The effects of ambient air pollution exposure on Thoroughbred racehorse performance

Abstract

Background

Limited research exists on impacts of air pollution on non‐human mammals, particularly animal athletes such as Thoroughbred racehorses. Athletes have a greater risk of exposure as heightened exertion and increased airflow carry more pollutants deeper into the respiratory tract.

Objectives

To provide insights into the impact of ambient air pollution, particularly fine particulate matter (PM_2.5), on race speed.

Study design

Retrospective observational study.

Methods

Data were obtained from The Jockey Club Information Systems, covering 31 407 winning races by Thoroughbred horses in California spanning 10 years (2011–2020) and evaluated the association between air pollution and winning race speeds. For race days, we collected PM_2.5 data from the nearest U.S. Environmental Protection Agency (EPA) monitoring site within 100 km of each racetrack (n = 12). We assessed the associations between daily average PM_2.5 concentrations and speed of winning horses with linear mixed effects regression. We adjusted for horse characteristics, race‐related covariates, temporal indicators (e.g., year), other air pollutants and temperature. We conducted sensitivity analyses by adjusting extreme air pollution days by reassigning values to the 95th percentile value and conducting linear mixed effects regression on series of datasets with incremental cutpoints of PM_2.5.

Results

In the cutpoint analysis, we found that for PM_2.5 between 4 and 23.6 μg/m³, speed decreased 0.0008 m/s (95% CI: −0.0014562 to −0.00018) for every 1 μg/m³ increase of PM_2.5.

Main limitations

Limitations include the use of offsite monitors leading to imprecise exposure measurements, not using training practice data, and generalisability as the study focuses on California racetracks.

Conclusion

This study highlights the need to create advisories to safeguard the performance of horses during periods of poor air quality. Further research is recommended to explore additional factors influencing the relationship between air pollution and equine welfare.

1. INTRODUCTION

Air pollution in human populations has consistently demonstrated adverse respiratory and cardiovascular health effects. ¹ Research studies have linked ambient air pollution to adverse health outcomes such as emergency department visits, hospitalisations and death, as well as adverse biological outcomes, such as changes in lung function, increases in heart rate variability and increases in systemic inflammatory markers. ¹ , ² , ³ , ⁴ Despite this, there is limited information on the health effects of air pollution in non‐human mammals. Laboratory animals and livestock have been studied to understand the effects of air pollution on mammalian systems from a mechanistic perspective, ⁵ , ⁶ , ⁷ , ⁸ but few studies have explored the effects of air pollution on animal athletes. ⁹ , ¹⁰ , ¹¹

Athletes are particularly vulnerable to the impact of air pollution as their total airway exposure to pollutants is magnified by their increased respiratory rate and volume during exercise. ¹² , ¹³ In human studies, acute exposure to fine particulate matter (PM_2.5) and ozone (O₃) during exercise can cause decreases in lung and vascular function in healthy individuals as well as people with asthma. ¹³ , ¹⁴ The increasing number and intensity of wildfires (e.g., in Western United States, Southern Europe, Australia) and related wildfire smoke have raised concern about the animal health effects of poor air quality ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ ; veterinarians in affected areas are being asked to advise regarding the management of horses during smoke events. The limited studies and lack of equine specific guidelines have left the U.S. Equestrian Federation (USEF) to default to using human recommendations such as avoiding the outdoors if the Air quality Index (AQI) is above 150 (~55.4 μg/m³ of daily average PM_2.5). ⁹ , ¹⁰ , ²⁰ , ²¹ However, when considering the lung capacity of horse (VO_2max ~150 mL/(kg.min)) being almost three times that of a human (VO_2max ~65 mL/(kg.min)), horses could be more affected by air pollution than currently recognised. ²² , ²³

Short‐term ambient air pollution is characterised by regional and temporal variability of pollutants. Under the U.S. National Ambient Air Quality Standards (NAAQS) promulgated by the U.S. Environmental Protection Agency (U.S. EPA), six air pollutants (i.e., criteria air pollutants) are monitored and regulated at an hourly or daily level, providing a retrospective observational study to understand the impact of ambient air pollution on animal athletes. Combining these environmental data with horseracing data results (winning times) provides an opportunity to understand the relation of pollution on an objective measure of athlete performance. Our aim in this study is to assess the association between racehorse winning speeds and short‐term regional air pollutants. We focused on PM_2.5 as U.S. EPA estimates that PM_2.5 is responsible for 90% of the health effects in human populations. ²⁴ Our study explores if changes in winning speeds are dependent on (i) exposure to PM_2.5, (ii) exposure to certain concentrations levels of PM_2.5, (iii) exposure to PM_2.5 at different racetracks. This information is needed to develop recommendations for management of equine athletes during adverse air pollution events.

2. MATERIALS AND METHODS

2.1. Study area

Our study included 12 equine racetracks across California (Figure 1 and Table S1). California has the second highest horse population in the United States, ²⁵ and due to the temperate climate, most Thoroughbred racetracks run throughout the year. Given California's area, topography and size and distribution of population (the most populous state in the United States), pollution trends are varied throughout the state.

FIGURE 1.

Map of California showing all racetracks in the study, and a closer look at the Golden Gate Fields racetrack. Coloured circles show the location of monitors and racetracks in relation to major roads.

2.2. Outcome data

We obtained study data for all California horse races from 2011 to 2020. Data were obtained from The Jockey Club Information Systems, which provided information on the winning horse, winning horse time, signalment, race date and track conditions. Given that races have varying lengths, we calculated the winning horse speed for each race using length of the race (m) over winning time (s) as the outcome.

2.3. Exposure data

We retrieved daily average air pollution concentrations (μg/m³) for fine particulate matter (PM_2.5), coarse particulate matter (PM₁₀), daily maximum 8‐h average (MDA8) O₃ concentrations (parts per billion—ppb) and daily maximum 1‐h nitrogen dioxide (NO₂) concentrations (ppb) from the United States Environment Protection Agency (EPA) Air Quality System Data Mart. ²⁶ All pollution monitors were selected based on proximity to each racetrack location and completion of data. For each racetrack, we excluded monitors further than 100 km from the racetrack, and those that have geographical barriers (e.g., mountain ranges). ²⁷ We then selected one monitor closest to the racetrack for each criteria pollutant. If a further monitor existed that provided more complete data on race days, that monitor was chosen instead if it was less than 20 km away from the closest monitor and had at least 3% more data available, based on a natural cutpoint in the data. MDA8 O₃ and NO₂ data were available daily; average daily PM_2.5 and PM₁₀ concentrations were often measured every third or sixth day. Like previous studies, ⁸ missing pollutant values were replaced by a weighted moving average on the day before and after the missing values. For pollutants measured every third day, three or more consecutive missing values were dropped from the analysis. For pollutants measured every sixth day, six or more consecutive missing values were dropped from the analysis. Months with less than 75% of available pollution data were excluded from the analysis. Daily average temperature data were retrieved from the Automated Surface Observing System (ASOS) from the National Weather Service. ²⁸

2.4. Covariate data

We included several covariates that could either be considered confounders (third variables that obscure the relationship between exposure and outcome) or precision variables (factors related to the outcome that may reduce the standard errors for the coefficient for the exposure of interest). ²⁹ Covariates for the regression model included horse‐level descriptors (sex and age), and race‐related variables (track location, track surface material, track conditions, race time and betting race types). Where there was minimal variability or low numbers, some categories were combined.

3. DATA ANALYSIS

Descriptive statistics were calculated for all dependent and independent variables to check for outliers, patterns in missing data and normality. Depending on normality, mean, standard deviation or median were used to describe continuous variables such as air pollutant concentration (i.e., PM_2.5, O₃, NO₂, PM₁₀) horse‐level variables (i.e., winning speeds, age), temporal indicators (year), temperature and monitor distance of air pollutants. Counts and percentages were used to describe categorical variables such as horse‐level variables (i.e., sex, furlong, race type, track conditions and track surface) and temporal indicators (i.e., time of day, weekend). Spearman correlations and generalised variance‐inflation factors (GVIF) were used to check for multicollinearity among the variables. For Spearman correlations, we considered a magnitude of |0.7| as highly correlated, and for GVIF, a threshold above 5 determined the presence of multicollinearity. This dataset includes multiple observations from repeated horses winning multiple races, therefore observations were considered to be non‐independent, meaning that winning speeds are more likely to be similar within a horse, than comparison of winning speeds across horses. To address this non‐independence of observations we implemented linear mixed effects regression, accounting for multiple wins per horse.

The initial regression model assessed the linear relationship between daily average PM_2.5 concentrations and the winning horse speeds. For a priori reasoning, the full model includes other air pollutants and racing related variables. ³⁰ , ³¹ , ³² , ³³ The final model form is:

$$Y_{\mathit{ti}}={\beta }_0+{\beta }_{1}X{1}_{\mathit{ti}}+{\beta }_{2}X{2}_{\mathit{ti}}+{\beta }_{3}X{3}_{\mathit{ti}}+b_i+e_{\mathit{ti}},$$

where Y represents winning horse speed for the tth day on the ith horse, ${\beta }_0$ is the intercept, ${\beta }_1$ are coefficients for the vector of pollutants (i.e., PM_2.5, O₃, NO₂, PM₁₀), which were added sequentially, one at a time, to an initial model with PM_2.5; ${\beta }_2$ represents coefficients related to the horses and racetracks (i.e., age, sex, furlong, race type, track conditions and track surface), ${\beta }_3$ represents seasonal and long term patterns for temperature (24‐h average) and temporal indicators (i.e., time of day, weekday/weekend, month and year). The term b _i represents the random intercept for each horse, but is constant over time for each horse, and e _ti is the model residuals for each horse‐specific race. Accounting for the non‐linear relationship between speed and temperature, we used a natural cubic spline for temperature with knots at 50 and 70 °F. ³⁴ , ³⁵ Categorical variables included horse sex, track conditions (fast track/other), track type (dirt or turf), furlong, weekend (y/n) and time of day (morning, afternoon, evening); age, month and year were modelled as a continuous variable. Model fit was determined using linearity diagnostics, Kolmogorov–Smirnov test to check for normality and Bayesian Information Criterion (BIC).

Using the linear model, we conducted several sensitivity analyses. To account for days that might be influenced by exceptional PM_2.5 events such as wildfire smoke, which results in major spikes in fine particulate matter, we truncated these extreme air pollution days by reassigning the PM_2.5 values at the 95th percentile (23.6 μg/m³), truncating all other pollutants at their 95th percentile value and reran the initial regression model. ⁸ Further, since there are no current thresholds for PM_2.5 to assess potential adverse impacts on race performance, we applied the full multilevel model to a series of data with incremental cutpoints of PM_2.5. For each concentration of PM_2.5 levels, the data set was divided into two subsets: (a) all races occurring at or above the specified PM_2.5 cutpoint and (b) all races occurring below the specified PM_2.5 cutpoint, rounded to the nearest whole number. We ran the model for different subsets of data, for cutpoints of PM_2.5 between 4 and 19 μg/m³ where there was enough data for linear regression. Finally, we tested for a possible linear interaction between PM_2.5 and racetracks to determine if individual racetracks modified the relationship between PM_2.5 and speed.

All analyses were conducted using R version 4.1.2. The RAQSAPI package version 2.0.3 was used to retrieve air pollutant data, ImputeTS package version 3.3 to calculate moving averages and riem package version 0.3.0 to retrieve meteorology data.

4. RESULTS

Data from 38 041 races were available for the 10‐year period. After excluding improbable values from data collection errors (i.e., negative air pollution concentrations, unidentifiable as female or male horse and event end times by the hour greater than 12) or missing values, there were 31 407 races included in our analytic data set (information on the distribution of missing values is included in Figures S1 and S2). All variables presented with mean and standard deviation were normally distributed. Overall, 80.9% of races were held from 1 to 5 p.m. and 55% of races were held on the weekends (Table 1). Approximately 77.5% of the races were run on a fast track, and 81% of the tracks had a dirt surface. Most races were between 4 (804.7 m) and 8 (1609 m) furlongs.

TABLE 1.

Summary of race‐related and horse‐related information.

	Overall (N = 31 407)
Race‐related variables
Race type
Claiming handicap (CLM)	12 923 (41.1%)
Claiming (MCL)	6773 (21.6%)
Maiden special weight (MSW)	3474 (11.1%)
Allowance or optional claiming (AOC)	2841 (9.0%)
Stakes (STK)	1906 (6.1%)
Starter allowance (STR)	1779 (5.7%)
Other	1711 (5.4%)
Furlong length
≤4	347 (1.1%)
Between 4 and 6	9798 (31.2%)
=6	6458 (20.6%)
Between 6 and 8	3365 (10.7%)
=8	7352 (23.4%)
Between 8 and 9	3119 (9.9%)
>9	968 (3.1%)
Weekend
Weekend	17 373 (55.3%)
Weekday	14 034 (44.7%)
Time of day
Morning (10 am to 12 pm)	2124 (6.8%)
Afternoon (12 pm to 5 pm)	25 421 (80.9%)
Evening (5 pm to 9 pm)	3862 (12.3%)
Track surface
Dirt	25 424 (81.0%)
Turf	5983 (19.0%)
Track conditions
Fast track	24 330 (77.5%)
Other	7077 (22.5%)
Horse‐related variables
Sex
Male	17 428 (55.5%)
Female	13 979 (44.5%)
Age
Mean (SD)	4.06 (1.41)
Speed
Mean (SD)	16.9 (0.524)

Note: A total of 14 182 racehorses won races, and 7308 of those were repeated winners.

4.1. Winning horses

The study population included 14 182 racehorses that won races during the study period; 7308 horses were repeat winners. In this study, 55.5% of the population were male horses, ranging in age from 2 to 11 years‐old (mean: 4.06 years; SD: 1.41). Mean winning speed for the study period was 16.88 m/s (SD: 0.52) (Table 1).

4.2. Air pollutant and meteorology measurements

Summary statistics for air pollutant values and meteorological data are presented in Table 2. Average daily PM_2.5 mean was 10.6 μg/m³ (SD: 7.3). There were 162 days (1549 races) with PM_2.5 higher than the 95% percentile of 23.6 μg/m³, and of those days the mean (SD) PM_2.5 value was 32.6 μg/m³ (13.3). There were 43 days (374 races) above the PM_2.5 24‐h NAAQS of 35 μg/m³. The mean for MDA8 O₃ was 38.9 ppb (SD: 13.6). A total of 163 days (1550 races) were above the 95th percentile of MDA8 O₃ (74.2 ppb), which exceeds the current NAAQS MDA8 O₃ of 70 ppb. For average daily PM₁₀, no days above the NAAQS were observed. Temperature mean (SD) was 61.7 °F (7.36). Table S2 details the distances of pollution monitor for each racetrack. The median (IQR) monitor distance was 17.01 (10.1–32.4).

TABLE 2.

Summary statistics for pollutants and temperature variables for all winning races (n = 31 407).

	Mean (SD)	Min	Q1	Median	Q3	Max
24‐h average PM2.5 (μg/m3)	10.7 (7.30)	0	6.20	9.30	12.8	152
MDA1 NO2 (ppb)	23.3 (13.3)	0	12.4	22.3	32.2	79.5
24‐h average PM10 (μg/m3)	21.7 (11.8)	1.00	14.0	19.0	27.0	146
MDA8 O3 (ppb)	38.9 (13.6)	1.00	30.0	38.0	46.0	115
24‐h average temperature (°F)	61.7 (7.36)	39.1	56.6	61.4	66.7	85.5

4.3. Correlations

Notable Spearman correlations (>|0.5|) were found between (i) PM_2.5 and PM₁₀, (ii) track surface and track conditions and (iii) Los Alamitos racetrack and evening races. Other correlations are depicted in Figure S3. Multicollinearity was not a concern as all variables had a GVIF^1/(2·df) value under 5 (Table S3). ³⁶

5. INFERENTIAL STATISTICS RESULTS

5.1. Aim I

5.1.1. PM_{2 .5} exposure

According to our model results from the full data set (without reassignment of extreme air pollutant values), PM_2.5 adjusted for other air pollutants showed no associations with speed (Tables S4 and S5). However, according to linear regression diagnostics for the full multilevel model (Figure S4) the relationship between speed and PM_2.5 was not linear nor are the residuals normally distributed (D (31 407) = 0.34595, p < 0.001).

5.1.2. PM_{2 .5} exposure without extreme values

Model results using the truncated data set (with reassignment of extreme air pollutant values) can be found in Table 3. Model 1 with only PM_2.5, showed speed increasing 0.0006 m/s for every unit increase of PM_2.5 (95% CI: 0.0002–0.0011). Model 2–4, all of which included other pollutants, had speed decreasing for each unit of PM_2.5 and speed, but this was not statistically significant. The truncated models also indicated a positive and significant relationship between speed and NO₂ and PM₁₀ that remained robust to adjustment to other air pollutants. For example, for every 1 μg/m³ increase in PM₁₀, winning horse speed increased by 0.0005 m/s (95% CI: 0.0002, 0.0009) when adjusted for the three other air pollutants. Although ozone was also positively associated with increases in winning speed, the association was not statistically significant in single pollutant or multiple pollutant models. Despite Model 3 having the highest BIC value we opted for Model 4 for subsequent to include MDA8 O₃ due to its known inverse associations with PM_2.5. ³⁷ However, though these data were truncated, regression diagnostics indicated similarly to the full model that the relationship between speed and PM_2.5 was not linear and residuals were not normally distributed (D (31 407) = 0.34595, p < 0.001).

TABLE 3.

Speed (m/s) $\beta$ coefficients and 95% confidence intervals with truncated data.

	Model 1		Model 2		Model 3		Model 4
	$\beta$	95% CI	$\beta$	95% CI	$\beta$	(95% CI)	$\beta$	(95% CI)
PM2.5	0.0006	0.0002, 0.0011	−0.0000002	−0.0005, 0.0005	−0.0006	−0.0011, 0.00002	−0.0005	−0.0011, 0.00009
NO2		_	0.0009	0.0006, 0.0012	0.0008	0.0005, 0.0011	0.0008	0.0005, 0.0011
PM10		_		_	0.0006	0.0003, 0.0010	0.0005	0.0002, 0.0009
O3		_		_		_	0.0002	−0.00004, 0.0005
BIC		−8126		−8155		−8159		−8151

5.2. Aim II

5.2.1. PM_{2 .5} exposure at specific ranges

Results shown in Figure 2 (truncated dataset) and Figure S5 (full dataset) summarise changes in speed for each cutpoint between 4 and 19 μg/m³. In the cutpoint analysis, we found that for PM_2.5 between 4 and 23.6 μg/m³, speed decreased 0.0008 m/s (95% CI: −0.0014562 to −0.00018) for every 1 μg/m³ increase of PM_2.5. Data were then split into two categories: observations with PM_2.5 values at and above the cutpoint (blue triangles) and observations below the cutpoint (yellow circles). For example, the cutpoints with PM_2.5 values at and above 6 μg/m³ had observations (n = 24 380) with PM_2.5 values between 6 and 23.6 μg/m³. For the cutpoints with PM_2.5 values below 6 μg/m³ had observations (n = 7027) with PM_2.5 values between 0 and 6 μg/m³. Regressions were applied independently to each of these data subsets. Using the example of 6 μg/m³, below 6 μg/m³, 1 μg/m³ increase of PM_2.5 was associated with an increased in winning speed of 0.00254 m/s (95% CI: −0.00205 to 0.00713). For PM_2.5 at or above 6 μg/m³, 1 μg/m³ 3 increase of PM_2.5 was associated with a decrease in winning speed of 0.00105 m/s (95% CI: −0.00177 to −0.000323). Upon review of only the races conducted with air pollution above the cutpoint (blue triangles), results showed negative and statistically significant changes in speed at cutpoints 4–11 μg/m³ of PM_2.5 (see Figure 2 for all confidence intervals), meaning as PM_2.5 increases racehorses are running slower. At 13 μg/m³, the estimates for the two data subsets converge, and speed increased approximately 0.0010 m/s for every increase of PM_2.5 (Figure 2). After 13 μg/m³, the data subsets below the cutpoints have confidence intervals that include the null value indicating that there is not a meaningful relationship between winning speed and PM_2.5 values below the cutpoint. For the data subset at and above 13 μg/m³, the winning speeds increase as PM_2.5 concentration increase. Observations at and above 19 μg/m³ indicate that for every 1 μg/m³ increase of PM_2.5, winning speed increases by 0.0114 m/s (95% CI: 0.0032–0.0196). The wide confidence intervals observed at the lowest and highest ranges of PM_2.5 indicate a high degree of variability due to sample sizes.

FIGURE 2.

Using the truncated dataset, the full sample (n = 31 407) was divided into two subsets repeatedly based on the specified daily average PM_2.5 cutpoint levels as shown on the x‐axis. We applied the full regression model to these series of data subsets for different ranges of PM_2.5 showing the complex relationship between speed and PM_2.5. Each point (yellow circle or blue triangles) with 95% confidence intervals (bars) shows the change in speed for every increase of PM_2.5 for each data subset. The estimates from the yellow circles came from data subsets that included PM_2.5 ranging between 0 μg/m³ to the indicated cutpoint level. The estimates from the blue triangles came from data subsets that included PM_2.5 between at cutpoint level to 23.6 μg/m³ of PM_2.5. Overall, changes in speed vary for different ranges of PM_2.5 concentrations. For the low ranges of PM_2.5 (values starting at 0 μg/m³ or yellow circles), the estimates show a gradual decrease in speed. For the higher ranges of PM_2.5 (values starting at specified cutpoint or blue circles) especially cutpoints 4–11 μg/m³ of PM_2.5, racehorses are running significantly slower for every increase of PM_2.5 as shown by the confidence intervals not crossing the red null line (p < 0.05).

5.3. Aim III

5.3.1. PM_{2 .5} exposure for each racetrack

Figure 3 shows PM_2.5 estimates from applying the regression model to the truncated dataset for each racetrack, with the full dataset estimates available in Figure S6 and Table S5. For the interaction model, we used Santa Rosa (SR) as the reference racetrack due to its low average pollutant concentrations across all pollutants in the study (Table S6). Compared with all other racetracks, SR had the most positive change in speed estimate. Horses who competed at SR, had a 0.01 m/s increase in speed for every increase of PM_2.5 μg/m³ (95% CI: 0.006–0.015). In contrast, a total of eight racetracks had a significantly statistically slower change in speeds compared with SR (p < 0.05), evident by the non‐overlapping confidence intervals in Figure 3. Overall, four racetracks fell below the red null line and showed statistically significant decreases in speed for every increase in PM_2.5. For instance, those competing at GG, racehorses had a 0.001 m/s decrease in speed for every increase of PM_2.5 (95% CI: −0.002 to −0.0002), meaning as PM_2.5 increases racehorses ran slower.

FIGURE 3.

After including racetrack as an interaction term in the full regression model for truncated data, we found racetracks modify the relationship between speed and PM_2.5. Each racetrack has a different change in speed estimate for every increase of PM_2.5. All racetracks ran slower than reference racetrack SR, a track with low average pollutant concentrations across all air pollutants in the study. Eight racetracks have statistically significantly different changes in speed, demonstrated by non‐overlapping confidence intervals, compared with the reference racetrack SR. Racetracks DMR, FER, FPX and GG showed racehorses ran slower as PM_2.5 concentrations increased.

6. DISCUSSION

This study examined the impact of short‐term regional air pollutants on the performance of equine athletes, accounting for meteorological conditions, race conditions and individual horse data. For our first aim, we adjusted for confounders and extreme values, and our findings suggest that linear regressions are insufficient in characterising the relationship between speed and PM_2.5. The model residuals were not normal thus leading us to explore the relationship between speed and specific ranges of PM_2.5 concentration levels. For our second aim, the cutpoint analysis, showed different associations for various ranges of PM_2.5 concentration levels. This highlights that there is a negative and non‐linear relationship between PM_2.5 and speed: horses ran slower for specific ranges of average daily PM_2.5, starting at 4 μg/m³. In other words, athletic performances declined when exposed to PM_2.5 values within the ranges commonly found in areas without heavy sources of anthropogenic pollution or wildfire smoke.

Our findings suggest horses are adversely affected by PM_2.5 even at concentrations as low as 4 μg/m³. This challenges the current USEF recommendations of 55.4 μg/m³ or an AQI of 150 for mitigating exposures to ambient air pollution. ²¹ Considering the US PM_2.5 annual average is 7.8 μg/m³, ³⁸ advising avoidance at 4 μg/m³ (AQI 22) is impractical; however, these results should still be heeded. Practical recommendations should incorporate awareness of seasonal air pollutant patterns, with particular attention to peak PM_2.5 periods in May to October from wildfires and higher background PM_2.5 during winter seasons. ³⁹ , ⁴⁰ This is especially important in CA where races occur year‐round. Similar to the human AQI guidelines, future advisories for horses could also include different AQI categories that correspond to a different level of health concern. ⁴¹ For instance, AQI 100–150 is especially critical for populations sensitive to health impacts of air pollution, including people with asthma. ⁴¹ Given that asthma is a condition also experienced by horses, AQI awareness among trainers and owners is critical to maintain the health and welfare of equine athletes.

Moreover, our findings for our third aim indicate different racetracks yield different changes in speed when exposed to daily average PM_2.5 concentrations. There were eight racetracks with speeds that were statistically significantly slower than SR (the track with the least pollution across all air pollutants) for every increase of PM_2.5 (see Figure 3 for all confidence intervals). Of those eight racetracks, four racetracks (DMR, FER, FPX and GG) showed statistically significant negative associations between PM_2.5 and speed; three racetracks (LA, SA and SAC) showed negative associations that were not statistically significant; one racetrack (HOL) showed positive associations that were not statistically significant. The different associations between speed and PM_2.5 for every racetrack may be due to varying sources of PM_2.5, which exhibit different levels of toxicity depending on source and residence time in the atmosphere. ⁴² For example, combustion‐related PM_2.5 from fossil fuels is considered more toxic than other sources of PM_2.5 from biogenic sources (e.g., wildfires). ⁴³ Certain areas may be more impacted by the pollution from traffic, industrial processes, agriculture and wildfires. ⁴⁴ In particular, PM_2.5 pollution exposure levels from vehicles in Southern California, specifically around Los Angeles, is 2.5 times greater than that in the Bay Area near San Francisco. ⁴⁵ Furthermore, the Central Valley, positioned between Los Angeles and San Francisco, remains one the most polluted areas in the United States for particulate matter. ⁴⁶ , ⁴⁷ Alternative hypotheses for the interaction of tracks and PM_2.5 could be due to athlete characteristics, diverse training regimens, horse management practices and prestige of certain racetracks that may results in distinct subpopulations of horses at different racetracks.

To our knowledge, there are only two studies that characterise the relationship between ambient air pollution and Thoroughbred horse race speed. ⁹ , ¹⁰ Unlike our results, one study found racehorses ran slower when exposed to elevated levels of O₃ compared with standard levels of O₃, and no other air pollutants were associated with speed. ⁹ However, these results may not be comparable to ours as the author sampled 27 races, and very few race days had elevated pollution levels. In Chile, the speed from racehorses found in handicapped races were negatively correlated with PM_2.5, PM₁₀, NO₂, NO and SO₂, and positively correlated with O₃. ¹⁰ The author also reported higher mean values for PM_2.5, O₃, PM₁₀ and NO₂ compared with our study. Unlike these studies, our analysis included a robust multilevel regression model accounting for multiple observations per horse and analysed over 31 407 races adjusting for multiple confounding factors. In comparison, most literature on human athletes primarily explores the impact of high air pollution on athletic performances. ⁴⁸ Some studies have noted a decline in lung function among athletes following exposure to high levels of PM_2.5. ⁴⁹ , ⁵⁰ However, the impact of lower‐level ambient air pollution, cumulative exposures and the influence of training versus race day exposures on performances are not fully understood, and provide opportunities for future research in both human and animal athletes. Recent studies have reported an increased likelihood of adverse cardiovascular outcomes in human populations when the daily average PM_2.5 concentration is between 7.1 and 15.4 μg/m³. ⁵¹ These results, along with findings from our study, demonstrate that ambient levels of PM_2.5 pollutants that do not include extreme values have adverse impacts on health.

Although limited, previous work has found exposure to air quality in stables and outdoor environments to impact health and performance in horses. Research on Ontario horses exposed to lagged weekly mean PM_2.5 and NO₂ were associated with greater risk of lower airway inflammation. ⁵² An investigation into racing Thoroughbreds found that exposure to the smaller respirable dust was statistically associated with equine asthma, and that speed figures were reduced by 2.9 and 1.2 points for each percent in mast cell and neutrophil proportions, respectively, present in bronchoalveolar lavage (BAL) samples. ⁵³ In contrast, another racehorse study did not find associations between mucous scores obtained from BAL and performance, but they did find higher mucous scores were more likely as AQI worsened. ⁵⁴ A final additional study evaluated 12 polo horses naturally exposed to wildfire smoke and elevated PM_2.5, the authors concluded that improved environmental conditions, rather than therapy, reduced their clinical signs and increased their athletic performance as measured by treadmill speed and VO₂ peak. ⁵⁵ Given the potential impact of air quality on equine performance and welfare, it is crucial to undertake as many feasible measures to mitigate the cumulative burden of air pollution. To reduce poor outdoor air exposure, check the AQI before exercising horses and modify training routines during periods of elevated air pollution. For indoors, soak or steam hay, use low‐dust bedding and minimise high‐pressure blowers for barn cleaning. These measures combined will decrease the overall exposure of air pollutants and its potential cumulative burden on the respiratory tract.

While our findings show a negative association for certain threshold of PM_2.5 (cutpoints 4–11 μg/m³) and speed, several uncertainties persist within our results. Notably, the reasons for the speed increases beyond PM_2.5 cutpoint of 12 μg/m³ remain unclear. One possible explanation is that these estimates lack reliability due to their smaller sample sizes and wider confidence interval, suggesting higher variability in the data subset. It could also be speculated that horses were rested or had reduced exercise leading up to race day due to the poor air quality leading to improved times on race day. A concentration of PM_2.5 of 12 μg/m³ would equate to an AQI of approximately 50. This could trigger some trainers to modify training. Also unexpectedly, all other air pollutants showed a positive association with speed. The positive associations could be due to delayed immunological processes responsible for systemic inflammation. ⁵⁴ , ⁵⁶ These uncertainties highlight the complex relationship between ambient air pollution and performance.

In addition to the uncertainties, the existing body of research on this subject is characterised by inconsistency in findings, emphasising the need for further studies to attain a comprehensive understanding of this relationship. Future investigations could explore lag effects to determine if immunological processes delay air pollutant impacts on speed. A previous study has found race day performance in humans impacted by ambient air pollution exposure 21‐days prior to competition for higher PM_2.5 concentrations (10 μg/m³ vs. 5 μg/m³). ⁵⁷ Additionally, studies could incorporate an analysis of training practices, as we expect that air pollutant exposure during training periods may be more intense and prolonged compared with race events. Lastly, to improve the generalisability of our findings, future studies could characterise exposure levels at various racetracks in other locations where pollutants may change. For example, high levels of SO₂, a byproduct of coal and diesel combustion, are found in the Northeastern United States but generally not in California.

Our study distinguishes itself by having a large sample size that spans over 10 years, incorporating multiple air pollutants and meteorological data. Additionally, it is the first study to incorporate repeated measures, racetrack‐level covariates and horse‐level covariates, allowing for a more comprehensive analysis. Also, our approach of examining PM_2.5 in a non‐linear fashion is noteworthy. Overall, these methodological strengths enhance the credibility and depth of our research findings. Limitations of the study include uses of offsite air pollutant and meteorological monitors, potentially affecting the precision of exposure measurements. Reliance on federal monitors for data collection resulted in missing observations. Using average daily PM_2.5, MDA8 O₃, and average daily PM₁₀ measurements instead of hourly ones may have reduced precision. Further, we could not discern handicapped races in our data, potentially affecting model precision. Lastly, we did not include nutrition, training regiment and management practices in our analysis.

7. CONCLUSION

In conclusion, our analysis of over 31 407 Thoroughbred races in California found that PM_2.5 levels below the national ambient air quality standards negatively impact the speed of winning horses in Thoroughbred races. As PM_2.5 continues to degrade air quality in the Western United States due to wildfire activity, horses training and competing in these areas will be increasingly exposed. There is a need for continued scientific work to build awareness, develop recommendations and management options for the health and welfare of the equine athlete.

FUNDING INFORMATION

This work was supported by a College Research Council award from the College of Veterinary Medicine and Biomedical Sciences, Colorado State University (Duncan, PI).

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

AUTHOR CONTRIBUTIONS

Linda D. Kim: Writing – original draft; visualization; formal analysis; software; data curation; methodology. Kimberly Kreitner: Writing – review and editing. Danielle M. Scott: Writing – review and editing. Katie Seabaugh: Conceptualization; funding acquisition; writing – review and editing. Colleen G. Duncan: Conceptualization; investigation; funding acquisition; supervision; writing – review and editing; project administration. Sheryl Magzamen: Conceptualization; investigation; funding acquisition; writing – review and editing; supervision; formal analysis.

DATA INTEGRITY STATEMENT

This was a secondary analysis; no original data were produced as part of this study.

ETHICAL ANIMAL RESEARCH

All data included in this analysis were secondary data and did not involve any interaction with animal subjects. Institutional IACUC approval was not required.

INFORMED CONSENT

Not applicable.

PEER REVIEW

The peer review history for this article is available at https://www.webofscience.com/api/gateway/wos/peer-review/10.1111/evj.14415.

Supporting information

Figure S1. The grid shows all combinations of missing as red bars. For example, 2959 observations had only PM₁₀ missing, and 142 observations had a combination of NO₂ and temperature missing. In total, there are 31 407 complete observations.

Figure S2. Number of missing air pollutant observations by track.

Figure S3. Results from Spearman correlation tests in a heatmap for variables in the final regression.

Figure S4. Model diagnostics plots of the final regression using full data. The green line from the linearity plot is not flat and horizontal, suggesting non‐linearity.

Figure S5. The full dataset (n = 31 407) with extreme values was divided into two subsets repeatedly based on the specified daily average PM_2.5 cutpoint levels as shown on the x‐axis. We applied the full regression model to these series of data subsets for different ranges of PM_2.5 showing the complex relationship between speed and PM_2.5. Each point (yellow circle or blue triangles) with 95% confidence intervals (bars) shows the change in speed for every increase of PM_2.5 for each data subset. The estimates from the yellow circles came from data subsets that included PM_2.5 ranging between 0 μg/m³ to the indicated cutpoint level. The estimates from the blue triangles came from data subsets that included PM_2.5 between at cutpoint level to 23.6 μg/m³ of PM_2.5. Overall, changes in speed do not vary for different ranges of PM_2.5 concentrations indicated by confidence intervals crossing the null (red line). For the low ranges of PM_2.5 (values starting at 0 μg/m³ or yellow circles), the estimates show a gradual decrease in speed. For the higher ranges of PM_2.5 (values starting at specified cutpoint or blue circles) especially cutpoints 4–11 μg/m³ of PM_2.5, are negative, but not statistically significant.

Figure S6. After including racetrack as an interaction term in the regression model for full data with extreme values, we found racetracks modify the relationship between speed and PM_2.5. Each racetrack has a different change in speed estimate for every increase of PM_2.5. All racetracks ran slower than reference racetrack SR, a track with low average pollutant concentrations across all air pollutants in the study. Seven racetracks have statistically significantly different changes in speed, demonstrated by non‐overlapping confidence intervals, compared with the reference racetrack SR. Racetracks DMR, FER, FPX and LA showed racehorses ran slower as PM_2.5 concentrations increased.

Table S1. Summary of California tracks (n = 12) code names and distribution for all winning races (n = 31 407).

Table S2. Summary of pollutant monitor distances (km) from track.

Table S3. Generalised variance influence factor (GVIF).

Table S4. Speed (m/s) coefficients and 95% confidence intervals with full data.

Table S5. Full regression model estimates using full data.

Table S6. Summary of average pollutant concentration by track.

Kim LD, Kreitner K, Scott DM, Seabaugh K, Duncan CG, Magzamen S. The effects of ambient air pollution exposure on Thoroughbred racehorse performance. Equine Vet J. 2025;57(3):712–722. 10.1111/evj.14415

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are not available from the author: Open sharing exemption granted by the editor as data are owned by a third party.