Defining and detecting malaria epidemics in the highlands of western Kenya

Abstract

Epidemic detection algorithms are being increasingly recommended for malaria surveillance in sub-Saharan Africa. We present the results of applying three simple epidemic detection techniques to routinely collected longitudinal pediatric malaria admissions data from three health facilities in the highlands of western Kenya in the late 1980s and 1990s. The algorithms tested were chosen because they could be feasibly implemented at the health facility level in sub-Saharan Africa. Assumptions of these techniques about the normal distribution of admissions data and the confidence intervals used to define normal years were also investigated. All techniques identified two "epidemic" years in one of the sites. The untransformed Cullen method with standard confidence intervals detected the two "epidemic" years in the remaining two sites but also triggered many false alarms. The performance of these methods is discussed and comments made about their appropriateness for the highlands of western Kenya.

Figure 1

Locations of the three study hospitals and the administrative districts they serve in the highlands of western Kenya. The inset map of Kenya shows the 15 districts designated by the Government of Kenya as at risk from unstable, temperature-limited, and hence epidemic malaria. The three districts shaded in red are those in the large map. The St. Joseph’s Catholic Mission Hospital (CMH) at Kilgoris, Tabaka CMH, and Kisii District Hospital are shown within their administrative boundaries of (1) Trans Mara, (2) Gucha and (3) Kisii Central District, respectively. The districts are shown against a backdrop of a digital elevation model for which a key is provided. North is to the top of the page.

Figure 2

Meteorologic time-series for study hospitals. Temperature and rainfall profiles for a synoptic (1980-1995). (a) minimum (bottom), mean (middle), and maximum (top) monthly temperatures (°C); (b) average total monthly rainfall (mm). The error bars denote standard errors of the calculated month means.

Figure 3

Time-series of child admissions and epidemic alerts for the three study hospitals.

Figure 4

Cumulative case graphs for child admissions in the three hospitals. Cumulative child admissions (<15 years) in Kilgoris (a), Kisii (b), and Tabaka (c). All years for which data were available are shown, 1980-1999, 1987-2000, and 1981-2000 time periods for Kilgoris, Kisii and Tabaka, respectively. Black dashed lines are all “normal” years. The blue line shows mean average cumulative child admissions over all years. Red lines show epidemic years, defined as the 2 years of highest total admissions. For Kilgoris these exceptional years are 1994 and 1998, for Kisii they are 1996 and 1997, and for Tabaka they are 1997 and 1996.