The Covid-19 hospitalization risk associated with air pollution in New York state counties after the 2023 Quebec wildfires

Abstract

Background:

Air pollution from the 2023 Quebec wildfires affected New York state (NY) with daily average PM_2.5 levels that peak on June 7. Increased Covid-19 hospitalizations were recorded weeks after the wildfires. This study analyses the trend of Covid-19 hospitalization in NY counties after the 2023 Quebec wildfires and estimates their association with higher PM_2.5 concentration levels, compared to 2022.

Design and methods:

A Bayesian spatiotemporal regression model was used to estimate the impact of wildfire smoke on Covid-19 hospitalizations. Four periods of pre/post-wildfire and 7-day post-wildfire daily hospitalization periods were considered to compare the association of daily average PM_2.5 levels, from May 1 to June 7, with daily Covid-19 hospitalization rates in NY counties in 2022 and 2023. The pre/post-wildfire and 7-day post-wildfire periods considered a lag of 2, 4, 6, and 8 weeks and 24, 48, 72, and 96 h, respectively. The model was adjusted for sociodemographic factors.

Results:

The Covid-19 hospitalization rate followed an increasing trend in the second, third and fourth pre/post-wildfire periods in 2023 in contrast with 2022 when no trends were identified. Each PM_2.5 unit increase was associated with a 2%; 6% and 7% Covid-19 higher hospitalization risk in periods 2, 3, and 4, respectively, in 2023 only. These findings identify a potential impact of wildfire smoke on the severity of Covid-19 morbidity after 2 weeks of the wildfires. Robust spatiotemporal analyses can be used to identify specific at-risk areas and communities to support public health decision-making and health strategies.

Conclusions:

This study identifies a higher risk of Covid-19 hospitalization in New York State associated with higher air pollution levels from the 2023 Quebec wildfires, in the first week and 2, 3, and 4 weeks after the wildfires. These findings concur with the increasingly investigated association of air pollution with severe Covid-19. The methodological approach of this study shows the utility of spatiotemporal epidemiological analyses and need for future research on wildfire smoke as a potential determinant of severe Covid-19. With more frequent and extreme climate events it is paramount to improve our understanding of many potential health impacts of wildfires to prepare strategies to deal with, and potentially anticipate, environmental health and healthcare responses in wildfire-prone regions.

Keywords

Covid-19 wildfires bushfires Bayesian spatiotemporal regression particulate matter New York counties environmental health hospitalization rate

Introduction

The wildfires that broke out on June 1, 2023 in Quebec, Canada burned 20 million acres of land with New York (NY) and other states in the United States recording high to very high levels of air pollution from the wildfire smoke in the following days and weeks. On June 3, the wildfire smoke reached the city of Buffalo in the northwest of NY state (NY) with half of state (west and north) covered in smoke by June 5 (NASA image). Between 6-7 June, the whole of NY was affected (NASA image) with a daily average peak of PM_2.5 74.75 µg/m³, reached on June 7. In response, the NY health services prepared for the potential increase of asthma and other cardiovascular and respiratory conditions, the most common morbidities after severe wildfires.¹ However, a high number of Covid-19 hospitalizations was also seen in the weeks following the Quebec wildfires which increased the demand on the health system (Figure 1). According to the Centre for Disease Control and Prevention data, daily Covid-19 hospitalizations increased up to 22% in July (post the wildfires) compared with May (before the wildfires), with a persistent increase of hospitalizations in the last 3 weeks of July.² Although higher Covid-19 morbidity following severe air pollution from wildfires has been previously studied, this is an atypical rise in severe Covid-19 morbidity in an area where, despite a high vaccination rate and a 6-month trend of decreasing cases, severe air pollution from recent wildfires could be a contributing factor.

Figure 1.

Smoothed trend of the mean of the Covid-19 hospitalization across all NY counties in 2022 and 2023, and daily PM_2.5 average levels and 95% Confidence intervals. The four pre/post and 7-day post-wildfire periods are highlighted in blue, green, purple, and pink, respectively.

Studies conducted during the first 18 months of the Covid-19 pandemic found that SARS-CoV-2 can survive in airborne particulate matter and potentially enhance its persistence in the atmosphere.³ This concurs with the increased risk of respiratory infections associated with higher levels of air pollutants such as particulate matter, nitrogen dioxide, carbon dioxide, and carbonaceous particles common in wildfires smoke.^4–6 A positive association of air pollution with higher risk of Covid-19 transmission has been previously identified in the United States^7,8 which coincides with estimates of an increased risk of Covid-19 associated with air pollution from wildfire smoke.⁹ Exploratory analyses of spatially aggregated data can provide important insights of the disease trend and specific areas at risk to support the health sector response.^10,11 Spatial epidemiological analyses can be useful to identify specific areas or communities with higher risk and to guide public health decision making regarding the higher risk of Covid-19 associated with air pollution in other regions affected by wildfires.^12,13

This study aims to analyze the trend of Covid-19 hospitalizations in the NY counties in the weeks following the 2023 Quebec wildfires and explore its statistical association with the higher levels of wildfire smoke, measured as PM_2.5 concentration levels. The analysis uses a Bayesian spatial hierarchical model that considers the spatiotemporal distribution of the Covid-19 hospitalization rate and the PM_2.5 levels in the 62 NY counties, comparatively between 2022 and 2023. The objectives are to map the risk of Covid-19 hospitalizations in NY counties in 2022 and 2023 and to estimate and compare the Covid-19 hospitalization risk associated with PM_2.5 in both years, after adjusting for sociodemographic confounders.

Methods

A preliminary analysis of PM_2.5 in NY identified an increasing concentration of daily levels from June 1, that peaked on June 7, 2023. To assess the 7 days of the wildfires and include an adequate pre-wildfire period for comparison according to best-practice,¹⁴ a 38-day pre/post wildfire (exposure) period was set to include daily PM_2.5 levels from 1 month before the start of the wildfires (May 1) until the peak PM_2.5 concentration (June 7). To estimate the association of PM_2.5 levels with Covid-19 hospitalizations, daily Covid-19 hospitalization rates were calculated in four 38-day lag periods starting 2, 4, 6, and 8 weeks after May 1, to compare the trend of Covid-19 hospitalization risk in different pre/post wildfire periods. The choice of an initial 2-week lag for the pre/post wildfire periods considers the Covid-19 incubation time that varies from 9 to up to 14 days,^15,16 and the potential risk of Covid-19 infection associated with higher levels of air pollution that can be related to changes in the immune response that could develop weeks after the exposure.¹⁷ The subsequent lags (4, 6, and 8 weeks) were set considering that air pollution effects should assess multiple lag periods¹⁸ and the need to compare the lag structure of the PM_2.5-Covid-19 associations to avoid over reliance on non-comparative U.S. Environmental Protection Agency analyses.¹⁹ An additional analysis was set to assess the acute effects of the wildfires on Covid-19 hospitalizations, to estimate the association of daily PM_2.5 levels during the 7 days of the wildfires with the Covid-19 hospitalization rate in four 7-day post wildfire periods (24, 48, 72, and 96 h after June 1).

Ethics approval was not required by the Queensland University of Technology ethics committee on the basis that all data are publicly available and exclude individual identifiers.

Data

Daily hospitalizations from May 15 to August 8 in 2022 and 2023, were obtained per county from the NY health department.²⁰ The daily hospitalization rate of Covid-19 per county was calculated using the counts of total population from the USA census, as the reference population. Air pollution data obtained from the USA Environmental Protection Agency were used to calculate daily averages of PM_2.5 levels for all NY monitoring stations and stations near the state boundaries from May 1 to August 8. An interpolation model implemented in R was used to estimate the daily PM_2.5 level in a 1 km grid over the whole of NY. Zonal statistics were used to calculate the average PM_2.5 level per county.

The following sociodemographic and environmental factors from the County Health Rankings (CHR)²¹ were included to adjust the analysis for potential confounders between state county quartiles: quality of life; health behavior; health outcomes; clinical care; socioeconomic factors and environmental factors. The CHR are measures based on vital statistics data, transmitted disease rates, and Behavioral Risk Factor Surveillance System survey data, calculated by staff at the National Center for Health Statistics.²² The number of people staying at home per county was obtained from the daily report on trips by distance from the Bureau of Transportation and Statistics. In addition, since higher population density has been associated with increased Covid-19 transmission in the United States counties,²³ the proportion of population per area (km²) in each county was calculated and included in the regression models.

Statistical analysis

The analysis used a Bayesian hierarchical regression model that considers the spatial dependency of the 62 NY counties across each pre/post wildfire period (38-day periods) and the 7-day post wildfire periods. For the $i^{t h}$ county in the $t^{t h}$ day, the Covid-19 hospitalizations ( $y$ ) were modeled following a Poisson distribution with the linear predictor defined on the logarithmic scale:

η_{i t} = \log (y_{i t}) = α + β_{c p} P M 2 . 5_{i} + β_{x} X_{x i} + T . p o p_{i t} + υ_{i} + ν_{i} + γ_{t} + ϕ_{t}

where α is the intercept, $P M 2.5$ is the average PM_2.5 concentration, $X$ represents the vector of sociodemographic covariates (scaled) with the respective regression coefficients $β_{x}$ , $T . p o p$ is the total population included as the offset, and the parameters $υ$ and $ν$ represent the spatial structured and the unstructured component according to the Besag-York-Mollie (BYM) specification.²⁴ The time parameters were defined as introduced by Knorr-Held²⁵ where $γ_{t}$ is the temporally structured effect, modeled using a random walk through the counties structure, and $ϕ_{t}$ is the unstructured temporal effect specified with a Gaussian exchangeable prior.

The choice of a Bayesian hierarchical model was based on the need to incorporate the spatial dependency between counties given the geographical level of the health data and the spatial nature of the exposure. Bayesian models produce robust and reliable estimates on spatially aggregated count data and are a strong statistical alternative in epidemiological analysis of infectious diseases, especially Covid-19.^26,27 In addition, the choice of a mixed regression model incorporating a spatial structure helps to control the potential ecological bias of analyses of data with a spatial level of aggregation²⁸ such as the NY counties Covid-19 hospitalizations.

To assess the effect of the prior on the spatial effects parameters, a preliminary sensitivity analysis compared the model with four vaguely informative priors to choose the best fit doing a comparison of the Deviance Information Criterion (DIC) value.²⁹ All priors had a similar impact on the results and the statistical credibility of the regression coefficients. Strong statistical associations were identified when the regression coefficient’s 95% credible intervals (95% CI) did not cross the null value of 1. The models were fit in R with the R-INLA package.³⁰ All maps were drawn in ArcGis-Pro (v.3.1) with the Mann Kendall trend test used to calculate and map the emerging spatiotemporal trends.³¹

Results

There were 11,919; 12,066; 13,148 and 14,814 Covid-19 hospitalizations in the whole of NY state for the four periods: May 15–June 21; May 30–July 06; June 14–July 21 and June 29–Aug 05, in 2022. The number of hospitalizations for the same periods in 2023 were 3104; 2886; 2562 and 3071, respectively. Table 1 shows the summary statistics for the Covid-19 rate per county and period for each year.

Table 1.

Covid-19 hospitalization rate per 1000 people per county across four pre/post wildfire and 7-day post-wildfire periods in 2022 and 2023.

Period	Min.	First Qu.	Median	Mean	Third Qu.	Max.
2022
Pre/post wildfire period 1 \| 15 May–21 June	0.00	0.015	0.017	0.017	0.021	0.031
Pre/post wildfire period 2 \| 30 May–06 July	0.00	0.009	0.014	0.013	0.017	0.022
Pre/post wildfire period 3 \| 14 June–21 July	0.00	0.010	0.013	0.013	0.017	0.027
Pre/post wildfire period 4 \| 29 June–05 Aug	0.006	0.011	0.014	0.015	0.019	0.032
7-day post wildfire period 1 \| 02 June–08 June	0.00	0.012	0.014	0.015	0.018	0.042
7-day post wildfire period 1 \| 09 June–15 June	0.00	0.010	0.014	0.014	0.018	0.042
7-day post wildfire period 1 \| 16 June–22 June	0.00	0.010	0.014	0.014	0.017	0.035
7-day post wildfire period 1 \| 23 June–29 June	0.00	0.009	0.014	0.013	0.016	0.030
2023
Pre/post wildfire period 1 \| 15/May−21/June	0.001	0.002	0.004	0.004	0.005	0.010
Pre/post wildfire period 2 \| 30/May−06/July	0.001	0.002	0.003	0.003	0.004	0.015
Pre/post wildfire period 3 \| 14/June−21/July	0.00	0.002	0.003	0.003	0.004	0.010
Pre/post wildfire period 4 \| 29/June − 05/Aug	0.00	0.002	0.003	0.003	0.005	0.008
7-day post wildfire period 1 \| 02 June–08 June	0.00	0.00	0.003	0.004	0.005	0.028
7-day post wildfire period 1 \| 09 June–15 June	0.00	0.001	0.003	0.004	0.005	0.028
7-day post wildfire period 1 \| 16 June–22 June	0.00	0.002	0.003	0.004	0.005	0.028
7-day post wildfire period 1 \| 23 June–29 June	0.00	0.001	0.003	0.004	0.005	0.028

Figure 1 shows the trend of daily PM2.5 average levels between May 1 and June 7, and the Covid-19 hospitalization rate across each of the pre/post wildfire periods and the 7-day post-wildfire periods in 2022 and 2023. The Covid-19 hospitalization rate was higher in all periods in 2022 compared with 2023. In 2022 the Covid-19 hospitalization rate consistently decreased during the pre/post wildfire period 1 while it consistently increased in pre/post wildfire period 3. In contrast, it did not exhibit a clear trend in pre/post wildfire periods 2 and 4. In 2023, there was an increasing trend of the Covid-19 hospitalization rate in the third and especially fourth pre/post wildfire periods, with no defined trend in pre/post wildfire periods 1 and 2. There was not a clear trend of the Covid-19 hospitalization rate between June 2 and 11, the days included in the 7-day post wildfire periods. There was only 1 day with no Covid-19 hospitalizations (July 5, 2023). Whereas there was no consistent trend in the PM_2.5 level in 2022 and 2023, the average daily concentration in 2022 was below the EPA standard 10 µg/m³,³² while the PM_2.5 levels since June 1, 2023 were higher compared to 2022, with daily average of more than 10 µg/m³.

Table 2 shows the exponentiated posterior mean of the fixed effect of PM_2.5 on the Covid-19 hospitalization rate once other sociodemographic confounders were taken into account, in the 4 pre/post wildfire and post wildfire periods in 2022 and 2023. There was a negative association of PM_2.5 with Covid-19 hospitalizations in all four periods in 2022 while higher PM_2.5 levels were positively associated with Covid-19 hospitalizations in periods 2, 3 and 4 in 2023. There was a slight but increasing association of PM_2.5 with the Covid-19 hospitalization rate from the second to fourth pre/post wildfire periods in 2023. Since the exponentiated regression coefficients can be interpreted as relative risk, the models indicate that each PM_2.5 unit increased the risk of Covid-19 hospitalizations by 2%; 6% and 7% in NY in the pre/post wildfire periods 2, 3, and 4 in 2023, respectively. For the 7-day post wildfire periods, there was a positive association between PM_2.5 and Covid-19 hospitalization in the first two periods in 2022 while this association was found in periods 1, 3, and 4 in 2023. The relative risk of Covid-19 hospitalization varied from −2% to 2% in 2022, and −5% to 11% in 2023, without a clear positive or negative trend in any of the 2 years.

Table 2.

Exponentiated posterior mean with 95% credible intervals (CI) of the spatial regression model in the pre/post wildfire and 7-day post wildfire periods in 2022 and 2023.

	Pre/post 38-day wildfire periods				7-day post wildfire periods
Variable	Period 1	Period 2	Period 3	Period 4	Period 1	Period 2	Period 3	Period 4
2022
PM2.5^a	0.95 (0.94–0.95)	0.99 (0.98–0.99)	0.9 (0.9–0.91)	0.96 (0.96–0.96)	1.02 (1.01–1.02)	1.01 (1.003–1.01)	0.98 (0.98–0.99)	0.99 (0.99–0.99)
People staying home	1.39 (1.38–1.4)	1.23 (1.22–1.24)	1.06 (1.06–1.07)	0.87 (0.87–0.88)	0.92 (0.91–0.94)	0.74 (0.73–0.75)	1.21 (1.2–1.23)	1.36 (1.34–1.38)
Quality of life	0.98 (0.65–1.47)	0.95 (0.65–1.4)	1.002 (0.67–1.49)	0.9995 (0.86–1.16)	0.97 (0.83–1.13)	0.93 (0.78–1.11)	0.81 (0.67–0.98)	0.76 (0.62–0.94)
Health behaviors	0.96 (0.63–1.49)	0.83 (0.55–1.24)	0.83 (0.54–1.26)	0.95 (0.8–1.13)	0.82 (0.7–0.97)	0.93 (0.78–1.12)	1.05 (0.86–1.28)	1.05 (0.84–1.3)
Clinical care	0.98 (0.72–1.35)	1.05 (0.78–1.41)	1.1 (0.81–1.49)	1.09 (0.97–1.22)	1.1 (0.99–1.23)	1.17 (1.03–1.34)	1.05 (0.91–1.21)	1.04 (0.89–1.21)
Socioeconomic factors	0.69 (0.5–0.95)	0.79 (0.58–1.06)	0.87 (0.63–1.19)	1.01 (0.89–1.15)	1.001 (0.89–1.12)	1.14 (0.99–1.3)	1.01 (0.87–1.18)	0.99 (0.84–1.17)
Environmental factors	1.08 (0.84–1.38)	1.1 (0.87–1.39)	1.13 (0.88–1.44)	0.92 (0.83–1.02)	1.02 (0.93–1.12)	0.98 (0.89–1.09)	0.94 (0.84–1.06)	0.91 (0.8–1.03)
Population density	0.99 (0.68–1.46)	1.04 (0.73–1.48)	1.11 (0.77–1.62)	1.04 (0.9–1.21)	0.77 (0.67–0.88)	0.92 (0.78–1.08)	0.76 (0.64–0.91)	0.72 (0.59–0.87)
Health outcomes	1.6 (1.05–2.44)	1.5 (1.02–2.23)	1.3 (0.87–1.97)	1.05 (0.89–1.22)	1.16 (0.99–1.36)	0.94 (0.79–1.12)	1.16 (0.96–1.4)	1.24 (1.01–1.53)
2023
PM2.5^a	0.93 (0.93–0.94)	1.02 (1.01–1.02)	1.06 (1.05–1.07)	1.07 (1.06–1.07)	1.08 (1.06–1.09)	0.95 (0.94–0.96)	1.09 (1.08–1.11)	1.11 (1.1–1.13)
People staying home	0.97 (0.95–0.99)	2.04 (2–2.07)	0.88 (0.86–0.89)	2 (1.96–2.03)	0.54 (0.49–0.6)	0.38 (0.34–0.43)	1.55 (1.39–1.74)	4.3 (3.86–4.8)
Quality of life	1.04 (0.8–1.35)	1.04 (0.75–1.44)	1.5 (0.95–2.39)	2.02 (0.996–4.13)	1.29 (0.89–1.86)	1.39 (0.98–1.97)	1.13 (0.72–1.77)	1.13 (0.47–2.73)
Health behaviors	1.23 (0.96–1.57)	1.16 (0.84–1.62)	0.86 (0.55–1.35)	0.65 (0.33–1.28)	0.98 (0.71–1.36)	0.8 (0.55–1.15)	1.11 (0.75–1.63)	1.34 (0.64–2.85)
Clinical care	1.05 (0.87–1.26)	1.03 (0.81–1.3)	1.32 (0.95–1.84)	0.96 (0.58–1.6)	1.38 (1.09–1.74)	1.49 (1.18–1.89)	1.15 (0.85–1.54)	0.96 (0.54–1.71)
Socioeconomic factors	1.01 (0.82–1.25)	0.78 (0.58–1.05)	0.75 (0.5–1.1)	0.47 (0.27–0.82)	1.19 (0.9–1.59)	1.05 (0.74–1.5)	0.65 (0.46–0.93)	0.35 (0.18–0.69)
Environmental factors	0.96 (0.81–1.13)	0.96 (0.77–1.2)	0.82 (0.6–1.12)	0.95 (0.6–1.5)	0.94 (0.75–1.18)	0.86 (0.67–1.09)	0.93 (0.71–1.21)	0.92 (0.55–1.54)
Population density	1.19 (0.95–1.49)	0.91 (0.67–1.23)	1.25 (0.84–1.88)	0.8 (0.43–1.47)	0.79 (0.59–1.06)	0.67 (0.5–0.91)	0.44 (0.31–0.61)	0.26 (0.14–0.5)
Health outcomes	0.86 (0.67–1.11)	1.19 (0.85–1.68)	0.92 (0.58–1.46)	1.46 (0.75–2.84)	0.6 (0.42–0.84)	0.75 (0.53–1.07)	1.12 (0.75–1.68)	1.83 (0.84–3.96)

Particulate matter 2.5.

Of the sociodemographic factors, only the number of people staying at home during the bushfire period had a strong effect on the Covid-19 rate, consistently across both the pre/post and 7-day post wildfire periods (Table 2). There were less consistent associations with the other sociodemographic covariates. The socioeconomic factors were negatively associated with the Covid-19 hospitalization rate in pre/post wildfire period 1 of 2022 and pre/post wildfire period 4 in 2023 only while the county health outcomes rank had a positive association with Covid-19 hospitalizations in the pre/post wildfire periods 1 and 2 in 2022 only. Quality of life, health behaviors, clinical care, population density, and health outcomes had strong associations in some but not all pre/post and 7-day post wildfire periods in 2022 and 2023.

Figure 2(a) shows the trend of the exponentiated posterior mean of the temporal effect in the pre/post wildfire periods, once the daily variations and spatial distribution of the NY counties were considered in the model, and after adjustment for sociodemographic confounders. The exponentiated daily posterior mean can be interpreted as the relative risk of Covid-19 hospitalization (relative to the whole period). In 2022, the relative risk of Covid-19 decreased in the first period, with an increasing trend in period 3 and an undefined trend in the second and fourth periods. In 2023, there was not a clear trend in the first two periods while there was an increasing trend in the second half of period 3 and the last 3 weeks of period 4. The sharp decreases in the same day in periods 2, 3, and 4 correspond to July 5, when there were no Covid-19 hospitalizations. Most days in periods 2, 3, and 4 had a higher risk (>1) compared with the same periods in 2022. Figure 2(b) shows the distribution of the county-specific spatial effect (cylinders) that once exponentiated can be interpreted as the relative risk of Covid-19 hospitalization (relative to all NY counties), which ranged from 0.004 to 2.93.8 without a defined clustering pattern in any of the four pre/post wildfire periods in 2022. There was a significant positive risk trend across the 4 periods clustered in counties in the southeast and clusters with a negative trend in west and central NY in 2022. The county specific Covid-19 hospitalization risk in 2023 ranged from 0.003 to 7.28 with some clustering in the southeast and west regions in the third and fourth periods. Most of the counties with the highest risk coincided with daily average PM2.5 level >10 µg/m³. The emerging risk trend across the four pre/post wildfire periods in 2023 shows counties with a positive trend clustered in west and central NY which coincide with most counties with average PM2.5 level >10 µg/m³, and counties with a significant negative trend clustered in central NY, also in counties with PM2.5 level >10 µg/m³.

Figure 2.

(a) Smoothed exponentiated posterior mean of the temporal effect. (b) Spatial distribution of the exponentiated county-specific posterior mean (cylinders) and the Man-Kendall statistics (emerging trend) in each county across the four pre/post wildfire periods in 2022 and 2023.

The exponentiated posterior mean of the temporal effect in the 7-day post wildfire periods is shown in Figure 3(a). All periods in 2022 show a similar trend with a decreased risk in the first 1–2 days followed by an increase in the next 2–3 days. There is a decrease in the risk in the last 2–3 days in periods 3 and 4. In 2023, there are dissimilar trends between the four periods, with decreases in the first 1–2 days in periods 1 and 2 and increases in periods 3 and 4, and varying trends in the last 3–4 days in each of the four periods. The distribution of the posterior mean of the spatial effect in each 7-day post wildfire period is shown in Figure 3(b). The county specific Covid-19 risk in 2022 (cylinders) ranged from 0.3 to 2.4 with some clusters of lower risk in the west, central and southeast regions in periods 3 and 4. A significant increasing risk trend was found in counties clustered in central and south NY while countries with a decreasing risk trend clustered in the central, north and southeast regions. In 2023, the county specific Covid-19 risk ranged from 0.03 to 21.9 with some clusters of lower risk in west and south NY in periods 3 and 4. There were multiple counties with a significant increasing risk trend clustered in the west, central, north and south regions, while some counties with a significant negative risk trend clustered in north, central, and southeast NY.

Figure 3.

Discussion

In this study, in a preliminary analysis, we identified consistent increasing Covid-19 hospitalizations four or more weeks after the start of the 2023 Quebec wildfires, in contrast with the Covid-19 hospitalization trend in 2022. To estimate the association of PM_2.5 with the Covid-19 hospitalization rate in the weeks before and after the wildfires, we defined a 38-day exposure period, from 1 month before the start of the 2023 Quebec wildfires to the day with highest PM_2.5 levels, and used four 38-day hospitalization periods with a 2, 4, 6, and 8-week lag, respectively. Additionally, we compared four 7-day post-wildfire periods to assess the association of PM_2.5 during the 7 days of the wildfires with the Covid-19 hospitalization rate 1, 2, 3, and 4 days after the fires. The analysis considered the spatiotemporal distribution of the daily Covid-19 hospitalization rate in the NY counties in both years, 2002 and 2023, to estimate and map the risk of Covid-19 hospitalization after adjusting for several sociodemographic confounders. We found a higher risk of Covid-19 hospitalization associated with higher PM_2.5 levels in the last three pre/post wildfire (38-day) periods and the last two 7-day post wildfire periods in 2023, compared to 2022 when this association was not found in any of the four pre/post wildfire periods and only the first two 7-day post wildfire periods, respectively. These findings underscore the potential effect more frequent and extreme environmental events such as the Quebec wildfires can have on the severity of infectious diseases such as Covid-19 that impact the capacity of the health sector response. The use of robust spatiotemporal models to do analyses of daily updated data can help support public health decision-making, with emphasis on identifying the most vulnerable communities or specific at-risk areas.

Our findings concur with previous research that found high PM_2.5 concentration levels were associated with increased rates of COVID-19 infections in California, USA,^33–35 Colorado, USA,³⁶ Italy,³⁷ and Australia.¹³ Whereas the causal links of these associations are not yet established, epithelial function damage and altered innate immune responses have been identified as potential mediatory mechanisms.³⁸ Some studies have found that exposure to air pollution may create a favorable environment for the transmission and increased virulence of SARS-CoV-2, especially among individuals with pre-existing respiratory and cardiac conditions.³⁹ A hypothesized mechanism for this correlation is the effect of pollutants like PM_2.5 on the expression of angiotensin-converting enzyme 2 (ACE2) in respiratory cells.⁴⁰ ACE2 is known to be the receptor utilized by SARS-CoV-2 for cell entry, and its upregulation due to air pollutants may heighten the vulnerability of the respiratory system to infection.⁴⁰ If these mechanisms can facilitate a greater viral attachment especially in people with comorbidities, this can reflect in an amplified transmission and potentially more severe Covid-19 morbidity that requires hospitalized treatment. In addition to PM_2.5, other air pollutants present in wildfires smoke such as nitrogen dioxide have been linked with broader COVID-19 transmission and mortality.⁴¹ Whereas the short-term risk estimated in the 7-day post wildfire periods was less consistent than the pre/post wildfire periods, the higher risk estimated in three of these periods can be an indication of increased severity of the diseases in people already infected with Covid-19, which has been considered in previous analyses of short-term air pollution exposure and Covid-19.⁴²

We used a spatiotemporal analysis to assess the trends and associations between PM_2.5 levels and COVID-19 hospitalization rates across NY, while accounting for regional and temporal variations. This approach allows estimating the Covid-19 hospitalization risk for specific counties, in addition to the risk associated with PM_2.5 across the whole of NY state. These estimates can be used as indicators of severe morbidity per county to rank at-risk areas, an approach that has been investigated to prioritize screening and detection for Covid-19 and other diseases.^43–46 This can be especially helpful during short-notice emergencies where health services can be quickly overwhelmed, while more results and data can be progressively added to estimate outcomes beyond the initial public health response.¹⁰ Risk maps provide a statistical and visual estimation of their correspondence with high air pollution levels. Although more research at the individual level is always needed to confirm the findings of these population-based studies, spatiotemporal analyses help to support the combined evidence regarding the higher risk of severe Covid-19 morbidity after wildfires to prepare strategies in wildfire-prone areas. Robust statistical approaches such as spatiotemporal regression modeling can be used on daily updated data to provide timely analyses to support the health sector, focusing on identifying vulnerable communities or specific areas to help prepare more informed immediate responses.

It is important to acknowledge the limitations of this study, especially regarding the spatial aggregation of the Covid-19 data, with the potential risk of ecological bias where factors at the spatial group level may produce spurious associations.⁴⁷ To reduce this risk, we adjusted the analysis for sociodemographic confounders and implemented a mixed effects model.²⁸ In addition, we used a Bayesian spatial regression to smooth estimates between geographical areas and increase the robustness of the regression coefficients which can help to reduce the effect of outliers in the spatially aggregated data.⁴⁸ The ecological design of this study means that our findings can provide only statistical estimates of association rather than measures of causality between wildfire smoke and COVID 19 hospitalization and residual confounders such as sex and age-structure differences may persist, along with temporal variations in public health policies or healthcare capacities, potentially affecting the risk estimates.

This analysis did not include immunity data from vaccinations or past Covid-19 infections which could help explaining the lower Covid-19 hospitalization risk in 2022, a factor not fully isolated from the PM_2.5 effect. The incorporation of data on the lesser virulence of more recent SARS-CoV-2 variants along with improved outpatient treatments can influence the risk of severe Covid-19 morbidity and affect the regression estimates in ecological studies. We could not include these factors as this would require the availably of some individual level data beyond the scope of the current study design, and other factors including within county differences such as healthcare infrastructure and rurality. Our spatiotemporal analysis helps to compound with other evidence of the risk of the potential impacts of wildfire air pollution on Covid-19 severity close to the incubation period but does not account for potential long-term impacts months after the exposure. This would require longer periods analyses and health data on other conditions such as fatigue, myalgias, and thromboembolic diseases which are becoming increasingly challenging to access given restrictions implemented by the current federal government administration.

Conclusions

The higher levels of PM_2.5 in NY state after the 2023 Quebec wildfires were associated with higher risk of Covid-19 hospitalizations, in the first week and four or more weeks after the wildfires started compared to the same period in 2022. This associations were identified using a spatiotemporal analysis to estimate and map the Covid-19 hospitalization risk in each NY county, a methodology useful to provide rapid assessments of the potential impacts and risks of wildfire air pollution on infectious and respiratory diseases such as Covid-19. These timely analyses can inform decision making in public health for specific areas or vulnerable communities in response to an increasing prevalence of extreme environmental events. The findings of this study can be strengthened by using higher spatial and temporal resolution data and additional information on comorbidities, virus variants and vaccination status to assess their effect on the Covid-19 risk. Nevertheless, further research with individual level data is required to identify the causality links between wildfire emissions and Covid-19 severity.

Footnotes

ORCID iDs

Javier Cortes-Ramirez

Vishal Singh

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

References

McArdle

Dowling

Carey

, et al. Asthma-associated emergency department visits during the Canadian wildfire smoke episodes - United States, April- August 2023. MMWR Morb Mortal Wkly Rep 2023; 72(34): 926–932.

Department of Health. Daily hospitalization summary, https://coronavirus.health.ny.gov/daily-hospitalization-summary. 2023.

Setti

Passarini

De Gennaro

, et al. SARS-Cov-2RNA found on particulate matter of Bergamo in Northern Italy: First evidence. Environ Res 2020; 188: 109754.

Cohen

Brauer

Burnett

, et al. Estimates and 25-year trends of the global burden of disease attributable to ambient air pollution: an analysis of data from the Global Burden of Diseases Study 2015. Lancet 2017; 389(10082): 1907–1918.

Zhang

Shao

Lin

, et al. Emission inventory of carbonaceous pollutants from biomass burning in the Pearl River Delta Region, China. Atmos Environ 2013; 76: 189–199.

Zhang

Chen

, et al. The relationship between air quality and respiratory pathogens among children in Suzhou City. Ital J Pediatr 2019; 45(1): 123.

Meo

Abukhalaf

Alomar

, et al. Effect of environmental pollutants PM-2.5, carbon monoxide, and ozone on the incidence and mortality of SARS-COV-2 infection in ten wildfire affected counties in California. Sci Total Environ 2021; 757: 143948.

Meo

Abukhalaf

Alomar

, et al. Wildfire and COVID-19 pandemic: effect of environmental pollution PM-2.5 and carbon monoxide on the dynamics of daily cases and deaths due to SARS-COV-2 infection in San-Francisco USA. Eur Rev Med Pharmacol Sci 2020; 24(19): 10286–10292.

Curtis

. PM2.5, NO2, wildfires, and other environmental exposures are linked to higher Covid 19 incidence, severity, and death rates. Environ Sci Pollut Res Int 2021; 28(39): 54429–54447.

10.

Nethery

Sabath

, et al. Air pollution and COVID-19 mortality in the United States: strengths and limitations of an ecological regression analysis. Sci Adv 2020; 6(45): eabd4049.

11.

Cortes-Ramirez

Gatton

Wilches-Vega

, et al. Mapping the risk of respiratory infections using suburban district areas in a large city in Colombia. BMC Public Health 2023; 23(1): 1400.

12.

Bronte

García-García

Lee

, et al. Impact of outdoor air pollution on severity and mortality in COVID-19 pneumonia. Sci Total Environ 2023; 894: 164877.

13.

Cortes-Ramirez

Michael

Knibbs

, et al. The association of wildfire air pollution with COVID-19 incidence in New South Wales, Australia. Sci Total Environ 2022; 809: 151158.

14.

Dominici

Zigler

. Best Practices for gauging evidence of causality in air pollution epidemiology. Am J Epidemiol 2017; 186(12): 1303–1309.

15.

Alene

Yismaw

Assemie

, et al. Serial interval and incubation period of COVID-19: a systematic review and meta-analysis. BMC Infect Dis 2021; 21(1): 257.

16.

Lauer

Grantz

, et al. The incubation period of Coronavirus Disease 2019 (COVID-19) from publicly reported confirmed cases: estimation and application. Ann Intern Med 2020; 172(9): 577–582.

17.

Zhu

Kuang

, et al. Effect of PM2.5 exposure on circulating fibrinogen and IL-6 levels: a systematic review and meta-analysis. Chemosphere 2021; 271: 129565.

18.

Adhikari

Yin

Lag effects of ozone, PM2.5, and meteorological factors on COVID-19 new cases at the disease epicenter in Queens, New York. Atmos 2021; 12(3): 357.

19.

Kim

S-Y

Peel

Hannigan

, et al. The temporal lag structure of short-term associations of fine particulate matter chemical constituents and cardiovascular and respiratory hospitalizations. Environ Health Perspect 2012; 120(8): 1094–1099.

20.

Department of Health. COVID-19 data in New York, https://coronavirus.health.ny.gov/covid-19-data-new-york (2023).

21.

University of Wisconsin Population Health Institute. County health rankings and roadmaps, https://www.countyhealthrankings.org/ (2023).

22.

Remington

Catlin

Gennuso

KP.

The County Health Rankings: rationale and methods. Popul Health Metr 2015; 13(1): 11.

23.

KTL

White

Nichols

BE.

Population density and basic reproductive number of COVID-19 across United States counties. PLoS One 2021; 16: e0249271.

24.

Besag

York

Molli

Bayesian image restoration, with two applications in spatial statistics. Ann Inst Stat Math 1991; 43(1): 1–20.

25.

Knorr-Held

Bayesian modelling of inseparable space-time variation in disease risk. Stat Med 2000; 19(17–18): 2555–2567.

26.

Kim

Yoo

Hong

, et al. Health behaviors and the risk of COVID-19 incidence: a Bayesian hierarchical spatial analysis. J Infect Public Health 2023; 16(2): 190–195.

27.

Kauhl

König

Wolf

Spatial distribution of COVID-19 hospitalizations and associated risk factors in health insurance data using Bayesian spatial modelling. Int J Environ Res Public Health 2023; 20(5).

28.

Wakefield

Multi-level modelling, the ecologic fallacy, and hybrid study designs. Int J Epidemiol 2009; 38(2): 330–336.

29.

Spiegelhalter

Best

Carlin

, et al. Bayesian measures of model complexity and fit. J R Stat Soc Series B Stat Methodol 2002; 64(4): 583–639.

30.

Lindgren

Rue

Bayesian spatial modelling with R-INLA. J Stat Softw 2015; 63(19): 1–25.

31.

ESRI. ArcGIS Pro: Release 3.1. Environmental Systems Research Institute, 2022.

32.

Environmental Protection Agency. National Ambient Air Quality Standards (NAAQS) for PM, https://www.epa.gov/pm-pollution/national-ambient-air-quality-standards-naaqs-pm (2023).

33.

Ademu

Gao

Thompson

, et al. Impact of short-term air pollution on respiratory infections: a time-series analysis of COVID-19 cases in California during the 2020 wildfire season. Int J Environ Res Public Health 2022; 19(9): 5057.

34.

Hsueh

Do wildfires exacerbate COVID-19 infections and deaths in vulnerable communities? Evidence from California. J Environ Manag 2023; 328: 116918.

35.

Zhou

Josey

Kamareddine

, et al. Excess of COVID-19 cases and deaths due to fine particulate matter exposure during the 2020 wildfires in the United States. Sci Adv 2021; 7(33): eabi8789.

36.

Metz

Bauer

Epperly

, et al. Investigation of COVID-19 outbreak among wildland firefighters during wildfire response, Colorado, USA, 2020. Emerg Infect Dis 2022; 28(8): 1551–1558.

37.

Frontera

Cianfanelli

Vlachos

, et al. Severe air pollution links to higher mortality in COVID-19 patients: the “double-hit” hypothesis. J Infect 2020; 81(2): 255–259.

38.

Liu

Shi

, et al. Effects of short-term high-concentration exposure to PM2.5 on pulmonary tissue damage and repair ability as well as innate immune events. Environ Pollut 2023; 319: 121055.

39.

Fattorini

Regoli

Role of the chronic air pollution levels in the Covid-19 outbreak risk in Italy. Environ Pollut 2020; 264: 114732.

40.

Paital

Agrawal

PK.

Air pollution by NO(2) and PM(2.5) explains COVID-19 infection severity by overexpression of angiotensin-converting enzyme 2 in respiratory cells: a review. Environ Chem Lett 2021; 19(1): 25–42.

41.

Copat

Cristaldi

Fiore

, et al. The role of air pollution (PM and NO(2)) in COVID-19 spread and lethality: a systematic review. Environ Res 2020; 191: 110129.

42.

Zhu

Xie

Huang

, et al. Association between short-term exposure to air pollution and COVID-19 infection: evidence from China. Sci Total Environ 2020; 727: 138704.

43.

Cortes-Ramirez

Wraith

Sly

, et al. Mapping the morbidity risk associated with coal mining in Queensland, Australia. Int J Environ Res Public Health 2022; 19(3): 1206.

44.

Courtemanche

Soneji

Tchernis

. Modeling area-level health rankings. Health Serv Res 2015; 50(5): 1413–1431.

45.

Cortes-Ramirez

Wilches-Vega

Caicedo-Velasquez

, et al. Spatiotemporal hierarchical Bayesian analysis to identify factors associated with COVID-19 in suburban areas in Colombia. Heliyon 2024; 10(9): e30182.

46.

Cortes-Ramirez

Mengersen

Morawska

, et al. The hospitalisation risk of chronic circulatory and respiratory diseases associated with coal mining in the general population in Queensland, Australia. Sci Total Environ 2024; 949: 174989.

47.

Greenland

Morgenstern

Ecological bias, confounding, and effect modification. Int J Epidemiol 1989; 18(1): 269–274.

48.

Kang

Cramb

White

, et al. Making the most of spatial information in health: a tutorial in Bayesian disease mapping for areal data. Geospat Health 2016; 11(2): 190–198.