Hot weather and suicide: a real risk or statistical illusion?

Page et al (2007) reported an association between increased risk of suicide and hot weather. We believe it is important that this finding is compared with similar associations reported in other countries and under similar conditions, particularly for countries with hotter climates but also for those moving through a period of climatic change.

We are a little disappointed that despite the authors' excellent statistical analyses and effective display of results, they determined the shape of their natural cubic splines ‘visually’ instead of using some model selection criterion, for example likelihood ratio tests, Akaike's information criterion (AIC), etc. Page et al also stated that Yip et al (2000) ‘failed to show any significant seasonality with recent UK data’. This may not be entirely accurate as we believe that Yip et al (2000) showed a decreasing seasonal pattern but not that the pattern had vanished.

The ‘unexpected’ reduction in suicide during the heatwave of 2003 reported by Page et al is difficult to explain on the basis of temperature alone, particularly as there was a 13.5-33% increase in general mortality during the 2003 heatwave (Kovats et al, 2006). It is clear that the association of increased mortality with high temperature is not specific to suicide. Hajat et al (2002) reported an almost identical increase in all-cause mortality of 3.34% (95% CI 2.47-4.23) for every 1°C increase in mean temperature above 18°C compared with the 3.8% increase in suicide reported by Page et al. This raises the possibility of an unaccounted confounder linking suicide, total mortality and daily mean temperature above 18°C. Such factors include climatic and non-climatic factors, whether acting independently or as interaction terms, for example number of sunshine hours (Salib & Gray, 1997), relative humidity, rainfall, unusual weather conditions, stress resulting in changes in the hypothalamic-pituitary-adrenal axis or even changes in the solar wind as measured by satellites (Richardson et al, 1994). Chronomics of suicides (Halberg et al, 2005) which do not rely on calendar year but on periodicity of solar wind (Richardson et al, 1994) may provide a plausible and alternative explanation to the findings of Page et al.

Perhaps the only conclusion that can be drawn from reading Page et al's paper is that high temperature may be associated with increased all-cause mortality. Given the very similar rate of increase in all-cause mortality and in suicide, the mechanism by which high temperature affects the rate of suicide should not be expected to differ from that operating for other causes of death.

Although high daily mean temperature may increase suicide risk, this is not an independent risk factor and may not have the implications for public health policy in relation to global warming that Page et al indicated.