Share this article on:

Statistical Modeling of Short-Term Effects of Meteorologic Variables on Mortality

Biggeri, Annibale*; Baccini, Michela*; Michelozzi, Paola; Accetta, Gabriele*; Katsouyanni, Klea; Analitis, Antonis; Anderson, Hugh Ross§; Forsberg, Bertil; Medina, Sylvia; Paldy, Anna**

ISEE/ISEA 2006 Conference Abstracts Supplement: Symposium Abstracts: Abstracts

*Department of Statistics, University of Florence, Florence, Italy; †Department of Epidemiology Local Health Authority ASL RM/E, Rome, Italy; ‡Department of Hygiene and Epidemiology, University of Athens Medical School, Athens, Greece; §Department of Community Health Sciences, St. Georges, University of London, U.K.; ¶Department of Public Health & Clinical Medicine, Umea University, Umea, Sweden; ∥Environmental Health Department, Institute de Veille Sanitaire, Saint Maurice, France; and **Department of Biologic Monitoring, National Institute of Environmental Health, Budapest, Hungary


Back to Top | Article Outline


Epidemiologic studies indicate that exposure to extreme meteorologic conditions is associated with an increase in mortality. We evaluated the health effects of apparent temperature during warm season (April–September) in 15 European cities participating to the PHEWE project. The aim of this work was to investigate the relationship between exposure and mortality and provide city-specific summary measures of the effect of high apparent temperatures to make straightforward comparison among cities.

All cities provided daily counts of deaths for all causes, cardiovascular and respiratory causes, and 3-hourly meteorologic data retrieved from the nearest airport weather station. Data on several confounders, including other meteorologic variables and air pollution variables, were also considered. A large variability in climatic characteristics and in mortality was observed in the cities involved.

The city-specific analyses were based on a GEE approach (Liang and Zeger, 1986). Daily mortality data were modeled by a marginal Poisson model assuming independence among separate summers and AR (1) correlation structure within summer, according to the results of an exploratory analysis based on dynamic models (Chiogna and Gaetan, 2005). Appropriate parametric terms were included in the model to adjust for time trend, seasonality, day of the week, holiday, air pollution concentration, wind speed, and sea level pressure.

First, the exposure–response curve was modeled by using a parametric regression spline. Then, we modeled the apparent temperature effect by a “V”-shaped function and used the estimated slope above the minimum as an indicator of high apparent temperature effect.

The city-specific thresholds, obtained by maximum likelihood approach (Muggeo, 2003), and the slopes were combined using Bayesian meta-analysis techniques.

Distributed lag models were specified to investigate the delayed effect of high apparent temperature and time-varying coefficients models were used to check sensitivity of results to different definition of warm season.

City-specific and pooled exposure–response curves show a clear “V” or “J” shape. The effects of high apparent temperature were very consistent among cities, and it appeared stronger in Mediterranean cities than in North Atlantic and continental cities. Threshold values appeared to vary among cities, ranging approximately from 20°C (north European cities) to 30° (Athens, Rome, and Milan). An excess of risk is concentrated in the first days and some evidence of harvesting was found, in particular for Mediterranean cities. The effect of high apparent temperature is quite constant and limited to the period from June to August for Mediterranean cities and from July to August for North Atlantic and Continental cities.

© 2006 Lippincott Williams & Wilkins, Inc.