Short-term exposure to fine particulate pollution and elderly mortality in Chile

The study investigates the short-term impact of fine particulate matter (PM2.5) on all-cause mortality among older adults (aged 75+) in Chile. Motivated by higher air pollution levels in Chile relative to other OECD countries and gaps in evidence outside developed regions—especially in rural areas—the authors leverage satellite-derived PM2.5 with complete national coverage. The research question is: what is the effect on monthly elderly mortality of within-commune, short-term (monthly) deviations in PM2.5 from expected seasonal levels? The purpose is to generate causal estimates using fixed-effects models that exploit temporal variation within communes, control for seasonal and national trends, and adjust for temperature. The importance lies in informing public health and policy in a country with substantial PM2.5 burdens, diverse sources (traffic, industry, wood burning), and rapidly aging populations, and in extending the literature by examining heterogeneity across geography, urbanicity, baseline exposure, elderly population shares, income, and time.

Prior work has established links between PM2.5 and adverse health outcomes (cardiovascular, respiratory, and all-cause mortality), with much of the evidence from developed countries at lower pollution levels. Remote sensing advances have enabled broader spatial and temporal coverage, facilitating studies in Africa, Indonesia, Brazil, and Latin American cities. In Chile, past short-term studies relied on ground monitors and focused on Santiago or select cities, limiting geographic representativeness; even satellite-based Chilean analyses have largely been city-focused. Evidence from rural areas remains scarce globally and particularly in Latin America. Additionally, literature often emphasizes heterogeneity by age and sex, with less attention to urban-rural differences, baseline exposure levels, or demographic composition such as the share of elderly. This study addresses these gaps using nationwide, commune-level data over 18 years and examines heterogeneity along multiple dimensions not fully explored previously.

Study area: Continental Chile, spanning diverse climatic zones (tropical north, Mediterranean center, Antarctic south). Analytical unit: commune (third-level administrative division). The panel covers 327 communes over 216 months (January 2002–December 2019), excluding 19 communes outside the continental area or with average 75+ population below 50. Data sources: - PM2.5: Satellite-based monthly estimates (0.01° ~1.11 km resolution) from the Atmospheric Composition Analysis Group (2000–2021), derived from aerosol optical depth with a chemical transport model and calibrated to ground observations. Validation used 3,858 monthly observations from 59 SINCA ground stations (2015–2021), yielding a Pearson correlation of 0.78 and RMSE of 12.76; satellite estimates overestimate at low concentrations (<12 µg/m3) and underestimate at high (>50 µg/m3). Regions XV, III, XI, XII showed poorer correspondence and were excluded in robustness checks. - Temperature: MODIS Terra Land Surface Temperature/Emissivity Daily (MOD11A1 v6.1) at 1 km resolution, aggregated to monthly means. Chosen over 2-m ambient temperature due to finer spatial resolution and high correlation (~0.9) with near-surface air temperature. - Population: Yearly commune-level estimates (2002–2019) disaggregated by single-year age and sex (INE), derived from the 2002 and 2017 censuses via demographic interpolation. 2017 Census enumeration-area (district zone/rural polygon) populations by five-year age groups and sex used for exposure weighting. - Mortality: Individual death certificates (1990–2020) from the Chilean Department of Statistics and Health Information (DEIS), including commune of residence, sex, age, date, and ICD-10 cause of death. Exposure estimation: - For each month, satellite PM2.5 and land temperature rasters were spatially intersected with enumeration-area polygons. Each polygon’s exposure was the area-weighted average of overlapping raster pixels. Commune-month exposures were computed as population-weighted averages across constituent polygons using 2017 enumeration-area populations as weights, yielding population-weighted PM2.5 and temperature exposures. Outcome construction: - Commune-month mortality rates for the 75+ population were computed as deaths in the month divided by the total 75+ population from yearly estimates. Cause categories: all-cause; cardiorespiratory; cardiovascular (ICD-10 I00–I99); respiratory (J00–J99); and all-cause excluding cardiorespiratory. Empirical strategy: - Primary model: generalized linear model with two-way fixed effects and a negative binomial outcome to accommodate over-dispersion and zero inflation. Canonical log link. Model: g(E[Y_imgt]) = β PM_imgt + γ T_imgt + α_i + θ_t where Y_imgt is the 75+ mortality rate in commune i, month m, quarter q, year t; PM and T are monthly commune-level population-weighted PM2.5 and temperature exposures. α_i are commune fixed effects; θ_t are quarter-year fixed effects (e.g., 2010-Q1). Standard errors clustered at the commune level. β and γ represent short-term effects. - Rationale: Commune FE remove time-invariant confounders (healthcare access, socioeconomic factors, baseline health, geography). Quarter-year FE flexibly capture national seasonal/time trends and potential time-varying confounders (e.g., hospital saturation, behaviors, other pollutants). Temperature included as a key confounder (linear in main model; polynomial/spline in robustness). Heterogeneity and robustness: - Separate models by region (16 first-level divisions), Metropolitan Region vs. rest, urban (≤70% rural population) vs. rural (>30% rural share), baseline PM2.5 (mean above/below 20 µg/m3), and 75+ population share (above/below median 4.5%). Additional stratification by income quintiles (based on 2017 per-capita income) and by time blocks (2002–2005, 2006–2010, 2011–2015, 2016–2019). - Sensitivity analyses included alternative outcome models (Poisson, OLS), alternative time fixed effects (month-year FE and variants), region-varying FE, flexible temperature terms (polynomial, spline), exclusion of regions with low satellite–monitor agreement, removal of small-75+ communes, sex-specific analyses, age 65+ outcome, and non-linear PM2.5 functional forms. - Code availability: https://github.com/pmbusch/PM25-Satellite-Chile

- Exposure distribution: Commune-level monthly PM2.5 exposures ranged 3–90 µg/m3 (mean 22.2). Strong seasonality with winter peaks; southern Chile exhibits higher PM2.5 and lower temperatures. Nearly all commune-months exceed WHO annual guideline (5 µg/m3); many exceed Chile’s annual standard (20 µg/m3). - Mortality baseline: Average national 75+ population 691,913 over study period; 862,818 deaths (≈3,995/month). Average monthly 75+ mortality rate 5.77 per 1,000; 7.7% of commune-months had zero deaths. - Main effect: A 10 µg/m3 increase in monthly PM2.5 is associated with a 1.7% increase in monthly all-cause mortality for ages 75+ (95% CI: 1.1–2.4%). - Geographic heterogeneity: Eight center-south regions show significant positive effects; average effect in center-south ≈4.7%. Metropolitan Region (Region M) effect: 5.3% (95% CI: 4.2–6.5%) per 10 µg/m3. Some regions’ estimates not significantly different from zero; Region XII shows a significant negative estimate but has poor satellite–ground correspondence. - Urbanicity and baseline exposure: No significant modification by urban vs. rural status (urban 1.8% [0.8–2.8]; rural 1.7% [1.0–2.5]) or by baseline mean PM2.5 above/below 20 µg/m3 (below 20: 1.6% [0.3–3.0]; above 20: 2.0% [1.1–2.9]). - Demographics: Communes with higher 75+ population shares (>4.5%) have larger effects: 2.3% (1.6–3.0) vs. 1.4% (0.4–2.4) below median (difference not statistically significant). Suggests demographic structure may partially explain larger effects in center-south. - Time trends: Effects estimated over 2002–2005, 2006–2010, 2011–2015, 2016–2019 show no clear trend, suggesting limited evidence of adaptation within the study window. - Income stratification: Largest effect in the poorest quintile; effects rise from the 2nd to 4th quintile then decline in the highest quintile; no monotonic trend or significant differences between income quintiles overall. - Cause-specific mortality: Respiratory causes show the largest effect: 2.1% (95% CI: 0.6–3.7). Cardiovascular: 1.3% (0.4–2.1). Cardiorespiratory: 1.6% (0.7–2.5). All-cause excluding cardiorespiratory: 1.9% (1.2–2.6). Urban areas have slightly larger effects for respiratory deaths (2.6% [0.4–4.8]) than rural (1.2% [−0.9 to 3.3]). - Sensitivity: Results robust to excluding regions with poor satellite accuracy, excluding small-75+ communes, and sex-specific analyses (male 1.7% [0.9–2.5]; female 1.8% [0.9–2.6]). Alternative model specifications generally yield positive effects; month-year FE absorb most PM2.5 variation, reducing statistical significance. Using more flexible temperature terms yields similar estimates. Omitting temperature increases the estimated PM2.5 effect to ≈6%, consistent with temperature confounding (warmer temperatures lowering PM2.5 and mortality at lower temperature ranges). Age 65+ results are similar. - Public health impact (counterfactuals): If all communes met Chile’s annual PM2.5 standard (20 µg/m3), an average of 27 avoided 75+ deaths per month (95% CI: 16–38). Meeting WHO guideline (5 µg/m3) implies 126 avoided deaths per month (95% CI: 76–176); over 2002–2019 this totals >25,000 avoided deaths (≈3.2% of all 75+ deaths). In the Metropolitan Region alone, meeting WHO guideline could avoid 178 deaths per month (95% CI: 140–217), ~11.6% of 75+ deaths.

The study provides causal evidence that short-term, within-commune monthly increases in PM2.5 elevate elderly (75+) all-cause mortality in Chile. By exploiting commune and quarter-year fixed effects and adjusting for temperature, the analysis isolates the short-term component of exposure, complementing literature that focuses on daily acute and long-term chronic effects. The magnitude of the monthly effect (1.7% per 10 µg/m3) aligns with or exceeds daily acute estimates reported elsewhere, while remaining below chronic effects, consistent with expectations. Findings are strikingly consistent across baseline exposure levels, urbanicity, sex, age threshold (65+ vs 75+), income, and time, suggesting a broad-based mortality response to PM2.5 in this context. The principal heterogeneity is geographic: center-south regions and the Metropolitan Region exhibit larger effects (~4.7–5.3%), which are not explained by urban–rural differences or baseline PM2.5 levels. Potential explanations include differences in PM2.5 sources and chemical composition (traffic/industrial emissions in center; residential wood burning in south), demographic structure (higher elderly shares), and behavioral factors (greater indoor time in colder regions, potentially increasing exposure to indoor sources). The slightly larger respiratory mortality effect in urban areas may reflect differences in pollutant composition and population density affecting respiratory infection transmission. The analysis underscores the substantial potential health benefits of lowering PM2.5, with sizeable avoided deaths under scenarios meeting national or WHO standards. The approach demonstrates the value of satellite-based exposure estimates for countries lacking dense ground monitoring networks and offers a scalable template for similar analyses elsewhere.

This study demonstrates that short-term monthly increases in PM2.5 significantly raise all-cause mortality among older adults (75+) in Chile, with an overall 1.7% increase per 10 µg/m3 and larger effects in center-south and the Metropolitan Region. Results are robust across numerous specifications and largely consistent across key subgroups, with limited evidence of modification by urbanicity, baseline exposure, income, or time. The geographic concentration of larger effects highlights priority regions for intervention. Policy implications include the urgent need to reduce PM2.5 exposures—particularly during seasonal peaks and wildfire episodes—through measures targeting major sources (traffic, industry, residential wood burning) and improved wildfire management. The satellite-based, population-weighted exposure framework enables comprehensive, high-resolution assessments where ground monitoring is sparse. Future research should quantify the role of PM2.5 chemical composition and sources, incorporate indoor exposure measurements, refine temporal alignment between exposure and mortality at sub-monthly scales, and further investigate how demographic shifts (increasing elderly shares) and potential adaptations over longer horizons influence vulnerability.

- Satellite PM2.5 accuracy is reduced at low (<12 µg/m3) and high (>50 µg/m3) concentrations, tending to overestimate at low and underestimate at high levels; this may bias effect sizes upward if true variability exceeds satellite estimates. Excluding regions with poor satellite–monitor correspondence (XV, III, XI, XII) did not materially change results. - Both satellite and ground monitors measure outdoor PM2.5, not indoor exposures. Elderly populations in colder southern regions may spend more time indoors with elevated indoor PM2.5 (e.g., wood combustion), leading to exposure misclassification. - Lack of data on PM2.5 sources and chemical composition limits inference on mechanistic pathways and source-specific health impacts. - Death certificates record commune of residence at death, which may not match the location of prior exposure. While elderly mobility is generally low in Chile, residual misclassification is possible. - Monthly aggregation introduces temporal mismatch between exposure peaks and deaths, potentially complicating causal attribution and masking immediate daily responses. - Alternative model specifications with stricter time fixed effects (month-year) reduce available variation in PM2.5, lowering precision; however, effect signs remain positive.