Global warming and malaria: a call for accuracy.

For more than a decade, malaria has held a prominent place in speculations on the impacts of global climate change. Mathematical models that " predict " increases in the geographic distribution of malaria vectors and the prevalence of the disease have received wide publicity. Efforts to put the issue into perspective 1–5 are rarely quoted and have had little influence on the political debate. The model proposed by Frank C Tanser and colleagues 6 in The Lancet and the accompanying Commentary by Simon Hales and Alistair Woodward 7 are typically misleading examples. The relation between climate and malaria transmission is complex and varies according to location, 2 yet Tanser et al base their projections on thresholds derived from a mere 15 African locations. Slight adjustments of values assigned to such thresholds and rules can influence spatial predictions strongly. 8 The authors invest considerable effort in assessing the sensitivity of their model to climate change scenarios but do not report the internal sensitivities to thresholds and rules. The predictive skill of their model is low (63% sensitivity, 95% CI 61–65%) but they consider projections acceptable if prevalence is projected " to within a month " (presumably +/– 1 month?), thereby biasing their model towards success. A model covering an entire year in a parasite-positive site would always be correct, although in such areas it would be relatively insensitive to climate. By contrast, sites in which transmission is seasonal would provide a more reliable test of accuracy, but estimation is more difficult because climate sensitivity is greater. Furthermore, because parasite clearance in communities is not instantaneous, 9 spot samples of parasitaemia on survey dates are not a suitable indicator of the duration of the transmission season. Lastly, " person/months " are unsuitable as a measure of transmission: an extension of season from 1 to 4 months will have more impact than from 10 to 12 months. According to their model, an extension of transmission from 11 to 12 months results in 10 6 more person/months in a population of 10 6 people, whereas an extension from 1 to 5 months gives the same increase in a population of 250 000. What Tanser and colleagues have modelled is merely the duration of the transmission season, which they interpret as " heightened transmission " and increased incidence. A greater failing is their reliance on " parasite-ratio studies ". The relations between transmission season and …