Nationwide assessment of mortality attributable to acute exposure to wildfire-related PM2.5 in Brazil

Recent epidemiological and toxicological studies suggest that fine particulate matter (PM2.5) from wildfires may be more harmful than PM2.5 from other sources, with stronger associations reported for respiratory outcomes and indications of higher toxicity of biomass particles. In Brazil, prior assessments of fire-related air pollution largely relied on total particulate matter during wildfire events and all-cause mortality, leaving uncertainties about cause-specific mortality risks and susceptible subpopulations. Evidence has also been limited to single cities or regions during burning seasons, with limited understanding of nationwide impacts and geographic or demographic variability. This study addresses these gaps by using a nationwide mortality dataset to quantify cause-specific, demographic, geographic, and temporal variations in the associations between short-term exposure to wildfire-related PM2.5 and mortality in Brazil.

Emerging evidence indicates that wildfire-related PM2.5 may exert stronger health effects than non-wildfire PM2.5. Aguilera et al. found greater increases in respiratory hospitalizations associated with wildfire-specific PM2.5 compared with non-wildfire PM2.5. Toxicological studies report biomass-derived particles exhibiting greater toxicity than those from fossil fuels. In Brazil, studies over the past five years estimated premature deaths attributable to fire emissions but generally used total PM during wildfire periods and focused on all-cause mortality, without disaggregating specific causes such as cardiovascular or respiratory deaths. Geographic coverage was often limited to single cities or regions near fire sources, constraining generalizability. The present study builds on this literature by employing source-specific wildfire PM2.5 estimates, nationwide coverage, and cause-specific outcomes to provide more comprehensive risk and burden estimates.

Study design and population: A nationwide, two-stage time-series analysis was conducted for Brazil from 2000 to 2016. Daily death records were obtained from the Brazil Mortality Information System (SIM), including municipality, age, sex, date of death, and primary cause coded by ICD-10. Cause categories included cardiovascular (I00–I99) and respiratory (J00–J99). Daily counts for all-cause, cardiovascular, and respiratory mortality were aggregated to 510 immediate regions defined by the Brazilian Institute of Geography and Statistics (IBGE). Age groups were 0–59 years and ≥60 years; analyses were stratified by sex. Exposure and meteorology: Wildfire-related PM2.5 was estimated using GEOS-Chem by taking the difference between simulations with and without fire emissions at 2.0° × 2.5° resolution, then downscaled to 0.25° × 0.25° as the ratio of wildfire-related to all-source PM2.5 via inverse distance weighted interpolation. All-source PM2.5 from GEOS-Chem was validated and calibrated against global ground-level monitoring using a random forest model with meteorological predictors (10-fold cross-validation: R2 = 86.5%, RMSE = 15.1 μg/m3). Calibrated wildfire-related PM2.5 was obtained by multiplying calibrated all-source PM2.5 with the downscaled wildfire-to-all-source ratio. Meteorological variables (temperature and dew point) were from ERA5 reanalysis at 0.25° × 0.25°, aggregated to daily temperature and relative humidity; region-level values were population-weighted averages. Statistical analysis: In the first stage, quasi-Poisson generalized linear models were fit for each immediate region to estimate associations between daily wildfire-related PM2.5 and mortality, using a distributed lag approach with a cross-basis defined by a linear exposure function and a natural cubic spline over lags 0–14 days (4 df). Models adjusted for: 21-day moving average temperature (natural cubic spline, 3 df), 7-day moving average relative humidity (natural cubic spline, 3 df), day of week, public holidays, and long-term/seasonal trends via a natural cubic spline of time with 7 df per year. In the second stage, region-specific estimates were pooled nationally by meta-analysis to obtain relative risks (RRs) and 95% CIs per 10 μg/m3 increase in wildfire-related PM2.5 for lag-specific and cumulative (lag 0–14) effects. Attributable deaths (AD), attributable fractions (AF), and attributable mortality rates (AMR; per million population per year) were computed using standard formulas with uncertainty propagated via RR confidence bounds. Sensitivity analyses varied the maximum lag (13, 15, 16 days), degrees of freedom for lag splines (3 or 5), meteorological spline df (4 or 5), and temperature moving average window (14 days) to assess robustness.

Descriptive statistics: Across 510 immediate regions (2000–2016), there were 18,681,906 all-cause deaths, including 5,271,936 cardiovascular and 1,954,849 respiratory deaths. Mean daily counts were 3008 (all-cause), 849 (cardiovascular), and 315 (respiratory). Adults ≥60 years accounted for 61.4% of all-cause deaths; females for 42.7%. Wildfire-related PM2.5 averaged 2.8 μg/m3 (SD 2.7) with smoke occurring in all months, varying by season and region. Exposure-response and lag patterns: Associations between wildfire-related PM2.5 and mortality were approximately linear. Effects were acute, with evidence of mortality displacement over lag days 2–6. A 14-day maximum lag captured lagged effects. Cumulative risk estimates (lag 0–14) per 10 μg/m3 increase in wildfire-related PM2.5: All-cause mortality RR 1.031 (95% CI 1.024–1.039, p < 0.001); cardiovascular RR 1.026 (1.015–1.038, p < 0.001); respiratory RR 1.077 (1.059–1.095, p < 0.001). Respiratory effects were stronger than all-cause and cardiovascular. Subgroup differences: Older adults (≥60 years) and females appeared more sensitive, with statistically significant differences for all-cause and respiratory mortality (p for difference < 0.05). Geographic heterogeneity was observed for all-cause mortality, with cumulative effects ranging from North RR 0.992 (0.971–1.012; not significant) to Southeast RR 1.060 (1.047–1.072; highest), with Northeast RR 1.023 (1.008–1.037), South RR 1.011 (0.995–1.027; not significant), and Central West RR 1.027 (1.016–1.038). Attributable burden (lag 0–14): An estimated 130,273 all-cause deaths (95% CI 76,534–183,346), 32,961 cardiovascular deaths (7,628–57,756), and 33,807 respiratory deaths (19,225–47,919) were attributable to acute wildfire-related PM2.5 exposure during the study period. Corresponding AFs were 0.70% (0.41–0.98) for all-cause, 0.63% (0.14–1.10) for cardiovascular, and 1.73% (0.98–2.45) for respiratory mortality. Annual attributable rates per million population were 54.8 (32.2–77.1; all-cause). Markedly higher AMRs were observed among adults ≥60 years: all-cause 383.9 (215.4–549.7), cardiovascular 107.7 (13.1–199.8), and respiratory 117.4 (67.2–165.8) per million per year. Regionally, the highest attributable mortality rate was in the Central West, followed by the Southeast. Some regions (e.g., North) showed non-significant or negative AFs consistent with null effect estimates. Robustness: Sensitivity analyses confirmed that a 14-day lag window was sufficient and results were robust to alternative spline df and meteorological adjustments.

The study demonstrates acute increases in all-cause, cardiovascular, and especially respiratory mortality associated with short-term wildfire-related PM2.5 exposure across Brazil, addressing prior gaps by using source-specific exposure estimates and nationwide coverage. The stronger respiratory effects align with toxicological evidence of heightened biomass particle toxicity. Geographic variation indicates greater susceptibility in the Southeast and lower or null effects in the North/Northeast, potentially reflecting differences in exposure intensity, population vulnerability, healthcare access, or pollution mixtures. Demographic analyses suggest older adults and females are more vulnerable to wildfire smoke, with significant differences for all-cause and respiratory mortality; however, literature on sex differences remains mixed and may depend on exposure and outcome specifics. The estimated total attributable deaths (about 130,000 over 2000–2016; roughly 7,700 annually) are consistent in magnitude with prior South American estimates despite methodological differences, reinforcing the substantial public health burden from wildfire smoke. These findings directly address the research question by quantifying cause-specific and subgroup-specific risks and burdens, elucidating lag structures, and highlighting spatial heterogeneity, thereby informing targeted risk communication and mitigation strategies.

Using a large, nationwide dataset and source-specific exposure modeling, this study provides robust evidence that short-term exposure to wildfire-related PM2.5 increases all-cause, cardiovascular, and particularly respiratory mortality in Brazil, with greater susceptibility among older adults and females and notable geographic heterogeneity. It is, to the authors’ knowledge, the first and largest study to characterize cardiovascular mortality risks from wildfire-related PM2.5 in Brazil. The results can inform public health preparedness and air quality management in regions with similar demographics and healthcare contexts. Future research should assess joint effects with co-pollutants (e.g., ozone, gases), evaluate long-term and repeated exposure impacts, refine exposure assessment to reduce measurement error (e.g., using higher-resolution or personal exposure data), and distinguish wildfire from deliberately set fires to better apportion sources and guide interventions. Further investigation into determinants of geographic and sex-specific susceptibility is also warranted.

Key limitations include: (1) focus on wildfire-related PM2.5 and short-term (acute) effects only; potential synergistic effects with co-pollutants (e.g., ozone, precursor gases) were not modeled; (2) exposure assessment relied on modeled, region-averaged wildfire-related PM2.5 (GEOS-Chem with calibration and downscaling), which may introduce exposure measurement error likely biasing effects toward the null; and (3) inability to distinguish wildfires from deliberately lit fires, limiting source-specific inference.