as the distance covered over a set amount of time prior to reaching Vmax) did change following training although this could not be clearly defined. This contradicts findings from previous studies evaluating untrained human athletes exercising for 6 weeks in which acceleration rates clearly improved following training (Bloomfield et al. 2007); however, in the present study, acceleration may have been influenced by the jockey (holding the horse back) under specific training strategies put in place by the trainer, as the horses were all exercised in groups of 2 and 3. It has been previously demonstrated in both man and horses that the efficiency of locomotion improves following training (Witte et al. 2006; Bloomfield et al. 2007), so although the acceleration rates for the horses in the current study did not improve following training their stride may have become more efficient. Although previous studies have shown significant training effects on gait and stride parameters in horses (Pfau et al. 2006; Witte et al. 2006; Parsons et al. 2008; Ferrari et al. 2009), with stride frequency increasing and protraction time decreasing at high speed following training (Parsons et al. 2008; Ferrari et al. 2009) to date there currently have been no studies evaluating the effects of additional variables such as jockey and gallop condition on measured speed parameters. Results form the present study suggest that all the extra uncontrolled variables like gallop condition, training strategy, jockey, etc. might confound the response to training and make it more difficult to accurately detect differences in the various speed indices.
It has been previously found in horses that males achieve higher speeds and heart rates than females (Mukai et al. 2003). These results are in contrast with our study, although a number of differences existed between the sample population and regime employed in our study compared with that of Mukai et al. (2003). For example, the sample size used in their study was much lower (20 horses), the horses were well-trained, the exercise test was submaximal in intensity and the HR measurements used were following a previous exercise test, when HR were stabilised. The effect of age was also not significantly associated with the majority of the speed parameters in the current study, although 2-year-olds were found to decelerate more rapidly than 3-year-olds: this may possibly be due to greater immaturity and fatigue of the 2-year-olds.
Although there has been one previous report evaluating the effect of rider weight on performance in which no association was found between rider weight and V200 (Kobayashi et al. 1999), as far as the authors are aware no other studies have evaluated the effect of factors such as gallop condition, jockey influence and total distance exercised on measured parameters of fitness. The results of the present study confirm the importance of taking these types of variables into consideration when assessing training responses. In the current study, variables such as gallop condition and jockey were chosen to be evaluated based on what a typical flat racehorse trainer would face in realistic conditions. Furthermore, it was considered important to evaluate these variables since the training environment in the current study could not be tightly controlled as would be the case with a treadmill investigation. In the present study, jockey, gallop condition and distance travelled were all shown to significantly impact on the various speed indices measured.
Gallop condition has been empirically considered to have a crucial effect on the performance of a flat racehorse. The current study supports this supposition with the results showing that gallop condition had a significant effect on the majority of the measured speed parameters, with firmer ground contributing towards increased speeds and accelerations. Gallop conditions have also been previously found to affect gait parameters (Ratzlaff et al. 2005) and have been strongly associated with the incidence of catastrophic injuries (Mohammed et al. 1991; Oikawa and Kusunose 2005), even during routine maintenance of the track (Peterson and McIlwraith 2008) so it is important to consider gallop conditions when attempting to minimise injury and maximise performance. Total distance exercised by the horse is also important to consider. In the current study, horses were trained over different distances with longer distances associated with decreasing Vmaxt. This could be explained by animal fatigue and potentially be used as an indicator to select distances best suited for individuals.
In our study, laid bags were used in order to equalise the amount of weight carried between horses during training resulting in the rider weight being a controlled factor whereas jockey influence (i.e. riding style, experience, capability, etc.) was not controlled and subsequently found to have significant influences on measured parameters. We found that jockey had a significant influence on most of the speed parameters, including Vmax and Vmaxt. This is not unexpected as the jockey's attitude on adjusting stimulus as well as keeping the horse's balance is essential for the animals' performance. Furthermore, a fitter and/or more experienced jockey may know how to better assist the horse with minimal interference on the horse's equilibrium.
In the current study, V200 was found to be greater for fillies than for colts, which is in disagreement with previous studies (Ohmura et al. 2002; Mukai et al. 2003). It is possible that in the current study fillies had greater V200 measurements than colts simply due to differences in rider or gallop condition rather than actual physiological differences. However, it is also likely that anticipation by the horses and in particular the fillies, in the current study prior to performing the sprint test, resulted in higher pre-sprint HRs affecting the calculation of the V200 since it has been previously demonstrated that fillies tend to be more nervous than males (Kusunose 1997) and that when a horse is exercised in a relaxed and non-excitable state, V200 is subsequently lower than if the animal were in an excitable state (Kobayashi et al. 1999).
In addition to V200, previous field studies have demonstrated VHRmax to be a reliable measure of fitness (Vermeulen and Evans 2006), although the present study did not find this to be the case. One possible explanation could be the potentially confounding effects of gallop condition, jockey and distance exercised on performance parameters resulting in a masking of the effect of training on these variables. VHRmax has also been previously shown to be a reliable measure of performance potential based on total earnings (Gramkow and Evans 2006) although again in the present study this was not found to be the case. In the present study, the classification of horses as TBE or TBO was based on retrospective racecourse performance records and did not take into consideration the various factors that were found to affect measured parameters during training (i.e. gallop condition, jockey, total distance exercised) making comparison of VHRmax and performance outcomes difficult. Future studies would need to standardise the above-mentioned variables to adequately compare measured parameters such as VHRmax to performance outcomes.
In the present study, plasma [LA] at T1 increased as the total distance exercised increased which is likely due to the fact that the horses in the current study were sprinting and therefore as the distance travelled increased the reliance on anaerobic metabolism also increased, resulting in increasing accumulation of plasma [LA]. Distance exercised has been previously found to significantly influence [LA] up to distances of 1600 m (Kubo et al. 1984) but not for distances greater than 1760 m (Evans et al. 2002); it is difficult to completely compare the findings from the current study and previous reports since the horses in the present report were performing a sprint test rather than a submaximal exercise evaluation. In addition, as horses in the current study covered a greater distance between Vmax and the finish line (i.e. VmaxDFd) meaning that they achieved Vmax earlier, plasma [LA] at T1 increased in a linear fashion with VmaxDFd likely reflecting the work intensity associated with the acceleration rate required to obtain Vmax during the sprinting. There was no other association of plasma [LA] or [CK] found with training or performance parameters.
In conclusion, we have found that the use of GPS technology in a training programme may allow a more objective and non-invasive evaluation of training responses. The speed variables investigated in the current study corrected for factors that may independently affect the interpretation of these speed indices may provide insight into training responses of flat racehorses. Combining these speed and distance indices with evaluation of locomotion through the use of accelerometer data has the potential to allow gait analysis of racehorses to be carried out under high speed, real-world conditions. Future field studies should include the effect of jockey, gallop condition and distance exercised when investigating locomotion parameters as well as associating speed profiles with the occurrence of injuries.
Funding for this project was provided by the Research Development Fund, University College Dublin. The authors wish to express their appreciation to the trainer Mr Jim Bolger and his staff and, in particular, Mr Pat O'Donovan and Mr Brian O'Connor.
Conflict of interest
The authors have no potential conflicts to declare.