Driver-specific deforestation as a modifier of wildfire PM_2.5-mortality associations in Brazil

Wildfire smoke is an increasingly important environmental health risk, and growing evidence links wildfire-attributable fine particulate matter (PM_2.5) to acute mortality. In Brazil, wildfires frequently co-occur with land-use change, yet it remains unclear whether the short-term mortality impacts of wildfire PM_2.5 vary according to the surrounding deforestation context and its underlying drivers. We conducted a two-stage, multi-location time-series study to estimate the short-term association between daily wildfire-attributable PM_2.5 and all-cause mortality in Brazil from 2003 to 2023, and to evaluate contextual effect modification by deforestation intensity. In the first stage, we fit quasi-Poisson regression models with distributed lag non-linear models (DLNMs) adjusting for long-term/seasonal trends, day of week, and moving averages of temperature and relative humidity. In the second stage, we pooled region-specific lag and cumulative estimates using multivariate meta-regression, entering driver-specific contextual deforestation intensity as an effect modifier. Across Brazil, higher daily wildfire-attributable PM_2.5 was associated with increased all-cause mortality, with effects distributed over lags up to 14 days. Effect modification varied by deforestation driver. At average contextual deforestation (z = 0), pooled cumulative RRs over lags 0-14 days were consistently elevated (≈1.015-1.017 per 10 μg/m³, depending on driver). Higher contextual deforestation was associated with stronger cumulative associations for several drivers, including renewable energy projects (low: 1.007 [0.998-1.016]; high: 1.027 [1.016-1.038]), reservoirs/dams (low: 1.012 [1.002-1.022]; high: 1.021 [1.010-1.033]), roads (low: 1.012 [1.006-1.019]; high: 1.020 [1.013-1.027]), and legal mining (low: 1.015 [1.009-1.022]; high: 1.017 [1.010-1.023]). In contrast, agriculture showed an attenuation at higher deforestation intensity (low: 1.027 [1.021-1.034]; high: 1.003 [0.996-1.009]). Lag-response patterns indicated that modification was often most evident at later lags for infrastructure- and extraction-related drivers. These findings underscore the importance of integrating air-pollution epidemiology with land-use science to identify which land-use trajectories may amplify or mitigate wildfire smoke-related health burdens.