Comparing approaches for studying the effects of climate extremes - a case study of hospital admissions in Sweden during an extremely warm summer

Abstract

Background: Health effects induced by climate, weather and climatic change may act directly or indirectly on human physiology. The future total burden of global warming is uncertain, but in some areas and for specific outcomes, mortality and morbidity are likely to increase. One likely effect of global warming is an increasing number of extreme weather events, such as floods, storms and heat waves. The excess numbers of specific health outcomes attributable to climate-induced events can be estimated. This paper compares approaches for estimating excess numbers of outcomes associated with climate extremes, exemplified by a case study of hospital admissions during the extremely warm summer of 2006 in southern Sweden.

Materials and methods: Daily hospital admission data were obtained from the Swedish National Board of Health and Welfare for six hospitals in the Skåne region of southern Sweden for the period 1998 to 2006. Daily temperature data for the region were obtained from the meteorological station in the city of Malmö. We used four established approaches for estimating the daily excess numbers associated with extreme heat. Time series of daily event rates were assumed to follow a Poisson distribution. Excess event rates were compared by using several approaches, such as standardised event ratios and generalised additive models to estimate the health risks attributable to the extreme climate event.

Results: The four approaches yielded vastly different results. The estimates of excess were considerably biased when not accounting for time trends in previous years' data. Three of four approaches showed a significant increase in excess hospitalisation rates attributable to the heat episode in Skåne in 2006. However, modelling the effect of temperature failed to describe the risks induced by the extreme heat.

Conclusion: Estimates of excess events depend greatly on the approach used. Further research is needed to identify which method yielded the most accurate estimates. However, one of the approaches used generally seem to perform better than the others in estimating the excess rates associated with the heat episode. Further on, estimating relative risks of temperature or other determinants of disease may fail to incorporate the unique characteristics of particular weather events, such as the effect caused by very persistent heat exposure. Unless this can be incorporated into predictive models, such models may be less appropriate to use when predicting the future burden of heat waves on human health.

Keywords: climate change; extreme event; heat; hospital admission; surveillance; temperature; weather.

Fig. 1.

To the left; excess admissions in summer 2006 derived from observed numbers in summer 2006 minus observed numbers during the two preceding summers. To the right; corresponding ratios of observed to expected numbers. The dot-dashed lines correspond to weekly mean temperatures during summer 2006 and the dashed lines to the weekly mean temperatures during the two preceding summers.

Fig. 2.

To the left; excess admissions during summer 2006 as observed minus expected, and to the right; ratios of observed to expected, with expected as predicted from a regression model based on patterns during the two preceding years. The dot-dashed lines correspond to weekly mean temperatures during summer 2006 and the dashed lines to the weekly mean temperatures during the two preceding summers.

Fig. 3.

To the left; excess admissions in summer 2006 derived from a regression model incorporating a smooth function of time of the heat event to capture associated risks, and to the right; corresponding ratios of modelled frequencies (RR) associated with the heat event divided by expected frequencies if no heat event. The dot-dashed lines correspond to weekly mean temperatures during summer 2006 and the dashed lines to the weekly mean temperatures during the two preceding summers.

Fig. 4.

Smooth functions of excess risk for hospitalisation during the record breaking warm summer of 2006 in cause-specific groups. The shaded region is the 95% confidence intervals.

Fig. 5.

The observed weekly count of hospital admissions in summer 2006 subtracted from the predicted count from the time series regression model with smooth functions describing excess risks during the heat event, also corresponding to the model residuals during summer 2006. The dot-dashed line corresponds to weekly mean temperatures during summer 2006 and the dashed line to the weekly mean temperatures during the two preceding summers.

Fig. 6.

To the left; weekly admissions in summer 2006 attributed to daily variation of temperature compared with mean temperature levels in 2004 and 2005. The model incorporates smooth functions of temperature effect in lag strata 0–1, 2–6 and 7–13, and to the right; corresponding risks relative to the risk attributed to mean daily temperatures of 2004–2005. The dot-dashed lines correspond to weekly mean temperatures during summer 2006 and the dashed lines to the weekly mean temperatures during the two preceding summers.

Fig. 7.

The risks of hospital admission as a function of temperature in lag strata of 0–1, 2–6 and 7–13. The shaded areas are 95% confidence intervals.