The confluence of the COVID-19 pandemic and wildfires on all-cause mortality in Colorado, 2010-2020
Sheena Martenies
8/9/2021
1 Overview
The COVID-19 pandemic has had a tremendous public health impact in the United States. As of July 2021, there have been more than 33.6 million confirmed cases of the disease and more than 600,000 deaths due to COVID-19 infection. COVID-19 has emerged as the third leading cause of death in the United States in 2020, falling right behind heart disease and cancer.
During a wave of infections in the fall of 2020, the US Centers for Disease Control and Prevention (CDC) published a study on the excess mortality associated with the COVID-19 pandemic in the United States. In this analysis, CDC reported that, nationwide, the COVID-19 pandemic had been linked to an estimated 299,000 excess deaths between January 26 and October 3, 2020. Of these deaths, roughly 198,000 were attributed to COVID-19 and the remaining 101,000 were attributable to other causes of death. There are several possible explanations for these additional excess deaths during the studied period, including an aging population and increasing trends in certain causes of death (e.g., drug overdose deaths). Alternatively, the number of excess deaths during the pandemic may be attributable to healthcare avoidance behaviors or disruptions in ongoing care access or utilization. A recent study by CDC found 41% of adults in the US had avoided seeking healthcare during the pandemic; of these avoided health care interactions, 12% were for urgent or emergency care and 32% were for routine medical care. Non-white individuals, individuals with disabilities, and individuals with two or more chronic health conditions were more likely to avoid healthcare during the ongoing pandemic.
In addition to the substantial public health burden associated with the COVID-19 pandemic, 2020 was also a historic year for wildfires in the western United States. In particular, Colorado experienced one of the worst fire seasons in the state’s history. The Cameron Peak fire, the largest on record, burned more than 206,000 acres as of October, 2020.Wildfires pose an immediate threat to human health and safety, but wildfire smoke is also associated with downstream morbidity and mortality among those not directly impacted by the flames. Wildfire smoke has been linked to morbidity and mortality in both children and adults, with the strongest evidence for respiratory disease outcomes.
Understanding the combined influence of these two public health challenges will be critical to developing strategies to protect public health in the future. In general, wildfire events are thought to result in increased health seeking behaviors.18 Access to care during a natural disaster such as wildfires is an important tool for reducing the overall morbidity and mortality associated with the event. However, evidence suggests people tended to avoided healthcare settings during the worst of the COVID-19 pandemic. Individuals with chronic health conditions exacerbated by wildfire smoke exposure may not have received appropriate care during these wildfire smoke events. Alternatively, because wildfire smoke tends to have the largest impact on ambient air quality and because Colorado residents were largely under stay-at-home orders during the 2020 wildfire season, exposures to wildfire smoke may have been lower during this time, results in reduced mortality during wildfire events. Thus, it is important to investigate how these two emergency situations jointly influenced health outcomes during the summer and fall of 2020.
2 Objectives
In this study, we aimed to examine the potential confluence of wildfire smoke and the COVID-19 pandemic on mortality risk in Colorado. Here we present the initial findings from this study investigating associations between the pandemic, key environmental exposures such as wildfire smoke or fine particulate matter, and all-cause mortality among residents living in the Front Range region of Colorado. We hypothesized that effect of wildfire smoke on the risk of death from all causes would be higher during the pandemic compared to previous periods.
3 Methods
3.1 Mortality and Population Data
Daily mortality counts from January 1, 2010 through December 31, 2020 were obtained from the Colorado Center for Health and Environmental Data (CHED). Each death that occurred during the study period was assigned to a census tract (using the 2010 census tract boundaries) based on residential address of the individual.
Population counts at the census tract level were derived from the American Community Survey (ACS) 5-year estimates. ACS data were obtained using the ‘tidycensus’ package in R. For each year of our analysis (2010-2020), we used the population estimates for the end of the five-year period (e.g., the 2010 population was derived from the 2006-2010 ACS data set). Because the 2016-2020 ACS survey is not yet available, and because we assigned deaths to the 2010 census tract boundaries, we assumed population estimates from 2019 were representative of 2020 populations.
In addition to population estimates, we also obtained four indicators of socioeconomic status from the ACS survey data. These four indicators were: the percentage of the population 25 years or older without a high school diploma (or equivalent); the percentage of households receiving food assistance (i.e., Supplemental Nutrition Assistance Program benefits) within the last year; the percentage of households occupied by renters; and median income (in inflation-adjusted dollars).
3.2 Exposure Assessment
3.2.1 Wildfire smoke exposures
Wildfire smoke plume data is available from the Hazard Mapping System (HMS) which run by the National Oceanic and Atmospheric Association (NOAA). Each day, HMS analysts visually inspect satellite imagery to develop smoke plume shapefiles. Daily shapefiles were downloaded for the entire study period (2010-2020). Wildfire smoke (WFS) exposures were assessed at the census tract level. We considered the population within a census tract to be exposed to WFS if an HMS-identified plume intersected with the boundary of the census tract.
3.2.2 Air pollutant and temperature exposures
Daily air pollutant meteorological variables were obtained from the Colorado Department of Public Health and Environment. For each day in the study period, we downloaded hourly measurements of fine particulate matter (aerodynamic diameter < 2.5 μm; PM2.5), ozone (O3), and temperature (°F). The number of monitors with available data varied by pollutant and day. The median (range) number of monitors available each day for PM2.5, O3, and temperature were 10 (0-13), 14 (0-15), and 13 (0-16), respectively. For each day and monitor, we calculated a daily mean concentration for all three variables when at least 75% of hourly observations were available. For O3, we also calculated the daily 1-hour maximum concentration.
We assessed daily air pollutant and temperature exposures at the census tract level using inverse distance weighted interpolation. Because the topography in the Front Range creates unique meteorological conditions that influence air and pollutant flows in the region, we excluded four monitors located the foothills to the west of the study area. We also excluded any census tracts where the elevation of the centroid differed by more than 25% from the elevation of the closest monitor. When interpolating daily concentrations at census tract centroids, we excluded all monitors that were farther than 50 km from the centroid.
3.2.3 COVID-19 Pandemic
The first confirmed case of COVID-19 was documented in Colorado on March 5, 2020. However, recent reports have suggested the virus was circulating before that time. Due to the difficulties in identifying when the virus was present in the Front Range, we elected to use March 5, 2020 as the nominal start of the pandemic. We operationalized the pandemic in our models as a binomial variable, where all dates on or after March 5, 2020 were assigned a value of 1.
3.3 Statistical Analysis
Because the daily risk of mortality is low at the census tract level, more than 90% of observations in our original census-tract level data set were zeros. Therefore, we chose to aggregate our daily data to (partial) counties. Mortality counts and populations in each census tracts were aggregated to the county level. Daily county-level air pollutant concentrations and daily temperatures were estimated using the population-weighted mean of values estimated at the census tract level. We considered a county to be exposed to WFS if any census tract within that county was exposed to WFS on that day. For indicators of socioeconomic status, we used values at the county level obtained from the ACS.
We modeled associations between our exposures of interest and mortality counts using single exposure models. We considered exposures on the same day at the deaths (lag 0) as well as one day prior (lag 1) and two days prior (lag 2). Because of the potential for delayed effects, we also considered the rolling average of the previous 11 days. The long-term WFS exposure variables were assigned a value of 1 if there was any WFS in the previous 11 days. We focused on an 11-day rolling average because this is the average length of time between contracting the virus and death among a cohort of Denver residents who contracted COVID-19 in 2020 (data not shown). All single pollutant models included indicator variables for county (where Denver was the reference county), year (where 2010 was the reference year), and county-level measures of socioeconomic status (as continuous variables). We included a penalized cubic regression spline term for day of year (1 to 365/366) to account for temporal trends in mortality and an offset term (as the log of the at-risk population). Continuous exposures and covariates were scaled prior to fitting the models and effect sizes were reported for a standard deviation increase in each pollutant.
To assess the potential joint effects of the COVID-19 pandemic and our exposures of interest, we fit GAMs that included interaction terms for the pandemic and each pollutant. We fit our updated GAMs using data from 2010 through 2020. Each model included the same penalized spline term for day of the year, indicator variables for county and year, and county-level measures of socioeconomic status. Interactions between the pandemic and pollutant of interest were visualized using marginal effects plots generated by the sjPlot package in R.
4 Results
4.1 Trends in mortality data
Our data set includes 52234 daily observations from 13 counties.
Weekly counts at the county level show the expected seasonal trends. Weekly crude mortality rates are higher in the winter months compared to summer. We also see a sharp increase in weekly mortality at the start of the pandemic and into the fall of 2020.
We also observed differences in annual rates by (partial) county over the eleven year period. Annual crude mortality rates tended to increase from 2010 to 2020, partially driven by changes in the age distribution of the population.
4.2 Trends in exposure data
Wildfire smoke exposures (assessed as a binary variable) varied by year. The fewest number of county-days impacted by wildfire smoke occurred in 2019 (3.1% of county-days). The greatest number of impacted county-days was in 2020 (21.4%), coinciding with that year’s severe wildfire season. Other heavily impacted years were 2012 (18.5% of county-days were impacted by wildfires) and 2018 (16.9 % of county-days were impacted).
| Year | County-Days | County-Days with WFS | Percent County-Days Impacted |
|---|---|---|---|
| 2010 | 4745 | 143 | 3 |
| 2011 | 4745 | 260 | 6 |
| 2012 | 4758 | 868 | 19 |
| 2013 | 4745 | 524 | 11 |
| 2014 | 4745 | 258 | 6 |
| 2015 | 4745 | 290 | 6 |
| 2016 | 4758 | 451 | 10 |
| 2017 | 4745 | 513 | 11 |
| 2018 | 4745 | 795 | 17 |
| 2019 | 4745 | 294 | 6 |
| 2020 | 4758 | 1015 | 21 |
Daily population-weighted estimates for PM2.5, O3, and temperature at the county level are summarized below:
| Pollutant | Period | Min | Mean | SD | Max |
|---|---|---|---|---|---|
| PM2.5 | Lag 0 | 6.0 | 51.4 | 13.4 | 143.0 |
| PM2.5 | Lag 1 | 6.0 | 51.4 | 13.4 | 143.0 |
| PM2.5 | Lag 2 | 6.0 | 51.4 | 13.4 | 143.0 |
| PM2.5 | 7-Day | 22.7 | 51.4 | 11.5 | 113.6 |
| PM2.5 | 11-Day | 24.7 | 51.4 | 11.2 | 97.1 |
| O3 | Lag 0 | 1.0 | 7.6 | 4.5 | 70.0 |
| O3 | Lag 1 | 1.0 | 7.6 | 4.4 | 70.0 |
| O3 | Lag 2 | 1.0 | 7.6 | 4.4 | 70.0 |
| O3 | 7-Day | 2.0 | 7.7 | 3.1 | 38.7 |
| O3 | 11-Day | 2.4 | 7.7 | 2.8 | 33.5 |
| Temperature | Lag 0 | -9.7 | 51.5 | 17.6 | 89.3 |
| Temperature | Lag 1 | -9.7 | 51.5 | 17.6 | 89.3 |
| Temperature | Lag 2 | -9.7 | 51.5 | 17.6 | 89.3 |
| Temperature | 7-Day | 6.1 | 51.6 | 16.3 | 82.1 |
| Temperature | 11-Day | 11.5 | 51.6 | 16.0 | 81.5 |
Trends in air pollutant and meterological data were also as expected. We observed higher population-weighted ozone concentrations in the summer months, corresponding with higher temperature exposures. Daily PM2.5 concentrations showed less temporal variability.
4.3 Correlations between exposures
We observed a high degree of correlation between the continuous exposure variables in our data set. In particular, temperature and ozone exposures were strongly positively correlated.
4.4 Single pollutant models: all-cause mortality
We first fit single-pollutant models used exposures with lags ranging from 0 to 2 days. We also assessed the 7-day and 11-day rolling average exposures. All models included a smoothed term for day, log(population) offset, and covariates representing county-level socioeconomic status.
All effect sizes are reported for a 1-unit increase in binary variables (pandemic or wildfire smoke) or an SD increase in continuous exposures (PM2.5, O3, and temperature).
No environmental predictors
In a single pollutant model adjusted for county, year, and county-level SES, we found that the pandemic was associated with an 18% increase in daily mortality risk (IRR = 1.18, 95% CI: 1.14 - 1.22).
| Predictors | Incidence Rate Ratios | CI | p |
| Pandemic | 1.18 | 1.14 – 1.22 | <0.001 |
| Observations | 52234 | ||
| R2 | 0.774 | ||
PM2.5
PM2.5 exposures assessed at the county level were not associated with increased mortality risk , regardless of the exposure period assessed.
| Predictors | Incidence Rate Ratios | CI | p | Incidence Rate Ratios | CI | p | Incidence Rate Ratios | CI | p | Incidence Rate Ratios | CI | p | Incidence Rate Ratios | CI | p |
| Lag 0 | 1.00 | 1.00 – 1.00 | 0.870 | ||||||||||||
| Lag 1 | 1.00 | 1.00 – 1.00 | 0.937 | ||||||||||||
| Lag 2 | 1.00 | 1.00 – 1.00 | 0.616 | ||||||||||||
| 7-Day Rolling Mean | 1.00 | 1.00 – 1.00 | 0.709 | ||||||||||||
| 11-Day Rolling Mean | 1.00 | 1.00 – 1.00 | 0.783 | ||||||||||||
| Observations | 49398 | 49398 | 49398 | 46425 | 45286 | ||||||||||
| R2 | 0.765 | 0.765 | 0.765 | 0.756 | 0.752 | ||||||||||
Ozone
We observed an small increase in mortality risk for same-day (lag 0) ozone exposures (IRR = 1.01, 95% CI: 1.00 - 1.01 for an SD increase in ozone) in a single pollutant model. Interestingly, an SD increase 11-day rolling mean ozone was associated with a decrease in mortality risk (IRR = 0.98, 95% CI: 0.97 - 0.99).
| Predictors | Incidence Rate Ratios | CI | p | Incidence Rate Ratios | CI | p | Incidence Rate Ratios | CI | p | Incidence Rate Ratios | CI | p | Incidence Rate Ratios | CI | p |
| Lag 0 | 1.01 | 1.00 – 1.01 | 0.049 | ||||||||||||
| Lag 1 | 1.00 | 1.00 – 1.01 | 0.565 | ||||||||||||
| Lag 2 | 1.00 | 0.99 – 1.00 | 0.308 | ||||||||||||
| 7-Day Rolling Mean | 1.00 | 0.99 – 1.01 | 0.697 | ||||||||||||
| 11-Day Rolling Mean | 0.98 | 0.97 – 0.99 | 0.001 | ||||||||||||
| Observations | 44042 | 44042 | 44042 | 43715 | 46167 | ||||||||||
| R2 | 0.716 | 0.716 | 0.716 | 0.717 | 0.755 | ||||||||||
Temperature
Same day temperature exposures were associated with increased mortality risk. Similar to ozone, lag 2, 7-day rolling mean, and 11-day rolling mean temperatures were associated with decreased mortality risk.
| Predictors | Incidence Rate Ratios | CI | p | Incidence Rate Ratios | CI | p | Incidence Rate Ratios | CI | p | Incidence Rate Ratios | CI | p | Incidence Rate Ratios | CI | p |
| Lag 0 | 1.01 | 1.00 – 1.02 | 0.010 | ||||||||||||
| Lag 1 | 1.00 | 1.00 – 1.01 | 0.206 | ||||||||||||
| Lag 2 | 0.99 | 0.98 – 1.00 | 0.020 | ||||||||||||
| 7-Day Rolling Mean | 0.99 | 0.98 – 1.00 | 0.012 | ||||||||||||
| 11-Day Rolling Mean | 0.98 | 0.97 – 0.99 | 0.001 | ||||||||||||
| Observations | 46794 | 46791 | 46788 | 46380 | 46167 | ||||||||||
| R2 | 0.758 | 0.758 | 0.758 | 0.756 | 0.755 | ||||||||||
Wildfire smoke
We did not observe any associations between our binary wildfire smoke variable and mortality risk.
| Predictors | Incidence Rate Ratios | CI | p | Incidence Rate Ratios | CI | p | Incidence Rate Ratios | CI | p | Incidence Rate Ratios | CI | p | Incidence Rate Ratios | CI | p |
| Lag 0 | 1.01 | 0.99 – 1.02 | 0.440 | ||||||||||||
| Lag 1 | 1.00 | 0.99 – 1.02 | 0.820 | ||||||||||||
| Lag 2 | 1.00 | 0.99 – 1.01 | 0.870 | ||||||||||||
| 7-Day Window | 1.00 | 0.98 – 1.01 | 0.457 | ||||||||||||
| 11-Day Window | 1.00 | 0.99 – 1.01 | 0.981 | ||||||||||||
| Observations | 51350 | 51350 | 51350 | 47895 | 46306 | ||||||||||
| R2 | 0.773 | 0.773 | 0.773 | 0.773 | 0.773 | ||||||||||
4.5 Interaction Models: Lag 0
Next we examined interactions between the pandemic and same-day (lag 0) environmental exposures. Again, all models included a smoothed term for day, log(population) offset, and covariates representing county-level socioeconomic status. Two-way interaction terms were included between the pandemic and the primary exposure of interest. These models were also adjusted for other exposures using the same exposure period (lag 0 days).
PM2.5
| Predictors | Incidence Rate Ratios | CI | p |
| Mean PM2.5 | 1.00 | 1.00 – 1.01 | 0.456 |
| Pandemic | 1.18 | 1.15 – 1.22 | <0.001 |
| Max O3 | 1.01 | 1.00 – 1.01 | 0.085 |
| Mean Temperature | 1.01 | 1.00 – 1.02 | 0.101 |
| Wildfire Smoke | 1.00 | 0.99 – 1.02 | 0.604 |
| Mean PM2.5*Pandemic | 0.98 | 0.97 – 0.99 | <0.001 |
| Observations | 41051 | ||
| R2 | 0.724 | ||
Ozone
| Predictors | Incidence Rate Ratios | CI | p |
| Max O3 | 1.01 | 1.00 – 1.02 | 0.018 |
| Pandemic | 1.18 | 1.15 – 1.22 | <0.001 |
| Mean PM2.5 | 1.00 | 1.00 – 1.00 | 0.703 |
| Mean Temperature | 1.01 | 1.00 – 1.02 | 0.083 |
| Wildfire Smoke | 1.00 | 0.99 – 1.02 | 0.802 |
| Max O3*Pandemic | 0.96 | 0.95 – 0.98 | <0.001 |
| Observations | 41051 | ||
| R2 | 0.724 | ||
Temperature
| Predictors | Incidence Rate Ratios | CI | p |
| Mean Temperature | 1.01 | 1.01 – 1.02 | 0.002 |
| Pandemic | 1.19 | 1.16 – 1.23 | <0.001 |
| Mean PM2.5 | 1.00 | 1.00 – 1.00 | 0.707 |
| Max O3 | 1.00 | 1.00 – 1.01 | 0.251 |
| Wildfire Smoke | 1.00 | 0.99 – 1.02 | 0.554 |
| Mean Temperature*Pandemic | 0.94 | 0.93 – 0.96 | <0.001 |
| Observations | 41051 | ||
| R2 | 0.725 | ||
Wildfire smoke
| Predictors | Incidence Rate Ratios | CI | p |
| Wildfire Smoke | 1.03 | 1.01 – 1.04 | 0.002 |
| Pandemic | 1.22 | 1.18 – 1.25 | <0.001 |
| Mean PM2.5 | 1.00 | 1.00 – 1.00 | 0.894 |
| Max O3 | 1.01 | 1.00 – 1.01 | 0.107 |
| Mean Temperature | 1.01 | 1.00 – 1.02 | 0.089 |
| Wildfire Smoke*Pandemic | 0.88 | 0.85 – 0.91 | <0.001 |
| Observations | 41051 | ||
| R2 | 0.725 | ||
Marginal Effects Plots
In general, we observed interactions between our environmental exposures (assessed at lag 0) and the pandemic, where mortality risk. The interaction between the pandemic and PM2.5, ozone, temperature, and wildfire smoke resulted in lower all-cause mortality risk.
4.6 Interaction Models: Lag 2
Next we examined interactions between the pandemic and lag 2 environmental exposures. Again, all models included a smoothed term for day, log(population) offset, and covariates representing county-level socioeconomic status. Two-way interaction terms were included between the pandemic and the primary exposure of interest. These models were also adjusted for other exposures using the same exposure period (lag 2 days).
PM2.5
| Predictors | Incidence Rate Ratios | CI | p |
| Mean PM2.5 | 1.00 | 1.00 – 1.01 | 0.332 |
| Pandemic | 1.19 | 1.15 – 1.22 | <0.001 |
| Max O3 | 1.00 | 1.00 – 1.01 | 0.229 |
| Mean Temperature | 0.99 | 0.98 – 1.00 | 0.027 |
| Wildfire Smoke | 1.00 | 0.98 – 1.01 | 0.594 |
| Mean PM2.5*Pandemic | 0.98 | 0.97 – 0.99 | <0.001 |
| Observations | 41043 | ||
| R2 | 0.724 | ||
Ozone
| Predictors | Incidence Rate Ratios | CI | p |
| Max O3 | 1.01 | 1.00 – 1.01 | 0.035 |
| Pandemic | 1.19 | 1.15 – 1.22 | <0.001 |
| Mean PM2.5 | 1.00 | 1.00 – 1.00 | 0.879 |
| Mean Temperature | 0.99 | 0.98 – 1.00 | 0.039 |
| Wildfire Smoke | 0.99 | 0.98 – 1.01 | 0.443 |
| Max O3*Pandemic | 0.96 | 0.94 – 0.97 | <0.001 |
| Observations | 41043 | ||
| R2 | 0.725 | ||
Temperature
| Predictors | Incidence Rate Ratios | CI | p |
| Mean Temperature | 1.00 | 0.99 – 1.01 | 0.594 |
| Pandemic | 1.20 | 1.16 – 1.23 | <0.001 |
| Mean PM2.5 | 1.00 | 1.00 – 1.00 | 0.939 |
| Max O3 | 1.00 | 1.00 – 1.01 | 0.517 |
| Wildfire Smoke | 1.00 | 0.98 – 1.01 | 0.674 |
| Mean Temperature*Pandemic | 0.94 | 0.92 – 0.95 | <0.001 |
| Observations | 41043 | ||
| R2 | 0.725 | ||
Wildfire smoke
| Predictors | Incidence Rate Ratios | CI | p |
| Wildfire Smoke | 1.02 | 1.00 – 1.04 | 0.027 |
| Pandemic | 1.22 | 1.18 – 1.26 | <0.001 |
| Mean PM2.5 | 1.00 | 1.00 – 1.01 | 0.614 |
| Max O3 | 1.00 | 1.00 – 1.01 | 0.268 |
| Mean Temperature | 0.99 | 0.98 – 1.00 | 0.031 |
| Wildfire Smoke*Pandemic | 0.87 | 0.85 – 0.90 | <0.001 |
| Observations | 41043 | ||
| R2 | 0.725 | ||
Marginal Effects Plots
Our lag 2 models showed similar results to those for the lag 0 exposures. The interaction between the pandemic and PM2.5, ozone, temperature, and wildfire smoke resulted in lower all-cause mortality risk.
4.7 Interaction Models: 7-Day Rolling Mean
Next we examined interactions between the pandemic and 7-day rolling average environmental exposures. Again, all models included a smoothed term for day, log(population) offset, and covariates representing county-level socioeconomic status. Two-way interaction terms were included between the pandemic and the primary exposure of interest. These models were also adjusted for other exposures using the same exposure period (7-day window).
PM2.5
| Predictors | Incidence Rate Ratios | CI | p |
| Mean PM2.5 | 1.00 | 1.00 – 1.01 | 0.341 |
| Pandemic | 1.19 | 1.15 – 1.23 | <0.001 |
| Max O3 | 1.02 | 1.01 – 1.03 | 0.004 |
| Mean Temperature | 0.98 | 0.97 – 0.99 | 0.003 |
| Wildfire Smoke | 1.00 | 0.99 – 1.01 | 0.721 |
| Mean PM2.5*Pandemic | 0.98 | 0.97 – 0.98 | <0.001 |
| Observations | 37637 | ||
| R2 | 0.725 | ||
Ozone
| Predictors | Incidence Rate Ratios | CI | p |
| Max O3 | 1.02 | 1.01 – 1.03 | 0.002 |
| Pandemic | 1.19 | 1.15 – 1.23 | <0.001 |
| Mean PM2.5 | 1.00 | 0.99 – 1.00 | 0.458 |
| Mean Temperature | 0.98 | 0.97 – 0.99 | 0.005 |
| Wildfire Smoke | 0.99 | 0.98 – 1.01 | 0.277 |
| Max O3*Pandemic | 0.95 | 0.94 – 0.97 | <0.001 |
| Observations | 37637 | ||
| R2 | 0.726 | ||
Temperature
| Predictors | Incidence Rate Ratios | CI | p |
| Mean Temperature | 0.99 | 0.98 – 1.00 | 0.116 |
| Pandemic | 1.20 | 1.16 – 1.24 | <0.001 |
| Mean PM2.5 | 1.00 | 0.99 – 1.00 | 0.672 |
| Max O3 | 1.01 | 1.00 – 1.02 | 0.087 |
| Wildfire Smoke | 1.00 | 0.98 – 1.01 | 0.461 |
| Mean Temperature*Pandemic | 0.93 | 0.92 – 0.94 | <0.001 |
| Observations | 37637 | ||
| R2 | 0.726 | ||
Wildfire smoke
| Predictors | Incidence Rate Ratios | CI | p |
| Wildfire Smoke | 1.01 | 1.00 – 1.03 | 0.069 |
| Pandemic | 1.25 | 1.21 – 1.29 | <0.001 |
| Mean PM2.5 | 1.00 | 1.00 – 1.01 | 0.877 |
| Max O3 | 1.01 | 1.00 – 1.02 | 0.050 |
| Mean Temperature | 0.98 | 0.97 – 0.99 | 0.002 |
| Wildfire Smoke*Pandemic | 0.87 | 0.85 – 0.90 | <0.001 |
| Observations | 37637 | ||
| R2 | 0.726 | ||
Marginal Effects Plots
The interaction trends persisted when examining the 7-day rolling average exposures.
4.8 Interaction Models: 11-day window Rolling Mean
Next we examined interactions between the pandemic and 11-day window rolling average environmental exposures. Again, all models included a smoothed term for day, log(population) offset, and covariates representing county-level socioeconomic status. Two-way interaction terms were included between the pandemic and the primary exposure of interest. These models were also adjusted for other exposures using the same exposure period (11-day window window).
PM2.5
| Predictors | Incidence Rate Ratios | CI | p |
| Mean PM2.5 | 1.00 | 0.99 – 1.01 | 0.880 |
| Pandemic | 1.20 | 1.16 – 1.24 | <0.001 |
| Max O3 | 1.02 | 1.01 – 1.03 | 0.002 |
| Mean Temperature | 0.97 | 0.95 – 0.98 | <0.001 |
| Wildfire Smoke | 1.01 | 0.99 – 1.02 | 0.379 |
| Mean PM2.5*Pandemic | 0.97 | 0.96 – 0.98 | <0.001 |
| Observations | 36136 | ||
| R2 | 0.725 | ||
Ozone
| Predictors | Incidence Rate Ratios | CI | p |
| Max O3 | 1.02 | 1.01 – 1.03 | 0.003 |
| Pandemic | 1.20 | 1.16 – 1.24 | <0.001 |
| Mean PM2.5 | 0.99 | 0.99 – 1.00 | 0.037 |
| Mean Temperature | 0.97 | 0.95 – 0.99 | <0.001 |
| Wildfire Smoke | 1.00 | 0.99 – 1.01 | 0.994 |
| Max O3*Pandemic | 0.95 | 0.94 – 0.96 | <0.001 |
| Observations | 36136 | ||
| R2 | 0.726 | ||
Temperature
| Predictors | Incidence Rate Ratios | CI | p |
| Mean Temperature | 0.98 | 0.96 – 0.99 | 0.006 |
| Pandemic | 1.21 | 1.17 – 1.25 | <0.001 |
| Mean PM2.5 | 1.00 | 0.99 – 1.00 | 0.092 |
| Max O3 | 1.01 | 1.00 – 1.02 | 0.086 |
| Wildfire Smoke | 1.00 | 0.99 – 1.02 | 0.728 |
| Mean Temperature*Pandemic | 0.93 | 0.92 – 0.94 | <0.001 |
| Observations | 36136 | ||
| R2 | 0.726 | ||
Wildfire smoke
| Predictors | Incidence Rate Ratios | CI | p |
| Wildfire Smoke | 1.02 | 1.00 – 1.03 | 0.009 |
| Pandemic | 1.26 | 1.22 – 1.30 | <0.001 |
| Mean PM2.5 | 1.00 | 0.99 – 1.00 | 0.320 |
| Max O3 | 1.01 | 1.00 – 1.03 | 0.063 |
| Mean Temperature | 0.97 | 0.95 – 0.98 | <0.001 |
| Wildfire Smoke*Pandemic | 0.88 | 0.86 – 0.90 | <0.001 |
| Observations | 36136 | ||
| R2 | 0.726 | ||
Marginal Effects Plots
The interaction trends persisted when examining the 11-day rolling average exposures.
5 Summary of Preliminary Findings
- When using data for an 11-year period, we found that the COVID-19 pandemic was associated with an 18% increase in daily mortality risk among residents of the Front Range region
- We found higher mortality risk associated with an increase in same-day ozone and temperature exposures in single-pollutant models. Associations were reversed when looking at longer exposure windows (e.g. 7-day rolling average)
- PM2.5 and wildfire smoke exposures were not associated with mortality risk in single-pollutant models
- In models where we included an interaction between the pandemic and environmental exposures, the interaction term was inversely associated with daily mortality risk
6 Next Steps
Additional analyses are needed to confirm our findings and more deeply examine these effects. We are planning the following as next steps in our analysis:
- Examine these trends when stratified by age (e.g., 65 years and older vs. other age groups)
- Examine how trends vary based on cause of death (e.g., CVD, respiratory, and deaths of despair)
- Examine confounding by other potential explanatory variables