The APHEA project as a turning point in modern environmental epidemiology

Abstract

In the late 80s and early 90s, it was unclear whether the observed levels of air pollution were related to health effects, with fragmented evidence based on a few single-city studies. The Air Pollution and Health: a European Approach (APHEA)-funded projects started in 1993 and continued until 2009 and, later, within the context of the APHENA collaboration with researchers from the United States and Canada. APHEA was among the first to introduce the multicity/multicountry approach, using standardized methods and specifically developed protocols, including examining exposure–response functions and investigating nonlinearity, as well as a two-stage analytical approach, with city-specific analysis followed by a meta-analysis and meta-regression to identify effect modifiers. This approach consolidated the time-series regression as a central design for studying health effects of short-term air pollution exposures, enabled extensions to other environmental exposures, specifically climate-related, and contributed to further methodological advances. Health-effects associations, thus, became more robust and easier to compare across locations, facilitating the consensus later reached on the importance of air pollution exposure for health, as well as the funding of multicenter cohort studies. The APHEA projects produced concentration-response functions that were used in health impact assessment and policies, e.g., setting the World Health Organization guidelines and legal standards for short period exposures. This can be particularly relevant, as the health burden of chronic exposures may be enhanced by short-term peaks that trigger health events that would not have otherwise happened for a long time. Although APHEA methods were developed for air pollution health effects, they are also particularly relevant for climate-change-related studies, where short-term exposures (such as temperature or other daily indices combining meteorological variables) are extremely important. Further advantages of time-series studies, making the design important for current and future research, include their relevance for episode-based risk communication and emergency response planning; their smaller susceptibility, compared with cohort studies, to the impact of measurement error leading to underestimating health effects; and their easier implementation, making them indispensable for an initial exploration of a hypothesis.

What this study adds

In this commentary, we briefly present the history of time series studies, in the context of the seminal APHEA multicountry projects, for investigating the health effects of short-term air pollution exposures and their importance in further developments of the methodology. We emphasize their current relevance for studying other environmental exposures, such as climate-change-related, and for policymaking, referring to the particular advantages of the design, including studying unusual events, resistance to measurement error, and potential for implementation. We underline that the results from time-series studies contribute independently from cohort studies to policymaking and standard setting.

Setting the stage

Modern studies on the short-term health impacts of air pollution in Europe began in the late 1980s and early 1990s.^1–12 During the same period, several studies were also published in North America.^13–15 At that time, evidence was largely derived from single-city studies that used different data sources, outcome definitions, and analytical strategies. These early investigations were important in showing that short-term increases in air pollution could be followed by increases in daily mortality or hospital admissions, but their findings were difficult to compare or combine, raising skepticism.¹⁶ Consequently, it was often unclear whether differences between studies reflected genuine variation across cities, stochastic variation, or methodological differences in study design. In addition, there were substantial questions over the appropriate way to control for seasonality.

In response to this scientific and public health challenge, the Air Pollution and Health: a European Approach (APHEA) project, which started in 1993, was the first to introduce a multicity approach spanning several European countries. This approach and its realization demonstrated the added value of moving from a set of local investigations toward an approach capable of producing multilocation comparable evidence with direct relevance for policy.

Building a standard methodological framework, including methodological innovations

Across 15 (and subsequently, in APHEA2, 29) European cities, with a total population over 43,000,000, the APHEA project introduced the first coordinated multicenter study, at a time when such collaboration was still uncommon in environmental epidemiology research.¹⁷ Its main innovation was, in addition to the inclusion of multiple locations, the use of a common protocol for exposure assessment, health outcomes, confounder control, seasonality control, and statistical analysis, including a priori planned meta-analysis.^18,19 Moreover, methods for examining exposure–response functions and investigating nonlinearity were introduced.^20,21 This framework helped consolidate time-series regression as a central design for studying the short-term health effects of air pollution and enabled extensions to other environmental exposures, specifically climate-related.²²

A two-stage analytical methodology was established, first estimating local associations within each city and then combining the risk estimates through meta-analysis and meta-regression, which later became a standard approach in environmental epidemiology.¹⁹ Beyond its analytical contribution, the project promoted explicit methodological rules, careful documentation, and transparent reporting, helping to establish reproducibility and comparability as core standards. Further, within the collaboration, there was extensive training in statistical analysis, with all groups analyzing their own data according to the protocol, and later, taking advantage of the EC call on “Accompanying Measures,” with the organization of Workshops, with contributions and attendance by members of all research groups.

The APHEA project had a substantial and long-lasting impact on environmental epidemiology. By providing a practical framework for standardizing time-series studies of short-term health effects, it helped improve the treatment of seasonality, long-term trends, weather, and other time-varying factors, nonlinearity, and delayed (lagged effects). As a result, health-effects associations became more robust and easier to compare across locations, and this facilitated the consensus that has been reached on the importance of air pollution exposure for health, as well as the funding of later multicenter cohort studies.

The APHEA results and publications also contributed to later methodological advances and helped popularize the usage of other appropriate methods, including more flexible modeling strategies such as generalized linear models and their extension, Generalized Additive Models, with spline terms,^23,24 and, in subsequent years, the development of distributed lag nonlinear models.²⁵ It also sustained the improvement of two-stage meta-analytic frameworks,²⁶ as well as more refined summary measures such as best linear unbiased predictions.²⁷

Scientific legacy and influence on future studies

A broader scientific legacy of the APHEA project was to show that large-scale collaboration in environmental epidemiology was feasible, productive, and methodologically credible.²⁸ The project served as a prototype for later multicenter and multinational initiatives that adopted the same general principles of harmonized protocol, local estimation, and pooled synthesis. Its influence can be seen in projects such as NMMAPS in the United States;^29–31 APHENA, which explicitly integrated evidence from APHEA, NMMAPS, and Canadian studies;³² ESCALA in Latin America,³³ PAPA in Asia,³⁴ and MEDPARTICLES in Southern Europe.³⁵ More recently, this collaborative model has been extended on a global scale through the Multicountry Multicity Collaborative Research Network, starting from the study of meteorological variables (nonoptimal temperature, climate change) and subsequently covering a much broader range of environmental exposures and health outcomes, including air pollution and other environmental hazards such as wildfire-related pollution, floods, and tropical cyclones.³⁶

Long-term scientific collaborations among epidemiologists, statisticians, clinicians, and exposure scientists were also strengthened through this work, shaping not only specific analyses but also the organizational model of collaborative environmental health research. In that sense, its legacy lies both in the results it generated and in the research culture it helped create.

From epidemiological evidence to policies for decreasing air pollution

APHEA produced concentration-response functions that could be used in health impact assessment (HIA) projects. Thus, it led to the APHEIS project, which directly estimated the health impact in 19 European cities using the APHEA project as a base but also used concentration response functions (CRFs) from cohort studies.³⁷ This was followed by the APHEKOM project, which provided the information economists need to calculate the real cost of air pollution to society.^38,39

Local and national stakeholders have used this information to produce policy-ready communication essential to developing actions aimed at effectively reducing exposure to air pollution.⁴⁰ Additionally, the CRFs fed directly into the Clean Air for Europe program launched by the European Commission in 2001,⁴¹ which used HIA methods built on the APHEA framework to justify tightening emission ceilings and ambient standards across Member States. The resulting EU Air Quality Directive 2008/50/EC,⁴² which consolidated and replaced earlier instruments, translated those assessments into legally binding limit and target values for particulate matter, nitrogen dioxide, ozone, and other pollutants—standards that remained the basis of European regulation for over a decade.

At the global level, evidence from APHEA contributed to both the World Health Organization Air Quality Guidelines of 2005 and the landmark 2021 update.^43,44 The 2021 revision was the first major revision in 16 years and set substantially more stringent guideline values on the basis of epidemiological studies, including those built on the APHEA two-stage framework. The 2021 revision of the Directive,⁴⁴ aligned more closely with updated scientific evidence, further tightened the annual PM_2.5 standard, again drawing on a body of epidemiological work addressing both the effects of long-term exposures, which account for more burden of disease at a population level, and studies on the effects of short-term exposures to set shorter time period guideline concentrations.

In this sense, the policy legacy of APHEA is concrete and traceable: it contributed to a chain of regulatory decisions that reduced ambient pollution levels affecting hundreds of millions of people across Europe and beyond.

Short-term and long-term evidence are complementary pillars

The APHEA project, and the time-series tradition it consolidated, was designed to answer a specific and well-defined question: Do day-to-day fluctuations in ambient air pollution trigger short-term increases in mortality and hospital admissions (and other health outcomes)? This design is well suited to detecting acute effects and providing time signals relevant to guideline levels for short time periods (e.g., 24 hours), episode management, and health surveillance. It is, however, not designed to estimate the health consequences of chronic exposure sustained over years. That question belongs to a different and complementary study design, the long-term cohort study, which has provided some of the most influential evidence in environmental epidemiology, from the Harvard Six Cities study⁴⁵ and the American Cancer Society cohort⁴⁶ to the European ESCAPE and ELAPSE projects.^47,48 The two approaches capture different biological phenomena. Short-term studies primarily reflect acute physiological responses (inflammatory cascades, autonomic dysregulation, and exacerbation of preexisting conditions) that manifest within hours to days. Long-term studies capture the structural consequences of sustained exposure: e.g., accelerated atherosclerosis, reduced lung function growth, increased incidence of lung cancer, and chronic obstructive pulmonary disease. Critically, the effect sizes estimated by the two designs are not directly comparable and should not be used interchangeably in HIAs. The issue of whether results from time-series studies should be used for HIA was raised many years ago.⁴⁹ There was a discussion on whether the effects of short-term exposures are incorporated in the estimated effects of longer-term exposures, e.g., annual exposures. If this is true, then in HIA there is only a need to use the long-term health effect estimates. The assumption was that deaths brought forward by short-term exposures would happen within the year and would be reflected in HIA based on chronic exposure CRFs, although this was not explicitly explained. Additionally, of course, it was accepted that CRFs from cohort studies would lead to person-years of life lost, which could not be directly calculated using the CRFs from time-series studies. Evidence from APHEA and other studies demonstrates that short-term harvesting does not explainthe acute increases in mortality, and there are events brought forward that might not have happened for a long time.^50,51 Also, the relationship between the two remains unclear—specifically, how repeated acute exposures accumulate into cumulative damage, which individuals are most susceptible across both time horizons, and through what mechanistic pathways transient insults become irreversible disease. There is some evidence of independence between effects of short-term peaks and those of more consistently increased long-term exposures.^52,53 Addressing this gap requires study architectures that link time-series data to individual-level longitudinal follow-up, higher-resolution exposure data to reduce misclassification, and integration of mechanistic evidence to connect epidemiological associations with biological pathways.

Furthermore, the studies on short-term effects have some other important advantages: (1) They are most relevant for episode-based risk communication and emergency response planning. (2) They are less prone, compared with cohort studies, to the impact of measurement error on the health effects estimates, which generally produces bias to the null, meaning that it is more probable to detect a true association using a temporal study rather than a cohort study.⁵⁴ (3) They are easier to implement, if the relevant daily exposure concentrations are available, and explore whether a specific exposure may be associated with a health outcome, thus giving incentive for the implementation of more expensive and time-consuming cohort approaches.

Lessons for current and future challenges

Looking ahead, several specific directions will define the next generation of research in this tradition. First, moving beyond single-pollutant models toward assessing independent and joint effects within multipollutant frameworks is essential. Real-world exposure is always to a mixture, and analyses that treat pollutants one at a time cannot adequately disentangle the contributions of individual components, capture interactions between them, or support source-specific intervention. Methods from “causal” inference, including instrumental variable approaches and difference-in-differences, have been applied in this context and deserve wider adoption.⁵⁵ Furthermore, as environmental epidemiology moves to the study of more climate- and meteorology-related exposures the problem of multicollinearity becomes more prominent. Methods to address this issue have been developed, such as ridge regressions and weighted quantile regression,⁵⁶ but further work will help stabilize risk estimates in climate-change contexts. Second, the persistent challenge of exposure misclassification, which has long been known to attenuate effect estimates and distort exposure–response curves, can now be addressed more directly.^54,57 Time series studies use ambient concentration levels as surrogate measure for personal exposure to air pollution from outdoor sources, thus entailing, by definition, exposure measurement error arising from not taking personal time-activity-location information into consideration. However, using personal variables to estimate personal exposure to outdoor-generated pollutants, which cannot practically be measured, incorporates measurement error inherent in the variables used for the estimation. This issue deserves further exploration in the future. Third, and perhaps most scientifically ambitious, is the combination of time-series designs with cohort studies that follow individuals over years. The administrative cohorts, taking advantage of routinely collected data for administrative purposes and now widely used, may be used in this direction. Such linked architectures would allow researchers to examine whether acute short-term triggers in susceptible individuals accelerate long-term disease trajectories, opening a window onto mechanisms that neither design can access alone. These combined designs will be specifically appropriate to assess climate-related temperature effects for which the short-term associations are very clear but the long-term remain ambivalent. Fourth, systematic attention to vulnerability, equity, and effect modification by socioeconomic status is no longer optional. Consistent evidence across settings shows that the health effects of air pollution and other environmental stressors are not evenly distributed—they are concentrated in populations with greater baseline burden, reduced adaptive capacity, and limited access to clean environments. Research that ignores this heterogeneity produces average estimates that may understate the risks faced by the most exposed; e.g., when CRFs based predominantly on studies from high-income countries are used, the results may not be valid when applied globally and vice versa. Thus, APHEA-type studies should be further expanded using standardized state-of-the-art methods and data to cover under-represented areas to strengthen validity and generalizability. Fifth, in terms of improving the methods related to the analysis of time series data, several of the methods that have been developed and widely applied generally rely on investigator decisions that may affect the results (such as placement and number of knots, constraints on the lags). Developing standardized protocols with less investigator-based decisions may lead to improved transparency and validation, although caveats of using a one-fits-all approach must also be addressed. Finally, on the policy side, the outputs of time-series research are directly applicable to operational tools that have yet to be fully developed: adaptive warning systems that account for heat and air quality, geographically targeted interventions for pollution hotspots, areas with pronounced urban heat island effect, and otherwise vulnerable neighborhoods. Thirty years after APHEA, the foundational methods are in place; the challenge now is to implement them with greater precision, greater equity, and greater urgency.

Conflicts of interest statement

The authors declare that they have no conflicts of interest with regard to the content of this report.

ACKNOWLEDGMENTS

The Air Pollution and Health: a European Approach (APHEA), collaborative group consisted of: K. Katsouyanni, G. Touloumi, E.Samoli (Greece; Coordinating center); D. Zmirou, P. Ritter, T. Barumandzadeh, F. Balducci, G. Laham (Lyon, France); H.E. Wichmann, C. Spix (Germany); J. Sunyer, J. Castellsague, M. Saez, A. Tobias (Spain); J.P. Schouten, J.M. Vonk, A.C.M. de Graaf (Netherlands); A. Ponka (Finland); H.R. Anderson, A. Ponce de Leon, R. Atkinson, J. Bower, D. Strachan, M. Bland (UK); W. Dab, P. Quenel, S. Medina, A. Le Tertre, B. Thelot, B. Festy, Y. Le Moullec, C. Monteil (Paris, France); B. Wojtyniak, T. Piekarski (Poland); M.A. Vigotti, G. Rossi, L. Bisanti, F. Repetto, A. Zanobetti (Italy); L. Bacharova, K. Fandakova (Slovakia); J. Schwartz (Harvard School of Public Health, Advisor).

The APHEA2 (extended network) collaborative group consisted of: K. Katsouyanni, G. Touloumi, E. Samoli, A. Gryparis, Y. Monopolis, E. Aga, D. Panagiotakos, A. Analitis, Y. Petasakis, K. Dimakopoulou (Greece; Coordinating center); C. Spix, A. Zanobetti, H.E. Wichmann (Germany); H.R. Anderson, R. Atkinson, J. Ayres (U.K.); S. Medina, A. Le Tertre, P. Quenel, L. Pascale, A. Boumghar (Paris, France); J. Sunyer, M. Saez, F. Ballester, S. Perez-Hoyos, J.M. Tenias, E. Alonso, K. Kambra, E. Aranguez, A. Gandarillas, I. Galan, J.M. Ordonez (Spain); M.A. Vigotti, G. Rossi, E. Cadum, G. Costa, L. Albano, D. Mirabelli, P. Natale, L. Bisanti, A. Bellini, M. Baccini, A. Biggeri, P. Michelozzi, V. Fano, A. Barca, F. Forastiere (Italy); D. Zmirou, A Balducci (Grenoble, France); J. Schouten, J. Vonk (The Netherlands); J. Pekkanen, P. Tittanen (Finland); L. Clancy, P. Goodman (Ireland); A. Goren, R. Braunstein (Israel); C. Schindler (Switzerland); B. Wojtyniak, D. Rabczenko, K. Szafraniek (Poland); B. Kriz, M. Celko, J. Danova (Prague, Czech Republic); A. Paldy, J. Bobvos, A. Vamos, G. Nador, I. Vincze, P. Rudnai, A. Pinter (Hungary); E. Niciu, V. Frunza, V. Bunda (Romania); M. Macarol-Hitti, P. Otorepec (Slovenia); Z. Dörtbudak, F. Ekran (Turkey); B. Forsberg, B. Segerstedt (Sweden); F. Kotesovec, J. Skorkovski (Teplice, Czech Republic); J. Schwartz (Harvard School of Public Health, Advisor).