[Meta-analysis of the Italian studies on short-term effects of air pollution]

A Biggeri, P Bellini, B Terracini
Epidemiologia e Prevenzione 2001, 25 (2 Suppl): 1-71

BACKGROUND: In recent years, much attention has been given to review reports on the early effects of air pollution on health, measured through daily series of deaths and/or hospital admissions. A number of large planned meta-analyses (in which methods for data retrieval and processing are commonly planned a priori for all participating centers) are on going both in the US and in Europe. The National Mortality, Morbidity and Air Pollution Study included data from 90 US cities, whereas APHEA (Air Pollution and Health, a European Approach) considers data from about 30 european cities. The present paper summarizes methods and findings of MISA, a meta-analysis of data from 8 Italian cities. It belongs to an ad hoc supplement of Epidemiologia & Prevenzione (Epidemiol Prev 2001; 25 (2) Suppl: 1-72), the official Journal of the Italian Association of Epidemiology, which contains a full description of the study. MISA was launched on March 2000, within the project "Statistics, Environment and Health" (GRASPA), funded by the Italian Ministry of Education. Additional support was given by the Authorities of the 8 participating cities (from North to South: Turin, Milan, Verona, Ravenna, Bologna, Florence, Rome and Palermo). DAILY HEALTH DATA: Deaths certificate and hospital admission data have been collected respectively from the Local Health Authority and regional files. The same programme for retrieval of data on selected hospital admissions for acute conditions was used in the 8 cities. Main data are summarized in Table 1. DAILY CONCENTRATION OF POLLUTANTS: Most data were obtained from Regional Environmental Protection Agencies, which are responsible for environmental monitoring since 1993. Verona, Palermo and Milan (1990-94) data were obtained from local sources. Monitors with more than 25% of missing data were excluded. Meteorological data were collected by the same monitors and completed with data from monitors situated in the suburbs or (in Milan and Bologna) in the airport. The monitors were selected by a group of experts to ensure comparability. For SO2 and NO2 daily averages of hourly measurements were used, whereas concentrations of ozone and CO were estimated as the maximum 8 hours moving average. Total suspended particulate or PM10 were measured as 24 hours deposition. All analyses used the whole range of observed values (Table 2). Daily data were considered as missing when more than 25% of hourly data were not available. Missing data in one monitor were imputed as average of data from the remaining monitors weighted by the ratio between the specific monitor's year average and the general year average of all the selected city monitors. Missing data in one day were imputed as average of four days (preceding and following day, the same day of the previous and following weeks). In the city of Florence and Palermo PM10 concentrations were available. For the other cities we applied a conversion factor from PTS to PM10 (0.6 for Turin and 0.8 for all the others) estimated through validation studies. Ozone concentrations were used only where background monitors were available (Turin, Verona, Bologna and Florence) and limited to the warm season (May through September).

METHODS: A common protocol for the city-specific analyses was defined on the basis of a structured exploratory analysis. The adopted basic model was a Generalized Additive Model for Poisson data. Effect estimates were age-adjusted (0-64, 65-74, 75+) and formal tests of interaction pollutant-age were conducted. In the first two age groups, indicator variables for seasonality were specified, and cubic splines with fixed number of degree of freedom were specified for the last age group and for all age groups for the morbidity data. Model adequacy was checked by residual analysis and inspection of the partial autocorrelation function. In a sensitivity analysis non linear pollutant effects were considered and overdispersed [table: see text] transitional models were fitted; the analysis was conducted for all lags 0-3 and some distributed lags (0-1, 1-2, 0-3); no multipollutant models were fitted. The same model was fitted to the city data. No model selection was done: Table 3 describes the steps in model building. In the meta-analysis, for each outcome, the estimates for each pollutant and for each city were combined using fixed and random effects models. Heterogeneity of effects was tested according to DerSimonian and Laird. Results were checked using a hierarchical bayesian model, which was used to investigate heterogeneity across cities in a meta-regression phase. Non informative priors were used. Posterior distributions of parameters of interest have been obtained with WinBUGS. 10,000 iterations (excluding [table: see text] the first 2000) were retained, while for the meta-regression 100,000 iterations (excluding the first 4000) were stored. To approximate the marginal posteriors only one sample out of five were used. Achieved convergence was assessed using the Gelman and Rubin approach. In the meta-regression the models specified were the following: [formula: see text] i denotes city, j calendar period (1990-1994; 1995-1999). The first model includes only period as effect modifier, while the second model other potential variables. The ui terms (which do not vary with j) represent city specific random effects.

RESULTS: For each pollutant, the meta-analysis detected a statistically significant association with mortality for natural causes. But for ozone, positive associations were commonly found for death and hospital admissions for both cardiovascular and respiratory diseases. Indeed, the only estimates whose lower 95% confidence limit bore a negative sign regarded the association between PM10 and mortality from respiratory diseases. Ozone in the warm season was positively and significantly associated with daily mortality and mortality for cardiovascular diseases whereas other estimates did not reach statistical significance and some were negative (only lag 0-1 for external comparability are reported in Table 4). Risks were highest (up to 4%) for respiratory conditions (Table 4). They were more pronounced at lag 1-2 for mortality, and at lag 0-3 for hospital admissions. Age was an effect modifier for mortality, the elderly being more susceptible. In the random effect meta-analysis, at lag 1-2, excess risks for unit increase of the pollutants at age 75+ and at age 0-64 were respectively: 4.9% and -0.4% for SO2, 1.7% and 0.6% for NO2; 2.3% and 0.2% for CO. Corresponding figures for PM10 at lag 0-1 were 1.1% and 0.2%. The effect of PM10 on mortality [table: see text] was greater during the warm season (2.8% vs 0.8%). A complete analysis is reported in the Italian text. Here we provide some details on the effects of PM10, about which the residual heterogeneity across cities was highest (Table 4). In addition, the epidemiological evidence on the hazards from this fraction of particulate matter is more controversial. Table 5 reports the excess risk estimated through the meta-analysis in 1995-99 for a 10 micrograms/m3 increase of PM10 for some outcomes. Proper prior distributions (overdispersed normal and inverse gamma) were adopted in the final bayesian analyses. The sensitivity of results to the choice of the priors were investigated (we defined proper and improper uniform, student's t), obtaining comparable results. Total natural mortality was significantly heterogeneous across cities (Q = 18.96, 5 df, p < 0.001). City-specific estimates are represented graphically in Fig. 1. As expected, the confidence (credibility) intervals are widest [table: see text] for bayesian estimates, intermediate for those obtained under a random effects model, and narrowest for those found under a fixed effects model. Nevertheless, differences in point estimates are negligible. A North-South gradient in risk is obvious. Table 6 shows, for the cities for which mortality data were available, the improvement in precision and the shrinkage of effect estimates toward the overall mean introduced by the bayesian modelling. In the meta-regression, total mortality and a deprivation score were associated with greater effects. The excess risks on hospital admission were modified by the deprivation score and by the NO2/PM10 ratio. Overall, the risk estimates were greater in the calendar period 1995-99 and there was a North-South gradient, with larger effects in cities located in Central and Southern Italy (Florence, Rome, Palermo).

CONCLUSIONS: The meta-analysis of the Italian studies on short-term effects of air pollution in 8 cities, MISA, exhibits the following features: With the exception of Naples, all greatest Italian cities were included; overall a population of 7 million was enrolled. The study protocol was accurate with regard to the selection of hospital admissions for acute conditions. Monitored data of concentration of pollutant were carefully evaluated before their inclusion in the meta-analysis. City specific analyses were carried out according to a common protocol controlling for seasonality, influenza epidemics, age and meterological variables; [table: see text] the protocol derived from a structured exploratory analysis. The meta-analysis was done using fixed and random effects models; a hierarchical bayesian model was fitted in a sensitivity analysis. The heterogeneity of effects across cities was investigated using a hierarchical bayesian model for meta-regression. While mortality data are of good quality, hospital admission data are more problematic. Since the filing criteria for the latter changed around 1995, comparability of results before and after such date is limited. Moreover, hospital admissions rely on availability of beds, the offer of which may be restricted during the warm season. Comparability of pollutant concentration estimates among cities may have been influenced by differences in monitor characteristics. (ABSTRACT TRUNCATED)

Full Text Links

Find Full Text Links for this Article


You are not logged in. Sign Up or Log In to join the discussion.

Related Papers

Remove bar
Read by QxMD icon Read

Save your favorite articles in one place with a free QxMD account.


Search Tips

Use Boolean operators: AND/OR

diabetic AND foot
diabetes OR diabetic

Exclude a word using the 'minus' sign

Virchow -triad

Use Parentheses

water AND (cup OR glass)

Add an asterisk (*) at end of a word to include word stems

Neuro* will search for Neurology, Neuroscientist, Neurological, and so on

Use quotes to search for an exact phrase

"primary prevention of cancer"
(heart or cardiac or cardio*) AND arrest -"American Heart Association"