The meta-analysis of the Italian studies on short-term effects of air pollution (MISA): old and new issues on the interpretation of the statistical evidences

The second meta-analysis of the Italian studies on short-term health effects of air pollution, known as MISA-2, was based on daily time series of indicators of both pollution and of health outcomes. It covered 15 cities during 1996–2002 for a total population of approximately nine millions. Health outcomes included mortality for natural causes, for respiratory diseases and for cardiovascular conditions, as well as hospital admissions for respiratory, cardiac and cerebrovascular diseases. Pollutants considered in univariate analyses were sulphur dioxide (SO2), nitrogen oxide (NO2), carbon monoxide (CO), suspended particulate matter (SPM) measuring less than 10 mm diameter (PM10) and ozone (O3, limited to the summer period). Results, including risk estimates, have been largely confirmatory of findings obtained in previous large meta-analytic studies carried out in North America and in Europe. A full report in Italian is available. The present contribution summarises the results of MISA-2 and addresses three major issues regarding their interpretation: robustness of the causal inferential process, the role of specific air pollutants and the reliability of risk estimates. The former issue is stressed according to Bradford Hill’s criteria and the conclusion is reached that at least for the association of air pollution with an increase in mortality the evidence for causality is strong. Assessing the role of each air pollutant is problematic: there is some evidence that the effects of PM10 are partly confounded by other pollutants, but PM10 may not be the best indicator of the role of air SPM (routine measures of PM2.5 have not been introduced in Italy). As for risk estimates, the per cent increase in risk of mortality for unit increase in PM10 concentration, measured in MISA-2, is remarkably similar to estimates in other studies and there is indication for linearity of the dose–response relationship.