Study design and cohort description
Fifteen trainers were contacted regarding study participation, and twelve agreed to participate. One additional trainer was recruited based on advertisement of the study. Training yard-level inclusion criteria were location (proximity to Oslo, Norway or Stockholm, Sweden), a licensed professional trainer in charge, and willingness to participate in the study over time. One additional stable in southern Sweden was included despite not fulfilling the proximity criteria due to the large amount of horses available at the yard. Trainers were either contacted directly and asked to participate, or were recruited through advertising in two Norwegian racing magazines.
Horse-level inclusion criteria were breed, age and training level; only Standardbred trotter yearlings that were broken to harness and within the first six months of driven exercise were recruited to the study. In addition, the horses were presumed sound, meaning that the trainer had not observed any lameness in the horse, and there was no history of other health issues believed to affect the movement of the horse. Exclusion criteria was an observed subjective lameness of ≥ 3/5 degrees according to the American Association of Equine Practitioners (AAEP) scale (0–5) during in-hand measurement.
Clinical examination and measurements of movement asymmetry
The horses were examined at their training yards or local racetrack. All horses underwent a general physical examination with auscultation of heart and lungs, measurement of height at the withers and pelvis, palpation of the musculature and distal extremities as well as an evaluation of their conformation. Examination was performed by one of the authors (ASK, EHSH or MH). The trainers filled out a questionnaire pertaining to their horses’ history of previous injury, veterinary treatment including dental examinations, hoof care/shoeing routine and radiographic screening for osteochondrosis (OC; radiographs included all four fetlock joints as well as hocks).
The horses were measured at the trot both in-hand and driven on a track with a sensor-based objective gait analysis system (Lameness Locator®, Equinosis, St. Louis, MO, USA). The horses were measured in-hand either before or after driven exercise. In-hand the horses were trotted in a straight line by their regular handler or one of the authors (ASK, EHSH, MR or EH). The ground surface was firm, consisting of either gravel, asphalt, packed dirt or hard packed snow/ice, and as even and level as circumstances allowed. Default software settings were used for stride selection for the asymmetry analyses; preferred stride selection was ≥ 25 steps. Data collection for in-hand trials were subjectively assessed as valid when a trot-up deemed representative for the horse had been completed. One trial per horse was used for analysis.
For track trials, the horses were exercised by their usual driver, with their regular tack and according to their planned schedule. All tracks were dirt tracks with a surface of packed dirt/sand, mixed with snow during the winter months. An electrode belt with a heart rate monitor (Polar Equine Belt, Polar Electro, Kempele, Finland) was placed around the trunk of the horse in front of the harness. Heart rate data was sent via Bluetooth to a device worn by the driver (Polar M450, Polar Electro, Kempele, Finland), which in addition registered speed, distance and route of the trial by means of an integrated GPS. Track trials were subjectively assessed as valid when a trial deemed representative for the horse had been completed. If a horse had more than one track measurement, the first measurement was used for analysis as long as the stride selection was ≥ 25 steps.
The IMU sensors of the movement analysis system were mounted on four locations on the horse: the poll, at the top of the withers, on the pelvis (between the tubera sacrale) and at the dorsal aspect of the right front pastern. For attachment on the poll, a purpose-made neoprene head bumper was attached to the head collar or bridle, and on the right front pastern a purpose-made pastern wrap was used. Sensors on the withers and pelvis were fastened with extra strong double-sided adhesive tape (Teppeteip, Clas Ohlson, Insjön, Sweden) and standard-issue duct tape; care was taken to ensure sagittal midline positioning. During track trials, the withers and pelvis sensors were covered with additional adhesive tape (Snøgg Animal Polster, Norgesplaster AS, Vennesla, Norway) to prevent loosening. The pastern wrap was secured with elastic tape (Norbind, Norgesplaster AS, Vennesla, Norway) to prevent rotation during exercise. The IMU sensors consisted of a tri-axial accelerometer, gyroscope and magnetometer that recorded the vertical acceleration of the head and torso and the angular velocity of the right front limb at 200 Hz with 8-bit digital resolution. Data transmission from the sensors was wireless via Bluetooth technology to a nearby computer tablet with the corresponding program software. For measurements on oval tracks, the IMU system tablet was placed in a small backpack worn by the driver to ensure continuous connection between the horse-mounted sensors and the receiving computer tablet.
Data processing
The program software mathematically converted the measured acceleration into a measurement of vertical displacement of the horse’s body using a double integration process (6). The biological basis for this approach is the natural locomotion of the horse. During trot, horses alternate between loading and unloading each diagonal leg pair with a corresponding sinking and rising of the head and pelvis. During this repeated diagonal loading and unloading, calculation of asymmetry relies on the difference between the minimum and maximum position of the head and pelvis between the two phases (left and right diagonal limb pair) of a trotting stride. Head movement is interpreted as reflecting forelimb asymmetry, while pelvic movement reflects hind limb asymmetry. Data from the withers sensor was not used in this study. Software data output consisted of four parameter values for each trial measurement calculated from the difference in head minimum (HDmin) and maximum (HDmax) positions between the right and left portions of the stride, and the difference in pelvis minimum (PDmin) and maximum (PDmax) position during the two portions of the stride. In addition, the vector sum (VS) of the mean HDmax and HDmin values was calculated as √(HDmax2 + HDmin2). Left front or hind limb attributed asymmetries were defined by negative values and right front- or hind limb asymmetries by positive values. Parameter output data were expressed in millimeters. A parameter value of 0 mm would indicate perfect symmetry between the left and right portion of the strides of a trial measurement. More detailed descriptions of the data processing can be found elsewhere (6,7).
Statistical methods
Descriptive data calculations
Criteria for movement asymmetry were based on recommendations for clinical use by the IMU system provider and correspond to published confidence intervals for repeatability of measurements with the system (6). This applies to the use of symmetry thresholds and subsequent division into asymmetry categories in this study, and not in regard to the recommended limits for standard deviation. For front limb values (HDmin, HDmax) the symmetry threshold was ±6 mm, for hind limbs (PDmin, PDmax) ±3 mm and for front limb VS 8.5 mm; parameter values below these pre-determined thresholds were consequently defined as ‘symmetric’. When measurements above these symmetry thresholds were observed, asymmetry categories defined as “mild”, “mild-moderate”, “moderate”, “moderate-severe” and “severe” were used based on a set increase in millimeter asymmetry by adding the threshold value (6, 3 or 8.5 mm, for front limb asymmetry, hind limb, and front limb VS, respectively) to the threshold for each parameter, corresponding to the classification presented in the IMU system output data (AIDE statement) with the exception of an additional ‘severe’ category. The resulting categories had thresholds for front limb/hind limb/VS values in mm as follows: Symmetric: 0–6/0–3/0–8.5, mild asymmetry: 6–12/3–6/8.5–17, mild-moderate asymmetry: 12–18/6–9/17–25.5, moderate asymmetry: 18–24/9–12/25.5–34, moderate-severe asymmetry: 24–30/12–15/34–42.5, severe asymmetry: >30/>15/>42.5. For all horses, trial output data (HDmin, HDmax, PDmin, PDmax, VS) for each trial condition (in-hand and/or track), was systematized according to the described thresholds to assess the distribution of limb asymmetry. Other categories were made based on the trial standard deviation magnitude, with three categories based on distance from the trial mean: Trials with SD > 120% of asymmetry mean, trials with SD > 50% and < 120% of asymmetry mean and trials with SD < 50% of asymmetry mean. These categories correspond to the levels of evidence (weak, moderate, strong) presented in the IMU system output data (AIDE statement).
A combined score per horse was created for front and hind limbs where the horse was classified as front limb asymmetric if one front limb parameter (HDmin or HDmax) was above threshold, and hind limb asymmetric if one hind limb parameter (PDmin or PDmax) was above threshold. Horses could be included in both front and hind asymmetry categories. The combined severity category was based on the highest asymmetry score.
For the horses with both in-hand and track trials, those with in-hand trial parameter values above threshold were noted, and corresponding values for the track trials of these horses were compared. Changes in parameter values were registered as one of three alternatives: The horse showed asymmetry of the same limb and above threshold during both trial modes; the horse showed same limb parameter values below symmetry threshold for track trials; or the horse showed parameter value asymmetry above threshold during both trial modes but changed the side of asymmetry between trials (left to right or vice versa).
An intra-class correlation coefficient (ICC) was calculated to investigate if horses at the same training yard resembled each other more closely in terms of magnitude of movement asymmetry than horses across the individual training yards.
Model building
Movement asymmetry data was analyzed using open software (R, version 3.6.1, The R Foundation for Statistical Computing, Vienna, Austria). Mixed models were created using the lmer function in the lme4 package. Four models were created, where each outcome variable was the absolute values of one of the four asymmetry parameters HDmin, HDmax, PDmin and PDmax. In all models, fixed effects were mode (with the levels: in-hand before track, in-hand after track, driven on a straight track or driven on an oval track) gender (male or female), height at the withers, height difference between the withers and pelvis, OC status (OC at radiographic screening; yes, no or missing data). Horse nested within trainer was entered as a random effect (random intercept) in all models. Normality of residuals was checked using q-q plots and homoscedasticity by plotting the residuals against the fitted values. Evaluation of statistical significance was made using type II p-values generated by a Wald F test with Kenward-Roger approximated degrees of freedom. The level of significance was defined as < 0.05.