The location of the three hospitals that provided inpatient clinical care and were identified for use in this study, along with details about the collection of clinical data and the local weather conditions, are provided in our companion paper.
Providing a precise catchment area of the population for admission's data was not possible as such information is not routinely collected in the hospitals; we assumed therefore that most inpatients came from the immediate surrounding high-elevation catchment area. Typically, long-term, facility-based data are difficult to interpret without some estimate of the populations served and how the population may have changed over time. To provide demographic information on the number of people served by the hospitals, we used population estimates from Kenyan national censuses in 1979, 1989, and 1999. District and lower level administrative boundaries changed with each census, so population growth rates were defined for three contiguous administrative areas, as close to the hospital as possible, which had not been subjected to boundary redefinition. Intercensal population growth rates (r) were calculated by using the formula r=loge(t2-t1), where t1 is the population estimate of the first census and t2 the population estimate of the second.
Data manipulation and statistical transformation were performed in Excel 2000 (Microsoft Corp., Seattle, WA) unless otherwise stated. Monthly mean adult (≥15 years) and child (<15 years) admissions were calculated and displayed as spider plots. The time series of admissions at each site was also plotted with a 25-point (month) moving average of the series to show more clearly the long-term movement in these data. For each hospital, we performed trend analysis through linear regression of the malaria admission data against a trend variable (observation month number/12). The coefficient of trend therefore indicates the annual trend (positive = increasing in time, negative = decreasing in time, zero = a stationary series). Such regression models are sensitive to seasonal variation, outliers, and heteroscedasticity (a term which refers to situations in which the variability of the residuals is not constant). To show the long-term trend unambiguously, seasonality was removed from each series by using an additive seasonal decomposition procedure[37,38] and the residuals were checked for normality and heteroscedasticity in SPSS version 11 (SPSS Inc., Chicago, IL).
Furthermore, we applied a proximate measure of transmission stability through a comparison of the numbers of adults to children with malaria admitted to the three hospitals. This adult:child ratio (ACR) of cases was calculated from total adult and child admissions for the duration of the available records. In areas of stable transmission, we assumed that the risks for complicated malaria in adulthood would be significantly lower than the risks in childhood; this assumption was based on expectations of the age distribution of malaria cases under varying transmission intensities[39,40,41]. Conversely, areas of infrequent parasite exposure lend themselves to equivalent risks in adults and children. As such, an ACR derived from hospital admissions approaching unity would suggest an increasing tendency toward unstable transmission, assuming that the typical age-structured population pyramid for developing country and rural communities prevailed and that there were no age-dependent biases in attendance rates.
Emerging Infectious Diseases. 2002;8(6) © 2002 Centers for Disease Control and Prevention (CDC)
Cite this: Clinical Epidemiology of Malaria in the Highlands of Western Kenya - Medscape - Jun 01, 2002.