MATERIALS AND METHODS

Ethical approval

The study protocol was reviewed and approved by the Laboratory Animal Center, Chiang Mai University, Thailand, which serves as the institution's Institutional Animal Care and Use Committee (IACUC) (approval code: AG001/2567). Written informed consent was obtained from the horse owners or authorized caretakers before enrollment. Only clinically healthy horses with owner authorization were included. All procedures were non-invasive and involved external 24-h HR and HRV monitoring using a chest strap sensor under routine management conditions. No experimental treatment, restraint beyond routine handling, invasive sampling, or procedure likely to cause pain or distress was performed. Horses were housed and managed according to their normal stable routine, with ad libitum access to water and regular feeding, grooming, and light activity schedules. During the monitoring period, animals were continuously observed by trained personnel for signs of discomfort, distress, abnormal behavior, or device-related irritation, and the monitoring device could be removed immediately if any welfare concern arose. All efforts were made to minimize disturbance and ensure animal welfare throughout the study.

Study period and location

The study was conducted in late January 2025 in Bangkok, Thailand, during the winter season in a tropical savanna climate. During the study period, ambient temperature ranged from 23°C to 33°C, while relative humidity ranged from 50% to 85%. These environmental conditions provided a representative tropical winter setting for investigating circadian autonomic regulation and the influence of environmental variables in horses maintained under field conditions.

Study design

This study was designed as an observational cross-sectional field study conducted under tropical environ-mental conditions. Horses were categorized into three age groups, and continuous 24-hour monitoring of HRV and environmental parameters was performed. Data collection was conducted over multiple days, with one horse from each age group monitored simultaneously during each recording session (three horses/day).

The protocol combined non-invasive 24-hour recordings using a Polar H10 HR sensor (Polar Electro, Kempele, Finland) in open-air tropical stalls under standardized management conditions with synchronized stall-level environmental monitoring using a WH2900C weather station (Shenzhen Fine Offset Electronics Co., Ltd., Shenzhen, China). This real-world approach provided an ecologically valid assessment of autonomic regulation in relation to both management-related and environmental influences under tropical winter conditions.

Horses

A total of 18 horses were initially enrolled in the study. Three animals were excluded because of excessive artifacts or poor compliance with the monitoring device. Consequently, only datasets meeting the predefined quality criteria were included in the final analysis, resulting in a study population of 15 horses.

The horses ranged from 4 to 20 years of age and comprised 12 geldings, two mares, and one stallion. Breed composition included eight Warmblood horses, six Warmblood-cross horses, and one Anglo-Arabian horse. The mean age and body weight were 12.06 ± 5.21 years and 530.13 ± 62.35 kg, respectively, with body weights ranging from 445 to 620 kg. Body condition scores ranged from 3 to 4 according to the scoring system described by Carroll and Huntington [19]. All horses were used for riding-school activities and were maintained under comparable management conditions, with no history of intensive athletic training.

The inclusion criteria consisted of clinically healthy horses with normal findings on physical examination, auscultation, electrocardiography, and echocardiography and without a history of systemic illness or administration of medications during the preceding two months. Horses with cardiac abnormalities or active medical conditions were excluded from the study.

Eligible horses were allocated to three age groups: Group 1 (4–7 years), Group 2 (8–14 years), and Group 3 (15–20 years). Detailed characteristics of the study groups are presented in Table 1.

Characteristic | Category | Group 1 (4–7 years) | Group 2 (8–14 years) | Group 3 (15–20 years)
Sex | Stallion (n) | 1 | 0 | 0
 | Gelding (n) | 4 | 4 | 4
 | Mare (n) | 0 | 1 | 1
Breed | Warmblood (n) | 2 | 1 | 5
 | Warmblood-cross (n) | 3 | 3 | 0
 | Anglo-Arabian (n) | 0 | 1 | 0
Age (years) |  | 5.5 ± 1.10 | 11.8 ± 1.92 | 17.4 ± 1.82
Body weight (kg) |  | 463.60 ± 12.97 | 575.40 ± 37.31 | 551.40 ± 58.20

Horses were allocated to age groups based on predefined age ranges, and no randomization was performed. During each recording session, one horse from each age group was monitored simultaneously under identical environmental conditions. Efforts were made to select animals with comparable body condition scores; however, some variation in body weight and breed distribution among groups remained because of population availability.

General physical examination

Before enrollment, all horses underwent a comprehensive physical examination, including assessment of vital signs (HR, respiratory rate, and rectal temperature), cardiac and pulmonary auscultation, evaluation of mucous membrane color and capillary refill time, and gastrointestinal auscultation. All animals were clinically healthy and had no history of medication administration during the previous two months. Furthermore, no horse exhibited signs of illness or was undergoing treatment at the time of the study.

HR and HRV monitoring

HR and normal-to-normal intervals (NN) were continuously recorded for 24 hours in all horses using a Polar H10 HR sensor (Polar Electro) with a sampling frequency of 1,000 Hz. Before sensor placement, horses were thoroughly groomed to remove debris from the electrode contact area. The sensor was secured with a non-invasive, equine-specific chest strap (Polar Electro), pre-wetted with water to optimize electrical conductivity between the skin and the electrodes, according to the manufacturer's instructions. Following skin preparation, the chest strap was positioned behind the withers, with the sensor unit aligned vertically between the shoulder and elbow (Figure 1).

Figure 1: Placement of the heart rate sensor on the horse. The sensor was fastened with a trotter strap behind the withers, with the unit aligned vertically between the shoulder and elbow.

Data acquisition began at 06:00 a.m. and continued uninterrupted for 24 hours. Sensor connectivity and data acquisition were managed using the Kubios HRV mobile application (version 1.7.12; Kubios Oy, Kuopio, Finland), which ensured synchronization and recording integrity. Signal quality and sensor position were regularly checked throughout the recording period to minimize motion-related artifacts.

Throughout the monitoring period, horses were housed individually in open-air stalls equipped with ceiling- or wall-mounted fans to facilitate air circulation. Horses maintained visual and auditory contact with neighboring animals and had ad libitum access to fresh drinking water.

The horses followed a standardized daily routine that included scheduled feeding, grooming, and light activity. Concentrate and roughage were provided at fixed times, whereas grooming and hand walking were performed during the daytime. Human interaction was minimized at night. The daily activity schedule is summarized in Table 2.

Time | Activity
04:00 | Concentrate feeding
06:00 | Heart rate sensor attachment
07:00 | Grooming
09:00 | Roughage feeding
11:00 | Concentrate feeding
13:00 | Body temperature monitoring
15:30 | Grooming
16:00 | Hand walking
17:00 | Return to stall
20:00 | Concentrate feeding
21:00 | Roughage feeding

HRV data analysis

The 24-h recording period was divided into eight-time intervals to evaluate circadian patterns of HR and HRV. For day–night comparisons, recordings were categorized into daytime (06:00–18:00) and nighttime (18:00–06:00). HRV parameters were also analyzed over the entire 24-hour period.

Raw inter-beat interval data were analyzed using Kubios Scientific HRV software (version 4.1.2.1; Kubios Oy, Kuopio, Finland). All RR interval tachograms were visually inspected for irregularities. Although electrocardio-graphy was not performed to confirm physiological arrhythmias, such as second-degree atrioventricular block, manual inspection was rigorously performed to minimize their influence on HRV analysis.

The threshold-based artifact correction algorithm implemented in Kubios Scientific HRV software was applied using a medium correction level (threshold = 0.25 s). Recordings were accepted only when the proportion of corrected beats remained below 5%, ensuring that >95% of RR intervals were suitable for analysis. Segments with excessive noise or poor signal quality were excluded. Missing or ectopic beats were corrected using cubic spline interpolation. To reduce low-frequency (LF) non-stationarity, detrending was performed using the smoothness priors method (λ = 500, cutoff frequency = 0.035 Hz). Each 24-hour recording underwent complete visual inspection to ensure overall signal integrity. Frequency-domain variables were automatically calculated using consecutive 5-minute epochs.

Time-domain variables included NN, standard deviation of NN intervals (SDNN), root mean square of successive differences (RMSSD), standard deviation of the averages of NN intervals for each 5-minute segment (SDANN), and the percentage of NN intervals differing by >100 ms (pNN100). In the present study, pNN100 was used instead of pNN50 as a species-specific adaptation because horses characteristically exhibit lower HR and longer RR intervals [20].

Fast Fourier transformation was used to convert NN intervals into frequency components. Frequency-domain analysis was performed using predefined frequency bands. The very low-frequency (VLF) range was defined as 0.001–0.01 Hz, the LF range as 0.01–0.12 Hz, and the high-frequency (HF) range as 0.12–0.6 Hz. The HF range was selected based on the respiratory frequency of the study population (0.18 ± 0.02 Hz) and previously reported equine HRV frequency ranges [21–23]. Variables analyzed included VLF power, LF power, HF power, total power, and the LF/HF ratio.

Environmental parameter monitoring

Environmental variables were continuously monitored over 24-hour periods to evaluate their associations with HRV parameters. Variables included ambient temperature, relative humidity, feels-like temperature, light intensity, air quality index (AQI), and particulate matter ≤2.5 μm (PM2.5). Feels-like temperature represented a composite index incorporating temperature, humidity, and wind speed. Both temperature variables were retained to capture ambient conditions and perceived thermal load. Because of their inherent correlation, they were not simultaneously incorporated into the same model.

Environmental data were collected using a factory-calibrated weather station (WH2900C; Shenzhen Fine Offset Electronics Co., Ltd.). The station was installed at a height of 1.8 m within the stable area to reflect the immediate environment of the horses. Environmental data were transmitted and stored in real time via the Ecowitt cloud platform (Ecowitt, Shenzhen, China), enabling synchronization with physiological recordings. Both the weather station and mobile devices used for HRV acquisition were synchronized to the same network time source. No significant data loss occurred during the study.

Statistical analysis

An a priori sample size calculation was performed using G*Power software (version 3.1.9.4; Heinrich Heine University Düsseldorf, Düsseldorf, Germany). RMSSD was selected as the primary outcome because of its sensitivity to parasympathetic modulation and relevance to age-related changes. Based on a previous equine study, a one-way analysis of variance model with three groups (α = 0.05, power = 0.80, effect size f = 1.23) indicated that 12 horses were required. Therefore, the inclusion of 15 horses was considered adequate.

Data are expressed as mean ± SD. Normality was evaluated using the Shapiro-Wilk test and quantile-quantile plots, whereas homogeneity of variance was assessed using Levene's test. No data transformation was required.

A mixed-model analysis of variance was used to evaluate the effects of time (within-subject) and group (between-subject), with horse identification included as a random effect to account for repeated measurements. Effect sizes for group, time, and interaction terms were reported using generalized eta-squared. When significant interaction effects were detected, simple main effects were explored. Post hoc comparisons were performed using pairwise t-tests with Holm correction, and effect sizes were expressed as Cohen's d.

Correlations between HRV parameters and environmental variables were assessed using Spearman's rank correlation coefficient. Environmental variables were analyzed individually in relation to HRV parameters and were not included simultaneously in multivariable models; therefore, multicollinearity was not considered a concern. All statistical analyses were performed using R software (version 4.5.0; R Foundation for Statistical Computing, Vienna, Austria), and statistical significance was set at p < 0.05**.**