Meta-analysis or pooled analysis? A comparison based on time-series used for the analysis of short-term effects of air pollution on human health

OBJECTIVES: to compare the meta-analysis and the pooled analysis approach to study short-term effects of air pollution on human health in Emilia-Romagna Region (Central Italy) cities, characterised by strong homogeneity of environmental and sociodemographic features. METHODS: application of fixed-effects meta-analysis and fixed-effects pooled analysis on time-series data of seven cities in Emilia-Romagna in the period 2006-2010. The relationship among adverse health events (deaths due to natural causes, cardiovascular disease, cerebrovascular disease and respiratory disease) and concentrations of PM10, PM2.5 and NO2 was investigated by means of GAM models, using the EpiAir protocol. RESULTS: the pooled analysis application entailed a gain in terms of precision of effect estimates in respect to meta-analysis approach. The interval widths of pooled analysis are lower than those of meta-analytic estimates, with percentage reductions between 7% and 43%. This power increase led to a major number of statistically significant pooled analysis estimates. It has been a generally good correspondence between the two methods in terms of direction and strength of the association among health outcomes and the various pollutants. An exception is the PM10 effect estimate on respiratory mortality, where the meta-analytic estimate was significantly higher and not in line with literature data. CONCLUSIONS: the study highlighted the increase in accuracy and stability of effect estimates obtained from a pooled analysis compared to a meta-analysis in a regional context such as the Emilia-Romagna Region, characterised by the absence of heterogeneity in exposure to pollutants and other confounders. In this context, the pooled approach is to be considered preferable to meta-analysis.