ANALYSIS AND META-ANALYSIS OF EPIDEMIOLOGICAL TIME SERIES DATA WITH APPLICATIONS IN AIR POLLUTION EPIDEMIOLOGY

ANALYSIS AND META-ANALYSIS OF EPIDEMIOLOGICAL TIME SERIES DATA WITH APPLICATIONS IN AIR POLLUTION EPIDEMIOLOGY PRACTICAL TRAINING COURSE The Contract included two Practical Training Courses (PTC). The first focused on statistical analysis methods of epidemiological time series, whilst the second on meta-analysis of observational studies. The two courses corresponded to the two stages of data analysis in the multi-centre, E.C. sponsored, APHEA project (Air Pollution and Health: a European Approach) which studies the short-term effects of air pollutants on health based on data from 30 European cities. In the first stage the individual city data sets, which consist of time series, are analyzed and in the second stage, the results from the individual city analyses are combined using meta-analytic methods adapted to epidemiological time-series. Both PTC took place within a week and combined theoretical presentations with practical applications using a computer lab and appropriate statistical software. The lectures were given by internationally known scientists working in Biostatistics and epidemiology. Practical sessions were organized in small groups and prepared and supervised by invited scientists. Lectures and students from about 20 countries in the European Union, Central-Eastern Europe, the U.S., Hong-Kong, Australia, Israel and Turkey participated. The first event was attended by 30 students and the second by 47. The place for both PTCs was Santorini Island in Greece. In October and November, when the PTCs took place, there are practically no visitors in Santorini, but the weather tends to be nice. The Conference facilities are good and the place offers an excellent setting for a quite and pleasant stay, helping the exchange between scientists from different places and cultures. 1

The Scientific Reports for the 2 events, which are attached, provide more details on the content and methodology of the PTCs. EVENT No 1. Introduction and General Objectives Time-series analysis has been developed mainly in econometrics and has been used widely to describe and predict the behavior of time-series. In epidemiology, the objectives of analyzing time-series data are more oriented towards explaining and quantifying effects of exposures on health outcomes, one or / both of which may be in time series form (i.e. daily number of deaths or daily ambient temperature). Sophisticated time series analysis developed in the framework of other disciplines, has to be appropriately adapted for use in epidemiological timeseries. The number of applications of such methodology in environmental epidemiology has recently greatly increased, and so has the interest of the scientific community and policy makers in environmental health problems. A major field of applications has been the investigation of short-term effects of air pollution. In the framework of the APHEA2 project (Air Pollution and Health: a European Approach-sponsored by the Environment & Climate 1994-1998 Programmecontract no ENV4-CT97-0534) 22 research groups from 19 different European countries are collecting and contributing time series data, but according to the Work Programme only four groups are involved in the analysis. The Practical Training Programme organised under the present contract had as primary objective to disseminate state-of-art knowledge and skills for the analysis of epidemiological time-series. Background Time-series analysis is concerned with data, which appear in time-ordered sequences of measurements of some phenomenon of interest. The methodology has been developed either in the framework of econometrics (emphasising forecasting) or in physical sciences (with applications in signal processing). In epidemiology there has not been traditionally much interest in the applications of time-series analysis, although variables which are in time-series form are of interest. In recent years (i.e. during the last decade), in the context of air pollution epidemiology, there was an increase in applications of time-series analysis, incorporating adaptations which aim mainly at the explanation and quantification of associations (and not forecasting). Several publications from the U.S. using time-series analysis gave consistent evidence that there are short-term effects of air pollution, especially of small ambient particles (those with aerodynamic diameter <10µm, PM 10, or < 2.5, PM 2.5 ) on health. These results were important from the public-health point of view: based on them the U.S. EPA was led (by court order)_ to start a procedure of revising the air quality standards for PM and WHO started a procedure of revising the 1987 Air Quality Guidelines. In Europe, in 1993, the APHEA (Air Pollution and Health: a European Approach) project started within the framework of Environmental 1991-1994 (contract number EV5V-CT92-0202). Before this project there were few publications in Europe concerning the current levels of air pollution short-term health effects, which were done with different methodologies not easily comparable. 2

The results of APHEA supported the existence of air pollution health effects, reported some heterogeneity of the effects in different regions in Europe and indicated a role for other pollutants in addition to the PM effects. These opened questions on the methodology and the study of heterogeneity of effects. Also in the U.S. a new large study was initiated, the National Mortality, Morbidity and Air Pollution Study (NMMAPS). In the framework of NMMAPS new methodology is being developed and data is collected in about 70 U.S. cities to analyze the air pollution and mortality association and a smaller number of cities to analyze air pollution and hospital admissions data. Specific objectives were: Under the above considerations, the specific objective of this first PTC 1. To establish communication and disseminate know-how between European scientists and scientific groups working in Environmental Epidemiology. 2. To enable young scientists from a wide range of European countries to analyse in a state-of the-art way, time-series data in their own country, in order to test hypothesis related to environmental epidemiology. 3. To increase the number and quality of relevant publications from Europe in high impact scientific journals (thus contributing to scientific advancement) and upgrade the European profile, in areas relevant to public health policies. Methodology Content of the PTC In terms of methodology, in epidemiological time-series it is important to describe the outcome-exposure association, adjusting for confounding effects of potential confounding factors. In the case of air pollution epidemiology exposure timeseries are daily measurements of pollutants such as carbon monoxide, sulphur dioxide, ambient particulate matter, nitrogen dioxide, ozone etc. Health outcomes are often daily numbers of total and cause-specific deaths and hospital admissions. Important potential confounders are meteorological variables (such as temperature and humidity) and temporal variables (such as seasonality and other cyclical patterns, day of the week, holidays, influenza epidemics). The outcome variables are the dependent variables in any model and, as count data, they follow a Poisson distribution. For the control of confounding, time-series methodology is necessary because the variables to be adjusted for show systematic variation in time and are autocorrelated. Several approaches exist to model these variables among which are: moving averages over specific number of days (15 to 30 in many applications) or the generalized additive model (GAM). The latter approach allows a Poisson regression to be fitted which controls for a smooth function of time using moving averages or other types of smoothing functions. Another alternative is to use regression spline functions. Another, completely different approach involves the use of sinusoidal terms to fit the long wave-length patterns, like the annual cycles, in the data. This type of analysis is less flexible in modelling trends and forces the models into specific shapes. Furthermore, issues like how much filtering and diagnostic tools used to decide 3

the adequacy of a model fit and modelling approaches for other variables (like the temperature) have also to be decided in each analysis. Structure of the PTC The course covered 5 full working days (Monday to Friday November 2-6-1998). Each day included lectures (about 4 hours) and practical training (about 4 hours). The lectures covered the following topics: Introduction to regression modelling; Introduction to S+; Air pollution studies in specific areas and their relevance; Generalized linear models; Means modelling; Splines and smoothing; Generalized additive models; Using STATA; Robust regression; Covariance models/mixed models; Time series of counts; Seasonal control; Diary data; Distributed lay models; Case-crossoner analysis. Extensive course material was distributed. For the practical training, students were divided into small working group (about 4 students) with one instructor and each group worked with two desktop or laptop computers applying in practice the contents of the morning lectures. Each small group analyzed real data sets which were contributed by the participants for training purposes. Students worked with the statistical package S+. The scientific programme of the course is attached. For the PTC, 7 Desktop computers were rented and shipped from Athens to the PTC site while participants also brought several laptop computers. Speakers/Instructors The principal lecture speaker was Dr. Joel Schwartz from the Harvard School of Public Health, Department of Environmental Sciences. Dr. Schwartz is one of the main U.S. scientists who published papers on the effects of air pollution and on related methodological issues. He is also an adviser in the APHEA project since its start. He has very long and high impact teaching experience in the same and related topics. Other speakers included Drs. A. Paldy and A. Goren who talked about problems specific to Central-Eastern and Middle East countries, Dr. A. Tobias who introduced applications with other statistical packages and Dr. Klea Katsouyanni (course organizer) who put the statistical applications in the wider context of Environmental Epidemiology. Eight instructors were each responsible for one small group during practical training. Students Thirty students attended, among which 16 from E.U. countries, 12 from Central- Eastern European countries, 1 from Israel and one from Hong-Kong. Site The PTC took place in Santorini Island in Greece. The site used belongs to the Bellonion Cultural Centre with excellent conference audio-visual facilities and space for the computers. All participants stayed in nearby hotels (photographic material is attached). Evaluation A special questionnaire was prepared for the anonymous evaluation of the course by all participants. Thirty-five questionnaires were returned completed and the average scores given are shown in the Table. 4

Question Workshop Administrative Arrangements How effectively and efficiently were the arrangements for your participation handled? Average Score 4.5(0.5) A. Please rate your hotel 4.4(0.7) B. Please rate Santorini as a Workshop destination 4.6(0.6) Workshop objectives A. Did you find the Workshop presentations useful for your research needs? B. The Workshop provided the opportunity for participants to receive current information about time-series analysis. 4.2(0.6) 4.3(0.6) C. The practical sessions were informative and useful. 3.9(1.0) D. The Workshop provided participants with the opportunity to learn practically how to handle an analysis. E. The Workshop provided participants with the opportunity to network with colleagues. 3.9(0.8) 4.0(0.7) *1: not at all, bad, inappropriate; 2-3: moderately; 4-5: exceptionally or excellent EVENT No 2. Introduction and General Objectives Attempts to review and synthesize knowledge are very old. Narrative reviews of the scientific literature are as old as scientific literature itself. However, the ever increasing amount of information published in the medical scientific press during the 20 th century made the need for reviewing more evident, necessary and demanding. Quantitative approaches to synthesize information have appeared more recently and have become powerful and influential over the past decade. The term meta-analysis (introduced by Glass in 1976) is taken from the Greek word meta literally meaning after and more metaphorically transcending. Although much attention has been given in recent years to meta-analysis in medical research, numerous conceptual and methodological issues particularly with respect to biases and the use of methods remain controversial. More controversy is surrounding its application in combining observational epidemiological studies compared to clinical trials. In the framework of the APHEA2 project (Air Pollution and Health: a European Approach; Environment and Climate 1994-1998 Programme-contract number ENV4-CT97-0534), 22 research groups from 19 different European countries collected and contributed data on exposure (air pollution), outcome (daily mortality and hospital admissions) and confounders (mainly meteorological data). The first stage of analysis involved individual city data analysis (and this topic was addressed by Event no 1 under the present contract) while the second involves the meta-analysis of all the individual city results. The present Practical Training Course (PTC) has as main objective to present, discuss and apply methods for meta-analyses with emphasis on those from observational studies. Ongoing analyses under the APHEA2 were used among other examples. 5

Background Meta-analysis may be defined as a quantitative approach for systematically assessing the results of previous research in order to arrive at conclusions about a body of research. Meta-analysis itself is an observational study, where the unit of analysis is the individual study. As published papers in scientific medical journals increase in number (in 1940 there were 2,300 biomedical journals; in 1994 there were 25,000!), so does the need for systematic reviews: from 1981 to 1988, 49 papers per year with meta-analysis could be identified in MEDLINE, whilst in 1989-93, 386 such papers per year were found. Meta-analysis of clinical trials is better established and accepted and in the early 90 s the Cochrane Collaboration was established for collecting data and producing reviews of clinical trials. Although observational studies meta-analysis developed at about the same time and evolved with a similar trend, there is more controversy attached to it. In spite of the controversy, meta-analysis of observational studies is developing and will develop in the future because it addresses two important aspects of scientific investigations: pattern recognition and study for the reasons of heterogeneity. Recently two large projects, the APHEA2 in Europe and NMMAPS in the U.S., concerning the effects of air-pollution on health incorporate meta-analysis in their protocol. APHEA2 is a continuation of APHEA, whose results supported the existence of air pollution health effects, reported some heterogeneity by region in Europe and indicated the role of particles as well as other pollutants. The APHEA2 is based on a much larger data base (30 cities) and aims to address methodological problems and explore heterogeneity. In the U.S. a similar project, the NMMAPS (National Mortality, Morbidity and Air Pollution Study) with data from the 90 largest U.S. cities has just been finalized. Specific objectives Under the above considerations, the specific objective of the second PTC were: 1. To introduce the concepts and methods of meta-analysis to young European scientists; 2.To discuss controversial aspects of meta-analysis; 3. To focus on meta-analysis of studies in air pollution epidemiology and the particular issues involved in this type of study; 4.To enable young researchers to critically apply the methods of meta-analysis; 5. To increase the number and quality of relevant publications from Europe in high impact scientific journals. Methodology Content of the PTC Although the focus and emphasis of this particular PTC was on prospectively planned meta-analysis of observational studies, it was decided for reasons of completeness and educational purposes, to introduce all aspects of meta-analysis. Thus, a historical perspective was given, basic meta-analytic methods were introduced and applied, using examples from clinical trials as well as observational studies. Subsequently, general presentations of more sophisticated and recent developments were introduced, like metaregression models which were also applied. The Bayesian approach of hierarchical modelling was presented and an overview of meta-analysis with several controversial issues about biases was discussed. Then, special issues in meta-analysis of epidemiological time-series data were discussed on the basis of several examples from the APHEA2 project. All sessions were accompanied by computer applications, for which students worked in small groups under supervision. 6

Structure of the PTC The course covered 5 and a half working days (Monday to Saturday October 23-28, 2000). Each day included lectures (about 4 hours) and practical training (about 3.5 hours). The lectures covered the following topics: Reasons for meta-analysis. Historical remarks. Basic methods for meta-analysis. Meta-regression models. Meta-epidemiology of meta-analysis. Hierarchical models for meta-analysis. Special issues in meta-analysis of air pollution epidemiological time-series data. Use of meta-regression to control confounding. Metasmoothing for estimating dose-response. Meta-analysis of distributed lag models. Mortality displacement in the short-term effects of PM. Exposure to nitrogen dioxide and mortality in APHEA2. Cause-specific mortality and exposure to particulate matter a 29 city metaanalysis. Exposure to ozone and mortality in the APHEA2 cities. Comparison of methods to combine dose-response curves. Exposure to ozone and carbon monoxide and cardiac hospital admissions in 8 European cities. Exposure to ozone and nitrogen dioxide and respiratory hospital admissions in 8 European cities. Distributed lags of weather and pollution effects on Mortality. Exposure to air pollution and hospital admissions for pneumonia. Can pollution patterns be detected in European cities? Extensive course material was distributed. For practical sessions, students worked in small groups using computers, applying the statistical package STATA. Each small group analyzed data sets and produced results. The scientific programme is attached. For the PTC, 7 desktop computers were rented and shipped from Athens to Santorini, while several participants also brought laptop computers. Speakers/Instructors Principal speakers included: Dr. Simon Thompson from the MRC Biostatistics Unit, Cambridge U.K. with well known contributions in the field of meta-analysis; Dr. Joel Schwartz from the Harvard School of Public Health, Department of Environmental Sciences who is one of the main U.S. scientists to publish on the effects of air pollution and more recently on newly developed methodology for combining results; Dr. Klea Katsouyanni (course organizer) from the Department of Hygiene and Epidemiology, the University of Athens who is the Coordinator of the APHEA project, Dr. Giota Touloumi, from the same University, involved in APHEA project from the beginning. Other speakers included Drs. M.A. Vigotti (Universita di Pisa, Italy), J. Peacock (St. George Hospital Medical School, U.K.), B. Segerstedt (Business Administration Foretagsekonomi, Umea Universitet, Sweden), S. Walker (MRC Clinical Trials Unit, U.K.), G. Rossi (Instituto di Fisiologia Clinica-CNR, Italy), P. Michelozzi (Agency for Public Health, Italy), S. Medina (InVs, Environmental Health, France), R. Anderson (St. Gerorge s Hospital Medical School, U.K.), P. Goodman (Dublin Institute of Technology Physics, Ireland), J. Sunyer (Municipal Institute of Medical Research Unit, Respiratory and Environmental Research Unit, Spain), E. Cadum (Environmental Prot. Agency of Piemonte, Epidemiology Area, Italy), A. Paldy ( National Public Health Centre, National Inst. Of Environmental Health, Hungary), B. Wotjyniak (National Institute of Hygiene, Department of Medical Statistics, Poland), B. Forsberg (Agency for Public Health, Sweden). Some topics were presented by students. Students Fourty seven students attended, among which 33 from E.U. countries, 11 from Central-Eastern European countries, one from Australia, one from Switzerland and one from Turkey. 7

Site The PTC took place in Santorini island in Greece. The site used belongs to the Bellonion Cultural Centre with excellent conference audio-visual facilities and space for the computers. All participants stayed in nearby hotels (photographic material is attached). Evaluation A special questionnaire was prepared for the anonymous evaluation of the course by all participants. Forty-one questionnaires were returned completed and the average scores given are shown in the Table. Question Average Score Workshop Administrative Arrangements F. How effectively and efficiently were the arrangements for 4.5 your participation handled? G. Please rate your hotel 4.3 H. Please rate Santorini as a Workshop destination 4.5 Workshop objectives I. Did you find the Workshop presentations useful for your 4.0 research needs? J. The Workshop provided the opportunity for participants to 4.5 receive current information about time-series analysis. K. The practical sessions were informative and useful. 3.7 L. The Workshop provided participants with the opportunity to 3.7 learn practically how to handle an analysis. M. The Workshop provided participants with the opportunity to 4.2 network with colleagues. 8

Identification: Scientist in charge: LIFmetaepid980435/EW/Jb KLEA KATSOUYANNI Address: DEPARTMENT OF HYGIENE AND EPIDEMIOLOGY MIKRAS ASIAS 75, 115 27 ATHENS, GREECE. Tel.: 30-1-7470577 / 7482467-+ Fax: 30-1-7704225 E-mail: Contractor: Contract n : kkatsoug@cc.uoa.gr University of Athens, Dept. of Hygiene & Epidemiology. ERBFMMACT980435 www site: - for events: - of contractor: www.uoa.gr - other relevant: PROGRAMME OF EVENTS Event n 1: Dates: 1-7/11/1998 Place: Santorini, Greece Event n 2: Dates: 23-28/10/2000 Place: Santorini, Greece TMR Home Page CORDIS Programme Home Pages 9