The impact of air pollution on petcare utilization

Significance

The vast majority of the estimated benefits of policies to reduce air pollution come from decreases in human morbidity and premature mortality. Yet despite a wealth of evidence on the impacts of air pollution on human health, the effects on animal health remain poorly understood. We provide large-sample empirical evidence of the impact of fine particulate matter pollution on pet health. Our estimates indicate a positive relationship between air pollution levels and veterinary visits for both cats and dogs, highlighting that veterinarians and pet owners should be mindful of the risks of poor air quality for companion animals. These findings can also help in quantifying the pet health benefits from policies to improve air quality.

Abstract

Air pollution is one of the leading causes of morbidity and premature mortality globally. A large literature documents the adverse impacts of ambient air pollution on human health. In contrast, there is a lack of comparable research studying the effects of air pollution on animal health. We fill this gap, utilizing 5 y of data on over seven million visits to veterinary practices across the United Kingdom. Leveraging within-city variation in daily monitor-measured air pollution levels, we find that increases in fine particulate matter (i.e., PM_2.5) are associated with significant increases in the number of vet visits for both cats and dogs. In aggregate, these estimates suggest that reducing ambient PM_2.5 levels to a maximum of 5 mg per cubic meter as recommended by the World Health Organization would result in a 0.7 to 2.5% reduction in vet visits.

One in every six human deaths in 2019 was attributed to air pollution (1). The primary driver of air pollution-related mortality is exposure to fine particulate matter (2–5). A large body of empirical evidence documents the link between increased exposure to fine particulate matter and a host of human health outcomes, including respiratory, cardiovascular, and neurological conditions that can lead to increased emergency department visits, hospitalizations, and premature mortality (6–12). In contrast, empirical research on the effects of air pollution on animal health remains limited, with no studies to date investigating impacts on pet health using large-scale population data. The limited existing evidence is primarily correlational, relying on associations observed in small-scale observational or clinical studies (13, 14). This is despite humans and pets sharing many of the same biological pathways and exposure levels that lead to morbidity and mortality (15).

Pet owners place particular value on the well-being of their pets. In 2022, households spent about $120 billion on pets in the United States and £10 billion in the United Kingdom, with spending patterns in pet healthcare closely reflecting those in human healthcare (16–18). The well-documented negative effects of air pollution on human health, coupled with the large pet population and the substantial economic resources devoted to their care, underscore the need for a rigorous evaluation of air pollution’s impact on pet health.

We use visit-level data from an extensive sample of veterinary practices across the United Kingdom to estimate the impact of fine particulate exposure (i.e., PM_2.5) on the utilization of pet healthcare. The sample includes seven million vet visits for cats and dogs over five years, which we combine with data on ambient air quality from nearby pollution monitors. Estimates from panel data regressions that leverage daily variation in air pollution and control for a wide range of confounding factors indicate that a one microgram per cubic meter increase in average PM_2.5 over the preceding week corresponds to a 0.7% increase in vet visits for both cats and dogs. This implies that moving from a low pollution period, with PM_2.5 below 5 $\mathrm{\mu }$g/m³, to a high pollution period, with PM_2.5 of 40 to 50 $\mathrm{\mu }$g/m³, translates into a 30 to 40% increase in expected total vet visits. The effects we find for pets are of a similar order of magnitude to studies that have looked at human health and hospitalizations (11, 19, 20).

We further substantiate these findings by demonstrating that the estimates remain similar when using an instrumental variables (IV) approach that helps mitigate concerns about possible measurement error in our measure of air pollution. In this IV framework, we focus on variation in PM_2.5 concentration levels driven by thermal inversions and changes in wind direction. Thermal inversions occur when atmospheric temperatures increase with altitude, inhibiting the vertical dispersion of air pollutants and trapping them near the surface. Changes in wind direction influence local air quality by either dispersing locally generated pollutants or transporting pollutants from distant sources. These weather phenomena are unlikely to be related to veterinary visits except through their effects on air quality (11, 21, 22).

In aggregate, our findings suggest that reducing ambient PM_2.5 levels to a maximum of 5 $\mathrm{\mu }$g/m³ as recommended by the World Health Organization (WHO) would result in a reduction in vet visits of 0.7% to 2.5% (approximately 80,000 to 290,000 fewer vet visits each year in the United Kingdom) (23). This entails an annual savings in petcare utilization costs, owner time, and travel costs of roughly 19 to 66 million pounds. Importantly, this is only some of the many benefits of reducing air pollution enjoyed by pets and their owners. For example, our annual savings do not encompass reductions in pet mortality and morbidity, owners’ emotional distress caused by pet sickness, or increased time spent enjoying companionship due to improved pet lifetimes. The full economic benefits of improved air quality are likely substantially larger than those captured in our calculation, highlighting the need for further work to quantify and monetize the full benefits to pets and their owners (24, 25).

While the relationship between air pollution and animal health has long been recognized—and has informed epidemiological insights into the health effects of pollution on humans (26, 27)—most existing research has focused on documenting statistical associations rather than establishing causal links. Prior studies typically rely on small-scale observational data or clinical laboratory experiments (28–31), including a limited number of small-sample studies examining pet cats and dogs (13, 14). Only a few studies to date have leveraged quasi-experimental variation in air pollution exposure to estimate its effects on wildlife and farm animals—such as on bird abundance (32) or dairy cow mortality (33). However, to our knowledge, no study utilizing quasi-experimental variation in air pollution exposure has focused on the large and economically significant pet companion population.

Here, we bring the same rigorous empirical methods that have been used in recent human health studies to provide evidence of the effect of air pollution on pet dogs and cats. Closest to this work in scale is research using the Italian National Canine Registry to examine the relationship between heavy metals pollution and the life expectancy of dogs (34). We contribute to this small literature by providing large-scale empirical analysis of the health impacts of air pollution on pets using econometric methods and data on 3.8 million unique cats and dogs. The estimated effects yield insights into how environmental stressors influence the provision of pet healthcare and offer empirical evidence that can be used to quantify the value of pet health improvements in regulatory impact analyses of air quality. The omission of animal welfare in such evaluations has been termed an “inexcusable gap,” highlighting the importance of further research in this area (24).

Results

All-Cause Vet Visits.

Fig. 1 presents estimates of the effect of PM_2.5 on the number of veterinary visits across all causes. In all cases, the estimates reflect the effect of an increase in weekly average PM_2.5 for the preceding 7-d period, including the date of the visit. We estimate the effect of PM_2.5 on vet visits separately for cats and dogs. Considering only air pollution on the seven preceding days reflects that scheduling veterinary visits in the United Kingdom is generally straightforward, with many appointments available on the same day or the following day.

Fig. 1.

This figure presents Poisson regression estimates and 95% CI of the impact of PM_2.5 concentration levels on the count of vet visits. The measure of air pollution used is daily values for the rolling weekly average up to that day. All regressions include controls for pet age, sex, and weather, as well as NUTS3 region and day-of-sample fixed effects. We utilize the binscatter method to show how visits change as pollution increases. We estimate separate Poisson regressions for each species. We weight our regressions by the population of each NUTS3 region. SE are clustered at the NUTS3 region level.

Since the dependent variable is the count of vet visits, we estimate the relationship using a binscatter Poisson regression model (35). This approach flexibly estimates the relationship between the number of vet visits and PM_2.5, while controlling for differences in pet age, sex, and weather. We also include NUTS3 region and day-of-sample fixed effects to control for all observed and unobserved factors that predict vet visits and vary across NUTS3 regions (e.g., persistent differences in socioeconomic status across region) or across days (e.g., seasonality of vet visits and overall economic trends). Each circle marker corresponds to a point estimate and the whiskers display the 95% CI for each estimate. We provide further details on the data sources and methodology in SI Appendix.

For both species, the effect of PM_2.5 on log expected number of visits is linearly increasing and statistically significant, with a clear trend of more visits on more polluted days. For example, moving from a low pollution period, with PM_2.5 below 5 $\mathrm{\mu }$g/m³, to a high pollution period, with PM_2.5 of 40 to 50 $\mathrm{\mu }$g/m³, translates into a 30 to 40% increase in expected total vet visits. Estimates at the upper end of the PM_2.5 distribution are noisier (though still statistically significant), and more divergent from the fitted line.

Table 1 reports point estimates from a Poisson regression model that controls for the same covariates and fixed effects as in Fig. 1 but fits a linear relationship between log expected vet visits and PM_2.5 (as suggested by Fig. 1). In the main specification (columns 1 and 2), we find that a one microgram per cubic meter increase in PM_2.5 corresponds to a 0.7% increase in daily expected number of visits for both cats and dogs ($P<0.05$).^*

Table 1.

This table presents Poisson regression estimates of the impact of PM_2.5 concentration levels on vet visits

Model:	(1)	(2)	(3)	(4)
Variables
PM2.5	0.0070∗ (0.0038)	0.0071∗∗ (0.0036)	0.0070∗∗∗ (0.0023)	0.0055∗∗∗ (0.0021)
Species	Cat	Dog	Cat	Dog
IV	No	No	Yes	Yes
Age controls	Yes	Yes	Yes	Yes
Sex controls	Yes	Yes	Yes	Yes
Weather controls	Yes	Yes	Yes	Yes
Dep. Var. Mean	5.042	13.45	5.042	13.45
Fixed-effects
NUTS3 area	Yes	Yes	Yes	Yes
Date	Yes	Yes	Yes	Yes
Fit statistics
Observations	355,705	355,872	355,705	355,872
Squared correlation	0.82377	0.84239	0.82377	0.84239
Pseudo R2	1.3457	1.1118	1.3457	1.1118
BIC	$6.2\times {10}^{11}$	$1.04\times {10}^{12}$	$6.2\times {10}^{11}$	$1.04\times {10}^{12}$

Signif. Codes: ***: 0.01, **: 0.05, *: 0.1.The dependent variable is the daily count of vet visits in a given NUTS3 region. The measure of air pollution used is daily values for the rolling weekly average up to that day. All regressions include controls for pet age, sex, and weather, as well as NUTS3 region and day-of-sample fixed effects. Dependent variable means are included in the table. We estimate separate Poisson regressions for each species. We also estimate Poisson regression specifications instrumenting for air pollution using wind direction and thermal inversion events (Columns 3 to 4). We weight our regressions by the population of each NUTS3 region. SEs are clustered at the NUTS3 region level.

We also estimate the impact of PM_2.5 on vet visits using an instrumental variable estimator that leverages variation in air pollution driven by thermal inversions and wind direction. Here, we find a similar effect: a one microgram per cubic meter increase in PM_2.5 increases expected total visits by 0.6 to 0.7%. As well as being more precisely estimated, the similar magnitude of the estimated effect provides additional reassurance that our baseline estimates are attributable to fine particulate matter (as opposed to other factors) and are not biased by measurement error.

These estimated effect sizes are of a similar order of magnitude to those found in human health studies on air pollution and hospitalizations. Systematic reviews have found that a 10 $\mathrm{\mu }$g/m³ increase in daily PM_2.5 leads to a roughly 1% increase in all-cause hospital admissions (19, 20).^† A more recent study that uses a similar empirical approach to ours found that a 10 $\mathrm{\mu }$g/m³ increase in daily PM_2.5 leads to a 2.2% increase in 1-d all-cause hospitalizations among US Medicare recipients (11). If we convert our estimates into comparable units, they suggest that an equivalent 10 $\mathrm{\mu }$g/m³ increase in weekly average PM_2.5 leads to a roughly 7% increase in expected all-cause veterinary visits. When estimating our regressions using contemporaneous daily average PM_2.5 instead of the weekly rolling average, we find an effect size of 2% (SI Appendix, section 3).

Vet Visits by Main Presenting Complaint.

Fig. 2 examines how ambient PM_2.5 affects the number of visits depending on the main presenting complaint (MPC). As in Table 1, we report the estimated effects in proportionate terms. Here, we see that the effect of PM_2.5 on overall vet visits is primarily driven by the “Other Unwell” category: a one microgram per cubic meter increase in PM_2.5 corresponds to a 1.3% increase in expected daily visits classified as “Other Unwell” for cats, and a 1.2% increase for dogs. Estimates from instrumental variables specifications reveal broadly similar effects and can be found in SI Appendix, section 3B.

Fig. 2.

This figure presents Poisson regression estimates and 95% CI of the impact of PM_2.5 concentration levels on the count of vet visits by MPC. The measure of air pollution used is daily values for the rolling weekly average up to that day. All regressions include controls for pet age, sex, and weather, as well as NUTS3 region and day-of-sample fixed effects. We estimate separate Poisson regressions for each species and for each MPC. We weight our regressions by the population of each NUTS3 region. SE are clustered at the NUTS3 region level.

The fact that we can detect a statistically significant effect of PM_2.5 on “Other Unwell” visits is plausible given that visits for this cause make up roughly a quarter of the total number of visits. Moreover, “Other Unwell” encompasses visits that are not easily categorized into the other more specific options provided, such as vaccinations and trauma. The main impacts of air quality on human health focus on the exacerbation of respiratory, cardiovascular, and neurological conditions. Since there is no MPC category for cardiovascular or neurological problems, it makes sense that pets with symptoms originating from these sources may be classified in the “Other Unwell” category.

There is a MPC category for “Respiratory” visits, and here we find no significant effect of PM_2.5. However, only about 1% of the visits are classified as Respiratory, indicating that respiratory conditions are rarely classified as the main reason a pet is brought in to the vet. We also do not find large statistically significant effects in the categories of MPC that almost certainly do not have any epidemiological relationship with air pollution exposure. For example, we do not see consistent statistically significant increases in vet visits for “Vaccination” or “Post-Op” on more polluted days.

However, the reporting of MPC is likely subject to substantial measurement error. Given this is a voluntary additional piece of data to enter after the visit, veterinarians have limited incentive to report MPC in a consistent and rigorous manner. This may help explain the large proportion of visits that are classified as “Other Unwell” or “Other Healthy.”

Vet Visits by Pet Age.

Much of the literature on how air pollution affects human health highlights that vulnerability varies across individuals, with age being a key risk factor (11, 36, 37). We therefore explore heterogeneity in our effects by age group. Figure 3 provides age-group-specific estimates of the effect of PM_2.5 on all-cause visits in proportionate terms. Though we observe an increase in effect size with age, we lack the statistical precision to detect differences across age group bins. Versions of these results using instrumental variables reveal broadly similar effects and can be found in SI Appendix, section 3B.

Fig. 3.

This figure presents Poisson regression estimates and 95% CI of the impact of PM_2.5 concentration levels on the count of vet visits by age group. The measure of air pollution used is daily values for the rolling weekly average up to that day. All regressions include controls for pet age, sex, and weather, as well as NUTS3 region and day-of-sample fixed effects. We estimate separate Poisson regressions for each species and for each age group. We weight our regressions by the population of each NUTS3 region. SE are clustered at the NUTS3 region level.

We cannot rule out that pet owners may combine multiple purposes into a single visit—such as addressing both acute health concerns triggered by air pollution and routine issues like vaccinations. However, vaccinations—the most likely source of schedulable visits—are heavily concentrated among younger pets, especially in the first year of life. If substitution from low-pollution days to high-pollution days was driving our effects, we might expect larger effects of PM_2.5 on vet visits among younger pets for whom vaccinations can be rescheduled. This is not what we observe in Fig. 3, suggesting that our estimates do not stem primarily from substitution.

Discussion

Using detailed data on more than seven million visits to veterinarian clinics spanning the United Kingdom, we uncover the extent to which ambient fine particulate concentration levels impact petcare utilization. We find that increases in PM_2.5 concentration levels are associated with significant increases in vet visits for pet cats and dogs. In aggregate, our estimates suggest that reducing ambient PM_2.5 levels to a maximum of 5 $\mathrm{\mu }$g/m³ as recommended by the WHO would result in a reduction in vet visits in the United Kingdom of 0.7 to 2.5% (approximately 80,000 to 290,000 fewer vet visits each year in the United Kingdom).^‡

Understanding the total economic value of these kinds of improvements in pet health is challenging. Clearly pet owners devote significant time and economic resources to caring for their pets—the petcare market has grown by over 66% in the last decade, significantly outpacing growth in the wider economy (39). If we focus purely on petcare utilization costs, the resulting savings to UK pet owners from moving into compliance with the WHO standard are roughly £18 to 63 million pounds per year.^§ Adding in the travel and time costs associated with vet visits adds a further £1 to 3 million pounds per year.^¶

However, the total economic benefits are likely to be considerably higher. This is because petcare utilization costs are only a small portion of the total willingness-to-pay to improve pet health. Despite an extensive literature on the economic value of health benefits for humans (42), it is striking that there is little equivalent research on the value people place on improving the health of their pets (24). There is only one study we are aware of that has examined this directly by estimating a contingent-valuation-based value of statistical dog life (43). Related work has also highlighted the significant value people place on protecting wild animals, including specific “charismatic” animal individuals (44, 45).

Here, we find that percentage increases in pet visits to veterinary clinics increase linearly in PM_2.5, providing an estimate that can be translated into a concentration–response function necessary to quantify the benefits of air quality improvements (46, 47). Further work is needed to better characterize the pet health—air pollution relationship for other health endpoints, including premature mortality. Ultimately, this growing body of evidence will support rigorous cost–benefit analyses of policies aimed at enhancing animal well-being, including that of companion animals (24, 25). In addition, further empirical evidence on pet health impacts may increase the salience of air pollution and other risks to pet health and change pet owner behavior.^#

Our research design faces several challenges and we want to acknowledge its limitations. First, daily average fine particulate concentration levels are highly autocorrelated: high pollution levels on one day are likely to be followed by higher levels on the following day. To assuage this concern, we utilize weekly average PM_2.5 levels and the same quasi-experimental methods used in recent studies of air pollution and human health (e.g., refs. 11, 21, and 22). Nevertheless, SI Appendix, Table S10—which presents estimated effects of weekly lags and leads of air pollution on vet visits—illustrates that autocorrelation in PM_2.5 may limit the extent to which we can interpret our primary estimates as reflecting the effects of solely last week’s pollution.

Second, PM_2.5 concentration levels are also correlated with the concentration levels of other air pollutants such as ozone and sulfur dioxide. For example, among other pollutants, burning coal emits sulfur dioxide, nitrogen oxides, and fine particulate matter (49). This “multipollutant” problem can make it challenging to isolate the causal impacts of fine particulate matter on human and animal health (50).

Third, while we study the relationship between PM_2.5 and vet visits, we are limited in our ability to identify the exact physiological pathways that underpin these changes in vetcare utilization. More detailed data on the illnesses pets experience and the exact treatments provided could help shed further light on these pathways. Measuring the effects of air pollution on pet mortality—ideally using data comparable to the human vital statistics datasets employed in prior studies—would also be of significant interest, if such data are available.

Finally, our analysis measures air pollution exposure using nearby readings of ambient outdoor air pollution. However, humans spend the majority of their time indoors. Indoor air quality is thus a key determinant of overall PM_2.5 exposure (23, 51). The same is true for pets, especially pets that remain exclusively indoors. While indoor and outdoor air quality are correlated, the two can diverge depending on building ventilation, air purification, and indoor sources of pollution such as cooking and heating (52).

Our focus on ambient air quality mirrors the existing research on human health, the vast majority of which also uses ambient outdoor measures of air quality (52). This is driven in part by the availability of data—pollution monitor and satellite derived measures of air quality are widely available whereas data from indoor pollution monitors are not. While our findings provide important insights into the impact of air quality on animal health, further research that focuses on the impacts of indoor air quality would be valuable.

Materials and Methods

Data.

The empirical analysis leverages daily visit-level data from veterinary practices across the United Kingdom over the period January 2017 to September 2022. The data are taken from the Small Animal Veterinary Surveillance Network (SAVSNET) database (53), which is administered by the University of Liverpool. The database includes visit information from around 10% of the five thousand veterinary practices in the United Kingdom. A more detailed description of SAVSNET can be found in ref. 54. We provide additional detail on the SAVSNET data in SI Appendix, section 1.

For each consultation/visit, the data include unique IDs for the veterinary practice and the pet, the species of the pet, age group, sex, the date and time of the visit, location of the practice at the NUTS3 region level (similar in granularity to U.S. counties), and the practitioner-derived MPC.^‖ We focus on cats and dogs, the two species that make up the bulk of vet visits. The resulting estimation sample contains data on approximately 3.8 million unique cats and dogs.^** We have 1.9 million visits for cats and 5 million visits for dogs.

We combine the vet visits data with hourly readings from air pollution monitors across the United Kingdom from UK Air, accessed via Openair (55, 56). The locations of the air quality monitors in the sample and figures presenting the geographic variation in average PM_2.5 concentration levels in the sample are shown in SI Appendix, section 1. We calculate air pollution levels first for each of the almost 10,000 middle layer super output areas (MSOAs) in the United Kingdom. These are census geographies each with a population of approximately 7,000 people (similar to US census tracts). For each MSOA, we take the inverse distance weighted average of the three nearest pollution monitors within 50 km of the centroid of the area.^†† We then calculate NUTS3 region level average pollution by taking the population-weighted average across all the MSOAs in each region. Full details can be found in SI Appendix, section 1.

In summary, we construct a panel dataset to study the effect of weekly average ambient PM_2.5 concentration levels on the daily number of veterinary practices visits in each NUTS3 region. Importantly, the longitudinal nature of the database allows us to eliminate the confounding effect of time-invariant unobserved determinants of pet health in each region (such as average economic status and long-term spatial differences in air pollution levels), and unobserved time-varying factors that are common across regions (such as recessions and nationwide reductions in air pollution levels).

We also control for daily temperature and precipitation in the empirical specifications, as weather can impact animal health and pet owners’ decisions to visit a veterinary practice. We compile hourly data on temperature and other meteorological variables from the ERA-5 reanalysis dataset, which provides consistent global estimates on a $0.{25}^{°}\times 0.{25}^{°}$ grid (resolution of approximately 25 to 30 km) (57, 58). In addition to controlling for weather in the models underlying the primary empirical analysis, we also use the weather data to identify the presence of thermal inversions and wind direction. In robustness checks, we employ both thermal inversions and wind direction as instrumental variables to generate exogenous variation in ambient PM_2.5 levels.

Table 2 provides summary statistics. The estimation sample contains roughly 1.9 million visits for cats and 5 million visits for dogs. Cats exhibit an older age distribution than dogs, reflecting their longer average lifetimes. When breaking visits out by the MPC that is recorded in the vet notes, the most common type of visit is for “Vaccination”, comprising 31% of visits. Another 47% of visits fall into the broad categories of “Other Healthy” and “Other Unwell”. The remainder of visits are made up of a variety of more granular categories related to issues such as “Trauma,” “Post-Op”, “Kidney Disease,” and so on. Visits classified with a MPC of “Respiratory” comprise only around 1% of visits.

Table 2.

This table contains summary statistics on pet veterinary visits data

	Cat	Dog
Sex: Female	0.510	0.487
Age: 0 to 4	0.320	0.371
Age: 12 to 16	0.177	0.113
Age: 16$+$	0.089	0.006
Age: 4 to 8	0.208	0.263
Age: 8 to 12	0.206	0.247
MPC: Gastroenteric	0.020	0.031
MPC: Kidney Disease	0.008	0.003
MPC: Other Healthy	0.264	0.276
MPC: Other Unwell	0.202	0.194
MPC: Post-Op	0.062	0.074
MPC: Pruritus	0.023	0.051
MPC: Respiratory	0.012	0.009
MPC: Trauma	0.045	0.043
MPC: Tumor	0.011	0.018
MPC: Unknown	0.002	0.002
MPC: Vaccination	0.350	0.299
N	1,861,334	4,960,176

An observation is a unique vet consultation event. All variables shown are indicator variables and the values shown are the means across the sample. MPC stands for “Main Presenting Complaint.”

Methods.

The goal of the empirical analysis is to determine whether there is a relationship between local air quality and the frequency of vet visits. We therefore aggregate the visit-level data to obtain a daily count of visits by species for each NUTS3 geographic region. Information on sex and age group category is converted to shares. We do this by dividing the count of visits for each sex or age group category for a given day and region by total all-cause visits for the same day and region. We also conduct versions of the analysis where we examine counts of visits for a specific MPC.

Since the number of visits to a vet clinic is a count variable, we use a panel Poisson regression model relating the daily total number of visits in a NUTS3 region ($N_{i,t}$) to daily average PM_2.5 concentration levels:^‡‡

$$\begin{array}{c}\hfill log(E[N_{i,t}|Z_{i,t}])={\alpha }_i+{\theta }_t+\beta {\text{PM2.5}}_{i,t}+\gamma X_{i,t}\end{array}$$ [1]

for each NUTS3 region $i$ and day $t$. We include age, sex, and weather controls (X_i,t), as well as NUTS3 fixed effects (${\alpha }_i$) and day-of-sample fixed effects (${\theta }_t$) in all specifications. The NUTS3 region fixed effects control for all time-invariant confounders that vary at the NUTS3 region level (e.g., persistent differences in socioeconomic status across region). The day-of-sample fixed effect control for unobserved time-varying confounders that are common across all NUTS3 regions (e.g., seasonality of vet visits and overall economic trends). The weather controls include precipitation and 2 ^°C bins of temperature. We weight our regressions by the population of each NUTS3 region to ensure that the sample is more representative of the geographic distribution of the population of pets.^§§

The weather controls are relevant not just to account for their direct effect on pet health, but also to control for their role in determining pet owners’ decisions to take their pet to the vet. For instance, we observe a reduction in vet admissions on very hot or very cold days, suggesting that pet owners are less likely to take their pet to the vet on those days. Pet owners may also avoid vet visits on high pollution days, depending on the salience of air pollution and its health effects. Consequently, our estimated effects of air pollution on vet visits should be interpreted as being net of this avoidance behavior.

The primary independent variable of interest is daily regional fine particulate concentration levels (i.e, PM2.5_i,t). Consistent with other studies on air pollution, we do not just focus on air pollution on the day of visit (11). Instead, in our preferred specification, we consider rolling averages of PM2.5_i,t in order to better measure sustained recent exposure to poor air quality. Namely, in Eq. 1, we define PM2.5_i,t as the rolling average of daily PM2.5 over a seven-day period ending with day $t$. In SI Appendix, section 3, we also consider models based only on contemporaneous exposure (i.e., the relationship between vet visits on date $t$ and PM_2.5 on date $t$ only). The estimated effects are noisier and smaller in magnitude than those reported in Table 1. This suggests that our main estimates are not inflated by short-term displacement effects (11).

Regressions are estimated separately for each species $s\in \{\text{cat},\text{dog}\}$ and outcome variable (i.e., all cause admissions, admissions by MPC, and admissions by age group). SE are clustered by NUTS3 region to account for autocorrelation in the error term across days within each geographic area.

As a robustness check, we also consider an instrumental variables approach in which PM_2.5 levels are instrumented with thermal inversions and wind direction. Both thermal inversions and wind direction are widely employed as instruments for air pollution (11, 21, 22). These meteorological phenomena can generate large fluctuations in air pollution exposure. Indeed, in our case, thermal inversions and wind direction are strong predictors of ambient PM_2.5, even conditional on the controls and fixed effects. These fluctuations are unlikely to be directly related to pet healthcare decisions, except through their effect on air quality—a necessary assumption to interpret estimates from the IV framework as causal. Our instruments are constructed using ERA-5 reanalysis weather data. Further details on the instrumental variables framework can be found in SI Appendix, section 4.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

The main dataset used for this paper contains information on individual veterinary consultations and so cannot be publicly shared in its raw format. The SAVSNET data are collected and maintained by the University of Liverpool and are available to anyone that wishes to apply for access (53). All remaining data used in the analysis (air pollution, weather etc.) and a full code file have been deposited on Zenodo (60).