Detectability of short-term associations between ambient air pollution and cardiovascular morbidity: An ecological time-series analysis under heterogeneous atmospheric conditions
Introduction: Short-term associations between ambient air pollution and cardiovascular morbidity remain inconsistent across different settings. This study evaluated these associations using Generalized Additive Models (GAMs) and examined the robustness of the findings through lag and sensitivity analyses. Materials and methods: An ecological time-series analysis was conducted using a publicly available environmental–health dataset comprising 5,811 daily observations of air pollutant concentrations, meteorological variables, and aggregated cardiovascular morbidity indicators. GAMs with penalized smoothing splines were fitted to assess potentially non-linear exposure– response relationships while adjusting for meteorological covariates. Lag structures (Lag 0–3 days), alternative spline dimensions, and multi-pollutant models were evaluated as sensitivity analyses. Results: The final GAM demonstrated limited explanatory capacity (AIC=25,602.9; AICc=25,604.6; explained deviance=1.04%). Pollutant smooth terms showed shallow non-linear exposure–response patterns (effective degrees of freedom: 8.36-9.40), but none were statistically significant (PM₂.₅: p=0.357; PM₁₀: p=0.950; NO₂: p=0.594; SO₂: p=0.339; O₃: p=0.530). Sensitivity analyses produced consistent results across alternative lag structures and spline dimensions, and model diagnostics indicated satisfactory fit without evidence of substantial over dispersion. Conclusion: Only shallow, statistically non-significant short-term associations were identified. These findings should be interpreted within the limitations of the ecological design, potential exposure misclassification, and residual confounding.