Malaria early warning in Kenya

Abstract

Kenya displays large spatiotemporal diversity in its climate and ecology. It follows that malaria transmission will reflect this environmental heterogeneity in both space and time. In this article, we discuss how such heterogeneity, and its epidemiological consequences, should be considered in the development of early warning systems for malaria epidemics.

Fig. 1

Spider-plots of malaria admissions in Wajir (a) and Kericho (b). The data are monthly averages for the 1991–1998 and 1965–1998 periods, respectively. Adult cases (unfilled circles) and child cases (filled circles) are shown.

Fig. 2

Malaria cases and rainfall by month (1991–1998) for Wajir, Northern Kenya, are shown in (a). The bars are malaria cases and the black line rainfall totals. Observed (blue bars) and predicted (black line) malaria cases in Wajir are shown in (b). The prediction is based on a simple quadratic relationship between present cases (x) and rainfall (y) 3 months previously; where x = 19.9635 − 0.0399y + 0.0018y. The 20-case baseline is thought to represent the background of malaria from very localized transmission, i.e. from around water sources, and imported malaria cases.

Fig. 3

(a) The monthly incidence (cases per 100 000) of Plasmodium falciparum malaria incidence in Kericho from April 1965 to December 1998 are shown in black. The dashed line shows a moving average of 61 months and the bold red line the stationary malaria incidence series (original value – moving average) on which spectral density analysis was performed. Note that cases are treated in most months and that the dramatic post-1980 increases in case numbers has been argued to result from the development of drug resistance. (b) The spectral density plot for malaria incidence at Kericho. A Tukey–Hamming window of three points was applied to smooth the spectral density plot. Details of the variance structure from the periodogram (unsmoothed spectral density plots of frequency) show that annual frequencies and less account for 69.8% of the total variance in the time-series and super-annual frequencies for 30.2%.