Assessing the Impact of Wildfire Emissions on the Seasonal Cycle of CO and Emergency Room Visits in Alberta and Ontario, Canada

Abstract

Exposure to wildfire smoke is a well‐known concern for public health and is anticipated to worsen with an increase in wildfire activity related to climate change. This study uses satellite and ground‐based carbon monoxide (CO) measurements from 2004 to 2019 to evaluate a change in its seasonal cycle due to wildfire emissions. Monthly average CO total columns from the Measurements of Pollution in the Troposphere (MOPITT) satellite instrument over Alberta and Ontario, and from a ground‐based Fourier transform infrared spectrometer in downtown Toronto are compared before and after 1 January 2012, following previous literature. Between the two time periods, a new peak emerges in the seasonal cycle of CO, centered around August. Monthly emergency room admissions from Alberta and Ontario for nine cardiovascular and respiratory diseases are assessed with a difference in difference analysis, using MOPITT CO as the exposure metric. This analysis was used to calculate the change in monthly hospital admissions per 100,000 people, given a 1 ppb increase in XCO post‐2012 compared to pre‐2012, along with the 95% confidence interval (CI). For Ontario, this term is positive and significant for hypertension (change = 1.88, CI = 1.18–2.57), ischemic heart disease (0.50, CI = 0.12–0.88), arrhythmia (0.12, CI = 0.03–0.22), and asthma (0.31, CI = 0.05–0.57). For Alberta, there is a significant and positive interaction for arrhythmia (0.48, CI = 0.12–0.85). These results indicate that there was a statistically significant increase in adverse health outcomes for five of the eighteen disease‐province pairings associated with the increase in atmospheric CO after 2011 coinciding with enhanced wildfire emissions.

Plain Language Summary

Previous literature has identified an increase of atmospheric carbon monoxide (CO) in the Pacific Northwest in August when considering the data before and after 2012, and linked this to forest fire activity in this region. In this study, a similar increase in CO was measured in the provinces of Alberta and Ontario, Canada using satellite data and in the city of Toronto using a ground‐based instrument. To determine whether the observed increase in CO has implications for public health, emergency room admissions for nine cardiovascular and respiratory diseases were examined over the same time period. The analysis used atmospheric CO as a measure of exposure and assessed differences between the pre‐2012 and post‐2012 periods. The findings are suggestive of a link between enhanced wildfire‐related CO concentrations after 2012 and worsening health outcomes, with statistically significant results for five of the eighteen disease‐province pairings.

Key Points

• The mean total column CO for August in the provinces of Ontario and Alberta increased between 2004‐2011 and 2012–2019 due to forest fires

• Hospital admissions for cardiovascular and respiratory diseases show worsening health outcomes after 2012 for five disease‐province pairs

• CO is demonstrated to be a potential metric for smoke exposure in wildfire health studies

1. Introduction

Long‐term monitoring of both environmental and health impacts of wildfires is increasingly important as both the frequency and severity are anticipated to rise in Canada (Hanes et al., 2019). Whitman et al. (2022) show that from 1970 to 2019, the province of Alberta had statistically significant increases in temperature and dryness, and a shift in fire regime characteristics whereby large fires began to dominate and re‐burn frequency increased. Season‐specific climate variables were shown to positively correlate with the fire activity, indicating that climate is a top‐down driver for wildfires in the region. In addition to climate change causing drier, hotter conditions and a lengthening of the forest fire season, fire suppression tactics have contributed to higher instances of extreme and uncontrollable burning events (Jolly et al., 2015; Kredier et al., 2024).

This study is motivated by the work of Buchholz et al. (2022), who found a change in the carbon monoxide (CO) seasonal cycle in the Pacific Northwest using the Measurements of Pollution in the Troposphere (MOPITT) satellite instrument. The change was observed when comparing the monthly average CO column from 2002 to 2011 with that from 2012 to 2018, revealing a new emerging peak in August that was attributed to an increase in wildfires in the area. The trend was also reported in the Central USA and Northeast North America. The work here builds on these findings to assess CO trends over the provinces of Alberta and Ontario, Canada and for the highly populated area of Toronto, Ontario, which experiences long‐range transport of air pollution from wildfires.

Over the decade from 2009 to 2019, the average burn area by province and territory is led by the Northwest Territories, then Saskatchewan, followed by Alberta, with Ontario ranking 8^th out of 13 in this list (CWFIS, 2024). Although Ontario, specifically the densely populated area of Southern Ontario, is less proximate to wildfire‐prone regions, it can still experience degraded air quality from long‐range transport of smoke plumes (Flood et al., 2025). Alberta has an active wildfire season locally, and also experiences long‐range transport from the surrounding provinces and territories. Health impacts are estimated to be greatest in areas with higher exposure to wildfire PM_2.5 and a large population (i.e., Alberta and British Columbia), however they are also relevant in areas with lower exposure levels (i.e., Ontario and Quebec) (Matz et al., 2020). Areas which are proximate to active fires will often have public health advisories for residents and intervention measures in place for vulnerable populations, while areas further away, which are not in the immediate disaster response area, may be less aware of their exposure (Magzamen et al., 2021).

In Canada from 2013 to 2015 and 2017 to 2018, acute health impacts from wildfire smoke exposure have been estimated to cost up to CDN$1.8 billion, while the cost of chronic health impacts may be as much as CDN$19 billion (Matz et al., 2020). In the same study, the continued exposure to wildfire PM_2.5 was estimated to reduce life expectancy in Canada (depending on the location) by 0.05–0.42 years. During the summer of 2015, the city of Calgary, Alberta experienced an episode of poor air quality resulting from the long‐range transport of wildfire smoke from wildfires in the Pacific Northwest. Outpatient data showed an elevated risk of respiratory disease morbidity of 33% during and after the event, related to increased PM_2.5 (Mahsin et al., 2022). Children (ages 0–9) were found to have an increased risk particularly for asthma, acute bronchitis and acute respiratory infection, while seniors were at a higher risk of congestive heart failure and ischemic heart disease (Mahsin et al., 2022). A study focused on an area of southeast British Columbia during the wildfire season of 2003, found an increase in respiratory‐related physician and hospital visits linked to wildfire smoke exposure using ground‐based PM₁₀ monitoring, a dispersion model, and satellite imagery (Henderson et al., 2011). The authors note the agreement between the results using PM₁₀ and using satellite monitoring, suggest that it can be effective exposure metric alternative to surface monitoring.

Studies commonly use PM_2.5 as a gauge for smoke exposure related to health effects, although Wettstein et al. (2018) identified the need for studies to incorporate other fire‐related emissions as exposure metrics. CO has been well established as a useful tracer for the transport of wildfire plumes using both space‐based measurements (e.g., Edwards et al., 2004; Pope et al., 2021) and with ground‐based measurements (e.g., Flood et al., 2025; Franklin et al., 2014; Lutsch et al., 2020; Viatte et al., 2013). Henderson et al. (2011) highlight the value of assessing acute and chronic exposure from wildfires using remote‐sensing data, particularly for rural and isolated areas where air quality monitoring may not exist. While correlations have been sought for long‐term exposure in proximate populations, it is important to also consider the impacts of the long‐range transport of pollutants from wildfires, which can be tracked using remote‐sensing data and has the advantage of providing both long‐term data sets and widespread coverage.

Based on the wildfire health impact studies in Canada discussed above, there is a deficiency in research that incorporates long‐term monitoring and long‐range effects, along with assessments that do not rely on the use of particulate matter (PM₁₀ or PM_2.5). This study aims to address this gap and contribute to a framework that may support similar studies in the future. To evaluate if the reported increase of wildfire‐related CO emissions has an impact on the health care system in Canada, emergency room (ER) admissions for cardiovascular and respiratory diseases are assessed over the same period discussed in Buchholz et al. (2022), as records are available. The objective is to assess whether emergency care utilization for respiratory and cardiovascular diseases in Ontario and Alberta, Canada is correlated with atmospheric CO from wildfires, and in turn, whether CO can be a useful exposure metric for health effects. The structure of this paper is as follows: Section 2 describes the data sets and the methods used to analyze them, Section 3 presents the results and discussion, and Section 4 provides conclusions.

2. Data Sets and Methods

2.1. CO Measurements

2.1.1. MOPITT

Operated by the Canadian Space Agency (CSA), the MOPITT instrument is on the Terra satellite, launched by National Aeronautics and Space Administration (NASA) in 1999. This nadir‐viewing instrument measures CO with local overpass times of 10:30 and 22:30 (Drummond et al., 1995). With a 22 km × 22 km footprint, there is global coverage approximately every 3 days. The MOPITT data used here are from the version 9, level 3 (L3) Joint thermal infrared (TIR) and near‐infrared (NIR) monthly product (Deeter et al., 2022). The joint TIR‐NIR product provides larger Degrees of Freedom for Signal (DOFS), better vertical resolution, and higher sensitivity to CO in the lower troposphere, compared to TIR or NIR alone (Deeter et al., 2013). Comparisons to CO vertical profiles measured by the National Oceanic and Atmospheric Administration (NOAA) using an airborne flask‐sampling system show that the MOPITT TIR‐NIR V9 product has a mean bias of 1.90% at the surface (Deeter et al., 2022). MOPITT validation using data from the Network for the Detection of Atmospheric Composition Change (NDACC) found that at Toronto the TIR‐NIR DOFS were 1.63 (over land), with a column averaging kernel that peaks in the mid‐troposphere and can be considered similar for other mid‐latitude areas (i.e., Ontario as a whole, and Alberta) (Buchholz et al., 2017).

MOPITT data are extracted for the areas which encompass the provinces of Alberta and Ontario (shown in Figures A1 and A2). To obtain a column‐averaged dry‐air mole fraction (XCO in ppb), the daytime retrieved total column CO is divided by the daytime dry‐air column, provided in the MOPITT data files. To account for the change in background atmospheric CO over the study period, not related to the wildfire enhancements that vary interannually, CO is detrended using the North American average change of −0.57% per year derived from 2002 to 2018 (Buchholz et al., 2021), prior to the seasonal cycle analysis, following the approach of Buchholz et al. (2022). To coincide with the available health data, MOPITT from 2004 to 2019 are used in this study.

2.1.2. FTIR Spectroscopy

The high‐resolution Fourier transform infrared (FTIR) spectrometer used in this work is located at the University of Toronto Atmospheric Observatory (TAO) in downtown Toronto (43.66°N, 79.4°W, 174 masl). The instrument records solar absorption spectra from which vertical profiles and columns of several atmospheric trace gases are retrieved using the SFIT4 code, based on the optimal estimation method (IRWG, 2020; Rodgers, 2000). Measurements started in 2002 and are contributed to the NDACC (De Mazière et al., 2018). Details of the CO retrieval can be found in Yamanouchi et al. (2023).

The TAO CO total column time series average DOFS for CO is 2.16 ± 0.29 and the average percent uncertainty is 2.97% (Yamanouchi et al., 2023). The column measurements are dominated by the troposphere, in both column density and retrieval sensitivity. This allows for CO enhancements measured at TAO to provide an indicator for the presence of wildfire smoke (e.g., Flood et al., 2025; Lutsch et al., 2016; Yamanouchi et al., 2020). For consistency with the MOPITT data, the TAO FTIR columns from 2004 to 2019 are made into monthly averages and detrended with the same North American average trend of −0.57% per year (Buchholz et al., 2021).

2.2. Health Data

The hospital ER data used in this work were provided by the Canadian Institute for Health Information (CIHI) National Ambulatory Care Reporting System (CIHI, 2024). The data were aggregated such that one count for each month in each health district was provided for each of the ICD‐10 (International Classification of Diseases 10th revision) code groupings (as listed in Table A1). The ICD 10 codes are of relevant cardiovascular and respiratory diseases, selected following Hahn et al. (2021) and Magzamen et al. (2021). The data contain no identifying information such as age or gender, and cell values less than 5 were suppressed by CIHI (a placeholder value of 4 was set, with only two instances of this).

There are five zones in the Alberta Health Service network (AHS, 2024) (North, Edmonton, Central, Calgary and South), and ER admission data were provided from 2010 to 2020 (as shown in Figure A1). There are 14 zones in the Public Health Ontario network (PHO, 2024) (Erie St. Clair, South West, Waterloo Wellington, Hamilton–Niagara Haldiman Brent, Central West, Mississauga‐Halton, Toronto‐Central, Central, Central‐East, South‐East, Champlain, North‐Simcoe‐Muskoka, North‐East, North‐West), and the data were provided for 2004 to 2020 (as shown in Figure A2). The ER counts were normalized by population (admissions per 100,000), to account for population changes over the time period studied (AHS and PHO, last access: 30 August 2024). Due to the large areas included in the regional health zones in each province (see Appendix Figures A1 and A2 for the regional maps), it was not possible to distinguish between urban and rural data. Data from 2020 were obtained but not included in the analysis due to anomalies in emergency care usage as a result of the COVID‐19 pandemic. The ER admissions are used as a point of reference for the health care utilization over time. Figure 1 shows the time series data for all nine diseases in each province to provide context for the admission rates in each province and complexity of the time series over time. The sudden increase in cerebrovascular disease and ischemic heart disease diagnoses likely reflects the transition from ICD‐9 to ICD‐10 codes in Canada (Walker et al., 2012) and should be interpreted as a data artifact rather than causally related to air pollution exposure. To limit possible biases in the results due to this, hospital data prior to 2009 for these two diseases in Ontario were omitted from the difference in difference (DiD) analysis.

Figure 1.

Time series of monthly hospital emergency room admissions per 100,000 for each province (Ontario in orange, Alberta in maroon), for the nine diseases studied.

2.3. Difference in Difference Analysis

A DiD analysis was applied to the hospital admissions data and MOPITT measurements. This approach compares a “pre‐treatment” and “post‐treatment” period, where the change in trends between the two periods is the “difference‐in‐differences.” In the absence of a change, the two regressions would remain the same. Unlike other DiD applications, the exposure metric and outcome metrics used in this work are not binary; although there are “before” and “after” periods, there is no available control group that did not experience the change in CO. The equation and parameters, presented in Table 1, are adapted from the DiD analysis in Card and Krueger (1993) and Zahran et al. (2014). The model assesses the monthly hospital admissions per 100,000 residents in each health district ($N_{mR}$), where exposure is the XCO (ppb) in a given month, with an interaction term ($i_{co}$) of before (0) and after (1) 1 January 2012 (corresponding to the timeframes discussed in Buchholz et al., 2022), and covariant controls for month (m) and health region (R). The coefficients are determined by using an ordinary least squares approach to minimize the sum of squared residuals with the change in $N_{mR}$, when only the respective variable is changed. The DiD was performed for each province individually. In short, we consider the pre‐2012 population by health region and province as the appropriate counterfactual for the post‐2012 (inclusive) population.

Table 1.

Summary of DiD Equation and Variables

Variable	Definition	Coefficient	Relevance
$N_{mR}={\beta }_0+{{\beta }_{1}X}_{CO}+{\beta }_{2}t+{\beta }_3{i}_{co}+{\beta }_{4}m+{\beta }_{5}R$
$N_{mR}$	Total admissions in a month for each health district (per 100,000 population)	–	–
–	–	${\beta }_0$	Intercept
$X_{CO}$	“Exposure,” detrended monthly mean from MOPITT in ppb	${\beta }_1$	The change in monthly hospitalizations per 1 ppb increase in XCO when the period is pre‐2012
$t$	Period: 0 for pre‐2012, 1 for post‐2012	${\beta }_2$	Change in hospitalizations post‐2012 versus pre‐2012 when XCO is held constant
$i_{co}$	CO interaction term (= $t\ast X_{CO}$)	${\beta }_3$	The monthly change in the effect of a 1 ppb increase in XCO on hospitalizations post‐2012 versus pre‐2012
m	Month indicator (only July, August, September)	${\beta }_4$	Change in hospitalizations per month, when the rest of the variables are held constant
$R$	Health region indicator	${\beta }_5$	Change in hospitalizations per health region, when the rest of the variables are held constant

The resultant is measured as hospitalizations (per 100,000), and the interaction term is dictated by before (2004 through 2011 for Ontario, and 2010 through 2011 for Alberta) versus after 2012 (2012 through 2019), to identify the difference in the hospitalization rate‐XCO relationship between the two time periods. The month and region are included to control for potential differences in XCO and hospitalizations that may occur regionally or temporally. Within the DiD, the ${\beta }_4$ and ${\beta }_5$ coefficients represent the change in hospital admissions compared to the month and region, respectively, when holding everything else constant. Since month and region do not interact with the other covariates, ${\beta }_4$ and ${\beta }_5$ can be interpreted independently of other variables in the model and are not relevant to the discussion of the relationship between XCO, hospital admissions, and period. Coefficient ${\beta }_1$ is related to the variable XCO and represents the change in hospital admissions per 1 ppb increase in XCO when the period is pre‐2012, and ${\beta }_3$ is related to the interaction term ($i_{co}$) and represents the change in the effect of 1 ppb increase in XCO on hospitalizations before 2012 versus after 2012 (inclusive). Therefore, the combined sum of ${\beta }_1$ + ${\beta }_3$ is the change in hospitalizations, per 1 ppb increase in XCO, in the post 2012 period. The ${\beta }_1$ and ${\beta }_3$ terms resulting from the DiD analysis are evaluated with two statistical parameters, the 95% confidence interval (CI) and the p‐value. The 0.025 and 0.975 range of the CI indicates the range at which 95% of the data is normally distributed, where a significant result will not contain a null value. The p‐value signifies the probability of the variable having no influence on the dependent variable ($N_{mR}$), and that the observed correlation happens by chance, where a lower value (<0.05) is indicative of a better relationship.

3. Results and Discussion

Following the findings in Buchholz et al. (2022), the analysis is performed for two periods, defined as “pre‐2012” (start to 31 December 2011) and “post‐2012” (1 January 2012 to end). All the data sets end with 2019 for consistency with the health data, and are analyzed from 2004 to match the available Ontario health data, with the exception of the Alberta hospitalization data, which are only available from 2010 onwards.

3.1. CO Seasonal Cycle

To assess if a change in the CO seasonal cycle is detected in the study regions, MOPITT and TAO FTIR CO data are examined pre‐2012 and post‐2012. Figure 2 shows the mean XCO measured by MOPITT in August over Canada, for the two corresponding time periods, 2004–2011 and 2012–2019. The post‐2012 period in Figure 2b shows large areas across the country with a higher monthly mean XCO than the pre‐2012 period displayed in Figure 2a. Figure 2c shows the difference between the two periods (2012–2019 minus 2004–2011), where yellow and orange represents a decrease in XCO, white is no change (0 difference), and red to black is an increase in XCO from the earlier period to the later period.

Figure 2.

MOPITT monthly mean August XCO for (a) 2004–2011 and (b) 2012–2019, and (c) the difference (2012–2019 mean minus 2004–2011 mean). The provinces of Alberta (AB) and Ontario (ON) are marked in panel (a).

To examine the differences for Alberta and Ontario, the detrended monthly mean XCO is plotted in Figure 3 by month, for 2004–2011 and 2012–2019, with ±1 standard deviation of the mean shaded. For both provinces, a larger value is seen in August (and a somewhat larger value in July and September) for the post‐2012 period, compared to pre‐2012, although with some overlap when considering the standard deviation of the mean. The monthly mean XCO for each year used in Figure 3 is plotted in Figure A3 to illustrate the interannual differences between the pre‐ and post‐2012 breakpoint. Similar to Figure 3, the detrended monthly mean total column CO from the TAO FTIR is plotted in Figure 4 to determine whether a change is detected from a ground‐based instrument in downtown Toronto. Here, the mean CO total column in August (and to a lesser degree in July and September) is seen to be larger for 2012–2019, however the year‐to‐year variability represented by the standard deviation shows significantly more overlap with the mean total column for 2004–2011. Yamanouchi et al. (2021) found a local decreasing trend in the TAO FTIR CO total column of −0.9% ± 0.07% per year from 2002 to 2019, which when used to detrend, results in a similar maximum between July and September. Furthermore, to demonstrate that summertime enhancements of CO are associated with fire activity, Figure A4 shows the MOPITT monthly mean XCO for Albert and Ontario versus the Moderate Resolution Imaging Spectroradiometer (MODIS) monthly burned area within Canada and the Continental United States from July‐September 2004‐2019.

Figure 3.

Detrended monthly mean XCO from MOPITT for (a) Ontario and (b) Alberta, for 2004–2011 (orange) and 2012–2019 (blue), with ±1σ indicated by the shading.

Figure 4.

Detrended monthly mean total column (TC) CO from the University of Toronto Atmospheric Observatory Fourier transform infrared spectrometer for 2004–2011 (orange) and 2012–2019 (blue), with ±1σ indicated by the shading.

3.2. DiD Results

Table 2 summarizes the DiD regression statistics for each disease in each province, using the equation and definitions provided in Table 1. The XCO β ₁ term represents the change in monthly hospitalizations per 100,000 associated with a 1 ppb increase in XCO, in the pre‐2012 period. The $i_{co}$ β ₃ term represents the change in the effect of a 1 ppb increase in XCO on monthly hospitalizations per 100,000 between pre‐2012 and post‐2012. The significance column is determined by the CI and p‐value: “no” means that both the CI and the p‐value suggest that the coefficient is insignificant, “possibly” indicates that the coefficient may be significant (CI contains the null but only minimally, and the p‐value <0.10), and “yes” signifies that the coefficient is significant based on both the CI and the p‐value. If the β term is shaded red, the relationship is negative; if it is green, the relationship is positive. When the $i_{co}$ β ₃ is marked with an asterisk (*), this indicates that β ₃ is positive and larger in magnitude than the absolute value of β ₁, which means that the relationship between XCO and hospitalizations has gone from negative in the pre‐2012 period to positive in the post‐2012 period, and that the interaction between XCO and period is significant.

Table 2.

Summary of DiD Regression Statistics: The β Term Corresponds to β ₁ for XCO and β ₃ for $i_{co}$, Shaded Red If It Is Negative and Green If It Is Positive, and Marked With * If β ₃ Is Positive and β ₃ > |β ₁|, and Significant; the 95% Confidence Interval and P‐Value Are Used to Determine Significance, Indicated With Yes (Bold Font), No, or Possibly

Disease	Province	Variable	β 1 for XCO, β 3 for iCO	95% confidence interval [0.025/0.975]	P‐value	Significant
Hypertension	ON	XCO	−1.72	−2.33/−1.10	<0.001	yes
		$i_{co}$	1.88*	1.18/2.57	<0.001	yes
	AB	XCO	−0.86	−3.50/1.79	0.52	no
		$i_{co}$	0.96	−1.66/3.58	0.47	no
Ischemic Heart Disease	ON	XCO	−0.41	−0.78/−0.03	0.04	yes
		$i_{co}$	0.50*	0.12/0.88	0.01	yes
	AB	XCO	0.42	−0.05/0.89	0.08	possibly
		$i_{co}$	−0.44	−0.91/0.03	0.06	possibly
Arrhythmia	ON	XCO	−0.13	−0.21/−0.04	0.003	yes
		$i_{co}$	0.12	0.03/0.22	0.009	yes
	AB	XCO	−0.52	−0.89/−0.1	0.006	yes
		$i_{co}$	0.48	0.12/0.85	0.009	yes
Heart Failure	ON	XCO	0.06	−0.02/0.15	0.12	no
		$i_{co}$	−0.09	−0.19/−0.001	0.05	yes
	AB	XCO	−0.20	−0.49/0.09	0.18	no
		$i_{co}$	0.22	−0.07/0.51	0.13	no
Cerebrovascular Disease	ON	XCO	−0.12	−0.28/−0.93	0.11	no
		$i_{co}$	0.11	−0.04/0.26	0.15	no
	AB	XCO	−0.15	−0.39/0.10	0.24	no
		$i_{co}$	0.14	−0.10/0.38	0.26	no
Pneumonia	ON	XCO	0.12	−0.13/0.36	0.34	no
		$i_{co}$	−0.22	−0.49/0.06	0.12	no
	AB	XCO	−0.48	−1.09/0.14	0.12	no
		$i_{co}$	0.48	−0.12/1.09	0.11	no
Acute Lower Respiratory Infections	ON	XCO	−0.01	−0.10/0.08	0.86	no
		$i_{co}$	−0.001	−0.10/0.10	0.98	no
	AB	XCO	−0.19	−0.59/0.21	0.34	no
		$i_{co}$	0.14	−0.26/0.54	0.47	no
Chronic Obstructive Pulmonary Disease	ON	XCO	−0.004	−0.20/0.19	0.97	no
		$i_{co}$	−0.02	−0.24/0.20	0.88	no
	AB	XCO	−0.14	−0.78/0.50	0.67	no
		$i_{co}$	0.16	−0.47/0.80	0.61	no
Asthma	ON	XCO	−0.14	−0.38/0.08	0.21	no
		$i_{co}$	0.31*	0.05/0.57	0.02	yes
	AB	XCO	0.37	−0.96/1.70	0.58	no
		$i_{co}$	−0.27	−1.58/1.04	0.68	no

Twelve out of the eighteen disease‐province pairs have a negative XCO β ₁ term and a positive $i_{co}$ β ₃ term, meaning that the negative effect of 1 ppb increase in XCO on hospitalizations is attenuated in the post‐2012 period, compared to the pre‐2012 period. Of those, five are considered significant (hypertension‐ON, ischemic heart disease‐ON, arrhythmia‐AB, arrhythmia‐ON, and asthma‐ON). In both Ontario and Alberta, XCO has a negative effect with arrhythmia hospitalizations in the pre‐2012 period that is greatly attenuated in the post‐2012 period, by a significantly positive interaction term. The interpretation of the $i_{co}$ β ₃ and XCO β ₁ terms is that the effect on monthly arrhythmia hospitalizations from a 1 ppb increase in XCO after 2012 is −0.002 (β ₁ + β ₃ = −0.127 + 0.125) admissions per 100,000 in ON and −0.036 (β ₁ + β ₃ = −0.521 + 0.485) admissions per 100,000 in AB. For hypertension in Ontario, the β ₃ coefficient is significant and greater than the absolute value of β ₁, which represents an inversion of the relationship between hospitalizations and XCO pre‐2012 (β ₁ = −1.716) to post‐2012 (β ₃ = 1.876), when a 1 ppb increase in XCO is associated with 0.16 (β ₁ + β ₃ = −1.716 + 1.876) additional monthly hospitalizations per 100,000 people.

To put those numbers into context, using the approximate population of Ontario of 12.9 million in 2009 and 14.6 million in 2019, a 1 ppb increase in XCO would correspond to 221 fewer hospitalizations in 2009 and 23 more hospitalizations in 2019. Cerebrovascular disease and hypertension in both provinces show an attenuated impact of XCO on hospitalizations, represented by a negative relationship with XCO (β ₁), and a smaller in magnitude, but positive $i_{co}$ coefficient (β ₃). In this case, both CIs for Alberta contain the null value, indicating that the association and interaction are not statistically significant. In contrast, the hypertension coefficients for Ontario are significant, with low p‐values and confidence intervals that do not contain null values. This may suggest that with a longer data set for the pre‐2012 period, the Alberta results could be significant, with the same applying to the shortened cerebrovascular disease time series for Ontario (see note in Section 2.2). For ischemic heart disease, heart failure, pneumonia, acute lower respiratory infections, Chronic Obstructive Pulmonary Disease (COPD) and asthma, the direction of the XCO β ₁ term and the $i_{co}$ β ₃ term are different for Ontario and Alberta, indicating that there is not consistent evidence to support a conclusion in the scenario. To evaluate the sensitivity of the findings, the DiD was also tested with the raw XCO values (not detrended); although the β values differed, the overall directionality and significance remained largely the same.

3.3. Discussion

The vast majority of epidemiological studies that investigate the impacts of wildfire smoke on health use base exposures to PM_2.5, through either ground‐based measurements of pollutants (Hahn et al., 2021), blended or fusion models that include ground‐based measurements coupled with remote sensing data, such as aerosol optical depth or indication of smoke in the atmospheric column (Gan et al., 2017, 2020; Magzamen et al., 2021), chemical transport models (Maji et al., 2024), or solely remote sensing data that approximates PM_2.5 in the atmospheric column (Wettstein et al., 2018). However, CO emissions are observed to increase during smoke periods (e.g., Buchholz et al., 2022) and ambient CO levels are associated with acute health outcomes in multiple settings. In Canada, ambient CO levels in Toronto were found to contribute 4.7% (95% CI = 3.4%–6.1%) of daily, non‐accidental deaths, while total suspended particles (TSP), which includes PM₁₀ and PM_2.5, contributed an additional 1.0% (95% CI = 0.2%–1.9%), based on changes in CO and TSP‐equivalent to their average concentrations (Burnett et al., 1998). In a study of the Medicare population in the United States, Bell et al. (2009) found that ambient CO levels below the US EPA's National Ambient Air Quality Standards were positively and significantly associated with cardiovascular disease (CVD) hospitalizations, including ischemic heart disease, arrhythmia, heart failure, cerebrovascular diseases, and total CVD hospitalizations; these results remained robust, though attenuated with the addition of NO₂ to the models (Bell et al., 2009). A study of daily ER visits from 2013 to 2019 in Lanzhou, China shows that an increase in atmospheric CO is associated with a rise in visits for CVD (I00–I99), ischemic heart disease (I20–I25), heart rhythm disturbances (I44–I49), heart failure (I50), and cerebrovascular diseases (I60–I69) (You et al., 2023). Although these studies were linked with ambient CO, no study directly linked to CO emanating from wildfire activity.

A systematic review of 36 publications on the long‐term health impacts of wildfire smoke exposure found that such studies are limited and largely focus on mental health (Gao et al., 2023). The majority of the studies assessed the cohorts for one to two years after exposure; nine of the 36 studies were based in Canada, where all but one was related to the 2016 Fort McMurray fires. The other Canadian report was a 20‐year population‐based cohort study that followed more than 2 million adults to examine the association between wildfire exposure and specific cancers related to wildfire emissions (Gao et al., 2023; Korsiak et al., 2022). Comparing health outcomes of populations based on wildfire exposure area, defined by proximity to burned area, suggests that an increased risk of lung cancer and brain tumors is associated with long‐term exposure to wildfires. Gao et al. (2024) applied a DiD analysis to cardiovascular mortality in relation to wildfire‐related PM_2.5 in Brazil, between 2010 and 2018, where an observed increase in wildfires occurred in the recent years. When controlling for seasonal temperature and socio‐economic status, they reported that wildfire‐related PM_2.5 was statistically significant in causing mortality with a 1‐year lag for all‐cause CVD (I00–I99, G45, G46), ischemic heart disease (I20–I25) and stroke (I60–I64, G45, I690–I694), but not for heart failure (I50), hypertension (I10–I13) or peripheral vascular disease (I70–I89). Use of a gas‐phase marker of smoke like CO may obviate challenges in PM_2.5 as a relevant smoke exposure as long‐term satellite measurements are available in remote areas, and source apportionment (e.g., background vs. smoke PM_2.5) is not required.

Unlike most wildfire smoke studies, this study found that most respiratory hospitalizations were not significantly associated with column CO in either the pre‐2012 or post‐2012 periods. However, pneumonia, COPD, and acute lower respiratory infections were all higher in Alberta in the post‐2012 period. Given the relatively smaller time series available for Alberta, as well as the relatively lower incidence of respiratory disease in the warm‐season months evaluated in this study, we interpret these findings to be suggestive of a relationship between increased column CO and these respiratory outcomes. Acute asthma events, the health outcome most consistently associated with wildfire smoke exposure, had a significant increase in the post‐2012 period in Ontario, suggesting that there was a meaningful change in slope of the relation between CO and asthma. Hahn et al. (2021) report that the cardiovascular outcomes showed less significance compared to the respiratory diseases examined with the same population and wildfire events. When examining the CVD ER admissions within the population as a whole, Hahn et al. (2021) found that the risk went down on the day of exposure and showed an increase with a 2–3 days lag, while the odds of respiratory peaked on the first day and decreased thereafter, with the strongest effects observed for asthma. Increased morbidity from respiratory diseases due to wildfire smoke exposure is a common finding in studies, whereas correlations between CVD and health care utilization and wildfire smoke are less frequently reported (Henderson et al., 2011). This may be somewhat attributed to differences in exposure assessment and health data usage (Hahn et al., 2021), which would be particularly relevant to the methods applied in this work as exposure and health metrics are aggregated on a monthly basis, while other studies have daily inputs.

The biological plausibility of the relationship between column CO and CVD may contribute to our current findings that are predominated by positive and significant relationships between CO and CVD outcomes in the post‐2012 period compared to the pre‐2012 period. Wettstein et al. (2018) found that smoke from 2015 wildfires in California were associated with a number of cardiovascular hospitalization outcomes, including myocardial infarction, ischemic heart disease, heart failure, dysrhythmia, pulmonary embolism, stroke, and transient ischemic attack (Wettstein et al., 2018). Exposure assessment was conducted using NOAA Hazard Mapping System (HMS) smoke polygons, classified as light, medium and heavy. The HMS polygons indicate if there is smoke in the atmospheric column, but does not indicate if the smoke is at the ground level. In a study spanning six wildfire seasons in Colorado, Magzamen et al. (2021) blended HMS polygon data with ground‐based US EPA monitoring data for PM_2.5 to indicate if smoke in the atmospheric column was present at ground level (Magzamen et al., 2021). This study indicated that all CVD hospitalizations (as well as hospitalizations for cerebrovascular disease, heart failure, and ischemic heart disease) in single‐lag model were significantly associated with a 10 μg/m³ increase in smoke PM_2.5; in distributed lag models, smoke PM_2.5 was significantly associated with ischemic heart disease, similar to findings in the current study.

Further, the Colorado analysis demonstrated that 10 μg/m³ increase in wildfire smoke was significantly associated with cardiac arrest mortality (Magzamen et al., 2021). A study in California indicated that exposure to wildfire smoke was significantly associated with out‐of‐hospital cardiovascular deaths over three seasons (Jones et al., 2020). An evaluation of the 2015 wildfire season in Calgary, Alberta, found that smoke exposure was positively and significantly associated with outpatient physician visits for congestive heart failure and ischemic heart disease (Mahsin et al., 2022), similar to findings in our study.

The scope of this analysis is limited by data availability and the aggregation thresholds of the CIHI, in frequency and duration. Although pre‐2012 data were limited for Alberta, we considered distance from fire to be an important metric in understanding potential health effects of wildfire proximity based on prior research (Magzamen et al., 2021). Alberta is in closer proximity to seasonal forest fires than Ontario, meaning the population has greater exposure to wildfire smoke and thus a greater potential for hospitalization records to reflect the change in wildfire frequency. However, Alberta hospitalization records were only available from 2010 onwards, which only provides two years in the “before” period and can potentially constrain the DiD signal. Other provinces were not included as they did not have data available for the appropriate duration. If the data were available on a daily basis, this would allow for an analysis more similar to those in Hahn et al. (2021) and Magzamen et al. (2021). Another consideration for the significance of trends is the sample size, the entire population of Canada is comparable to the state of California (PPIC, 2024; STATS CAN, 2024). Further, utilization of health care during fire events care may vary based on several confounding factors, such as smoke density and toxicity (can depend on proximity and severity of burning fire, temperature, ecoregion, etc.), human behavior (e.g., availability of medical treatment, ability to self‐treat), and local intervention measures (e.g., evacuation orders, distribution of protective equipment) (Magzamen et al., 2021).

The DiD uses the monthly average total column XCO as measured by MOPITT; as stated in Section 2.1.1, the measurement is most sensitive to the middle troposphere. The change in seasonal cycle is observed in both the MOPITT data and the CO measurements from the TAO FTIR (as shown in Figures 2 and 3 and 4), which is sensitive to the lower troposphere, and it should be noted that enhancements in the surface level CO that more directly influence air quality, will vary from the column measurements. However, enhancements in the CO tropospheric column measured at TAO have been linked to episodes of degraded air quality as a result of long‐range transport of wildfire plumes (Flood et al., 2025). The overlap observed in the standard deviation of the monthly mean CO presented in Figures 3 and 4 may indicate that a longer time series is needed to better resolve the CO increase in Alberta and Ontario, which may also reflect the varying results of Table 2. Nonetheless, the prominence of a positive uptick in outcomes between the pre‐2012 and post‐2012 periods is suggestive of a shift in the relationship between health outcomes and a more intense wildfire season, as marked by enhanced atmospheric CO columns. The wide confidence intervals can be somewhat attributed to the limitations as stated, and underscore need for further research. Overall, the absence of daily health care encounter data limited the ability to conduct a traditional air pollution epidemiology study, such as a time‐series or case‐crossover analysis. Instead, this work serves as a preliminary investigation to highlight the potential value of CO as a marker for evaluating health impacts related to air quality, particularly from large‐scale impacts of wildfire smoke events. This provides future research opportunities to use CO as an indicator for smoke exposure, locally or after long‐range transport.

4. Conclusions

This study investigated a change in the seasonal cycle of atmospheric CO from satellite measurements in the provinces of Alberta and Ontario, in addition to ground‐based monitoring in Toronto, Ontario, partitioning measurements from before and after 1 January 2012. Monthly hospital ER admissions for nine cardiovascular and respiratory diseases were obtained for the health districts in Alberta (2010–2019) and Ontario (2004–2019) and analyzed with respect to the satellite‐derived monthly total column CO using a difference‐in‐difference approach. The objective was to assess whether the hospitalization records follow a similar pattern to that observed with the atmospheric CO and to evaluate whether CO can be used as a an appropriate metric for exposure.

A comparison of MOPITT XCO measurements across Canada shows areas of higher mean August values for 2012–2019 compared to 2004–2011, even without removing the overall North American downward trend of −0.57% CO per year. When assessing the detrended monthly mean XCO for Alberta and Ontario specifically, a positive shift is seen for the period from July to September, with a maximum in August. The TAO FTIR monthly mean CO total columns show a similar increase over the same time period. The standard deviations of the pre‐2012 and post‐2012 monthly means overlap due to year‐to‐year variability. Overall, the findings are consistent with the results in Buchholz et al. (2022), which showed an increase in seasonal August CO observed by MOPITT in large areas of the United States when comparing pre‐2012 and post‐2012, attributed to an increase of wildfires in the Pacific Northwest.

A DiD regression was applied to monthly hospital ER admission counts for nine cardiovascular and respiratory diagnoses to assess if the observed increase in column CO due to wildfire emissions is reflected in health diagnostics. Despite the sometimes‐wide confidence intervals, several diseases show a significant change in the DiD regression when considering the interaction term between exposure and outcome. In the instances when the model shows a positive $i_{co}$ β ₃ interaction term, an increasingly adverse effect is observed during the more intense wildfire years (post‐2012). The positive $i_{co}$ β ₃ coefficients indicate how many more hospitalizations per 100,000 are associated with a 1 ppb increase in XCO after 2012 compared to before 2012 for the same month. This positive interaction is seen for hypertension (AB and ON), ischemic heart disease (ON), arrhythmia (AB and ON), heart failure (AB), cerebrovascular disease (AB and ON), pneumonia (AB), acute lower respiratory infection (AB), COPD (AB) and asthma (ON). However, the interaction was only found to be statistically significant for hypertension (ON), ischemic heart disease (ON), arrhythmia (AB and ON), and asthma (ON). Overall, these findings demonstrate the potential of satellite‐derived CO as a metric for exposure in wildfire health analyses.

As climate change impacts the occurrence of wildfires, it is important to consider the effects of the increased emissions on human health, both locally and through long‐range transport. Understanding the implications for the health care system can help build a framework for shaping policy, informed decision making, and effective risk mitigation. Establishing a connection between adverse health outcomes and satellite measurements of CO offers the potential to enhance public health advisories, leveraging the broader spatial coverage provided by satellites compared to regional ground‐based monitoring methods. While outside the scope of this work, ground‐based in situ CO measurements could be explored to further evaluate the use of satellite‐derived CO for use in health effects studies. Overall, studies examining the long‐term and long‐range health impacts of wildfires are limited, underscoring the importance of further research to expand upon the work done in this study. Further research, including broader geographical coverage, and longer‐term finer‐scale patient data will provide greater insight into the long‐term impacts of long‐range wildfire emissions on human health.

Conflict of Interest

The authors declare no conflicts of interest relevant to this study.

Appendix A.

Table A1.

Name and ICD‐10 Codes for the Diseases Examined

Disease	ICD‐10 codes
Hypertension	I10–I13, I15
Ischemic heart disease	I20–I25
Arrhythmia	I46–I49
Heart failure	I50
Cerebrovascular disease	I61–I69, G45
Pneumonia	J12–‐J18
Acute lower respiratory infections	J20–J22
Chronic obstructive pulmonary disease (COPD)	J40–J44, J47
Asthma	J45

Figure A1.

Map of Alberta provincial health districts from which emergency room admission data were obtained (STATS CAN, 2018).

Figure A2.

Map of Ontario provincial health districts from which emergency room admission data were obtained. The upper left panel shows the districts in southern and central Ontario and the lower right panel shows the districts in northern Ontario (STATS CAN, 2018).

Figure A3.

Detrended monthly mean XCO from Measurements of Pollution in the Troposphere for (a) Ontario and (b) Alberta, for 2004–2011 (orange) and 2012–2019 (blue), showing each year individually.

Figure A4.

Monthly mean MOPITT XCO (detrended, ppb) versus monthly summed burned area from MODIS for the Continental USA and Canada, for July‐September from 2004 to 2019.

Flood, V. A. , Strong, K. , Buchholz, R. R. , Kuiper, G. , & Magzamen, S. (2025). Assessing the impact of wildfire emissions on the seasonal cycle of CO and emergency room visits in Alberta and Ontario, Canada. GeoHealth, 9, e2024GH001317. 10.1029/2024GH001317

Data Availability Statement

The TAO FTIR data are publicly available from Yamanouchi et al. (2022) and from the NDACC Data Host Facility (NDACC, 2024).

NASA provides open‐source data for the MOPITT instrument (NASA/LARC/SD/ASDC, 2024).

The hospital admission data used in this work are protected by an Aggregated Data Single Use License with the Canadian Institute for Health Information. These data may be requested by completing a data inquiry form for the use of data from the National Ambulatory Care Reporting System on the Canadian Institute for Health Information website (CIHI, 2024).