Climate change epidemiology: methodological challenges

Abstract

Climate change is now thought to be unequivocal, while its potential effects on global and public health cannot be ignored. However, the complexities of the causal webs, the dynamics of the interactions and unpredictability mean that climate change presents new challenges to epidemiology and magnifies existing methodological problems. This article reviews a number of such challenges, including topics such as exposure assessment, bias, confounding, causal complexities and uncertainties, with examples and recommendations provided where appropriate. Hence, epidemiology must continue to adapt by developing new approaches and the integration of other disciplines such as geography and climatology, with an emphasis on informing policy-making and disseminating knowledge beyond the field.

Appendices

Appendix 1

What is a heat-wave?

The World Meteorological Organisation (WMO) has no formal definition for the term “Heat-wave”, since the characteristics tend to be event-specific (WHO 2009). Table 2 describes, with an emphasis on the methods used to derive criteria for triggering such systems, some national Heat-Health Watch Warning Systems (HHWS) in operation from five countries. All information presented is from a non-systematic search using publicly available sources such as national institutional and ministerial websites.

Table 2 An example of operational HHWS systems in five countries

HHWS for Europe has been reviewed previously by Kovats and Kristie (2006) as well as WHO’s EuroHEAT project. It is clear that according to the examples of HHWS’s presented in the table below, the definitions of heat-waves differ markedly, in the type and number of meteorological variables taken into account, absolute thresholds, duration, as well as the weight given to implications of health effects. It would seem that regional- or city- specific warning plans tend to involve more sophisticated approaches to determine criteria for triggering alerts: examples include Shanghai (Tan et al. 2004) and Philadelphia (Kalkstein et al. 1996), which use synoptic approaches to identify air masses that have been associated with increased mortality.

But does increased complexity in the methods of heat-wave prediction add value to HHWS’s? The EuroHEAT project investigated the predictive value of a number of meteorological characteristics used by current HHWS and their associations with increased mortality (WHO 2009); these include heat-wave duration, intensity, maximum and minimum temperatures (90th percentile of daily distribution), and interval between heat-wave episodes. It was found that maximum and minimum temperatures and duration were significantly associated. This result suggests that perhaps the simplest definitions are adequate for the purpose of public-health planning which is the primary aim of any HHWS’s. Also predictions of simple meteorological variables such as temperatures are usually of higher confidence, with the added advantage of a longer lead-time (WHO 2009), to enable timely implementation of planned actions. However, real-time health-surveillance for the duration of alerts as seen in the French systems may be the most fitting way to maintain reactivity to events such as heat-waves that could quickly escalate into crisis.

Appendix 2—Examples of bias

Example 1

Balanyà et al. (2006) tested the hypothesis that chromosomal inversion polymorphisms of Drosophila subobscura are evolving in response to global warming. However, according to a critique (Rodríguez-Trelles 2007), their conclusions may not be adequately supported by their data owing to a potential systematic bias in their sampling approach. Balanyà et al. compared inversion frequency records collected up to 50 years ago latitudinally across three continents with the corresponding current records gathered on the same dates. Using calendar dates instead of climatological or biological dates could be systematically misleading for two reasons. First, because global climate warming has lengthened the growing season, increasingly at higher latitudes, current biological dates are not expected to represent their corresponding calendar dates from decades ago, the disparity being greater toward the poles. Second, because chromosomal inversion polymorphisms of D. subobscura, a temperate zone species, undergo pronounced seasonal cycles, with seasonal transitions in inversion frequencies occurring in a matter of weeks. Thus, according to the critique, it is possible that the long-term global genetic shift reported by Balanyà et al. is, at least in part, a sampling artefact ensuing from a biological lag between the old and new samples—especially those from higher latitudes(Rodríguez-Trelles 2007). The new samples were collected systematically later than the old ones with respect to the historical onset of the biological spring.

Example 2

Global climate model (GCM) -based output grids can bias the area identified as suitable when these are used as predictor variables for species distribution, because GCM outputs, typically at least 50 × 50 km, are biologically coarse. Seo et al. (2009) tested the assumption that species ranges can be equally well portrayed in species distribution operating on base data of different grid sizes by comparing species distribution statistics and areas selected by four species distribution models run at seven grid sizes, for nine species of contrasting range size. Area selected was disproportionately larger for distribution models run on larger grid sizes, indicating a cut-off point above which model results were less reliable. Up to 2.89 times more species range area was selected by distribution models operating on grids above 50 × 50 km, compared to distribution models operating at 1 km². Spatial congruence between areas selected as range also diverged as grid size increased, particularly for species with ranges between 20,000 and 90,000 km².

Appendix 3

Heat-waves—non-linearity

The PHEWE (Assessment and Prevention of acute Heath Effects of Weather conditions in Europe) project assessed the effect of the “apparent” maximum temperatures (represents the combined discomfort due to heat and humidity) and mortality in 15 countries. It was found that despite differences between the minimum mortality thresholds between the cities, the shapes of the heat-health dose–response curves were remarkably consistent: they were V- or J-shaped curves (Baccini et al. 2008), mirroring results from another large international study, the ISOTHURM project (McMichael et al. 2008). In addition, when the PHEWE cities were stratified into “Mediterranean” and “Northern-Continental” by meteorology and geography, the meta-analytic curves suggest that the heat effect, defined as change in mortality associated with 1°C increase in maximum apparent temperature above city specific threshold, is larger in Mediterranean (3.12%, 95% CI 0.60–5.72) than Northern-Continental (1.84%, 95% CI 0.06–3.64), despite potential acclimatization, which contradicts a previous study in the US (Curriero et al. 2002).

Results such as this have important implications for public health policies, as some national heat-health response plans reply on simple temperature thresholds for activation (Appendix1) and their tiered alert structure also needs to take into account the non-linearity described above to set appropriate thresholds for each level of response. For instance, by identifying the minimum mortality threshold and slope of the response curves by city or region, extrapolation from an analogue location can be misleading due to differences between the distribution of characteristics such as socio-economic-status, demographics, underlying morbidity, race, and adaptation such as use of air conditioning (Kinney et al. 2008).

Moreover, how far should an alert threshold for HHWS stray from minimum mortality threshold on a heat-mortality curve? Although the most reliable evidence of correlation between heat-waves and health is restricted to mortality only, there are a small number of recent studies that suggest that this may extend to morbidity. The association between heat and respiratory mortality was also seen with hospitalization in the PHEWE study (Baccini et al. 2008) and in the elderly (Kovats et al. 2004; Linares and Díaz 2008). While for cardiovascular and circulatory admissions, the evidence is more contradictory (Kovats et al. 2004; Linares and Díaz 2008; Michelozzi et al. 2009; Panagiotakos et al. 2004; Schwartz et al. 2004), although potential biological plausibility has been proposed previously (Pan et al. 1995). This is thought to be attributed to the speed of progression of such events being exacerbated by the high temperature and therefore manifest as mortality instead (Kovats et al. 2004; Linares and Díaz 2008; WHO 2009).