Abstract
BACKGROUND: The clinical presentation of individuals infected with Plasmodium falciparum is exceptionally diverse, ranging from asymptomatic parasitemia to life-threatening disease. Frequent previous exposure to Plasmodium spp results in partial protection from severe disease; however, this protection wanes in individuals emigrating from holoendemic regions, and there are currently no reliable biomarkers that accurately indicate this semi-immunity. METHODS: Data were analyzed from 1392 adults infected with P falciparum in Gabon and Germany. Full blood count parameters and ratios were evaluated individually and as a combined ensemble-based machine learning classifier to predict disease severity, ranging from asymptomatic infection to severe malaria. As a secondary objective, the influence of previous exposure to Plasmodium spp was assessed. RESULTS: Comparing asymptomatic parasitemia with uncomplicated malaria in Gabonese and comparing uncomplicated with severe malaria in German patients revealed significantly lower platelet counts (218 vs 150 ×10(3)/µL, P < .0001; 85 vs 40 ×10(3)/µL, P < .0001, respectively) and higher neutrophil counts (2.32 vs 2.57 ×10(3)/µL, P = .0037; 3.08 vs 4.49 ×10(3)/µL, P < .0001) in those with greater infection severity. The machine learning classifier outperformed single parameters in differentiating infection severity in both comparisons (area under the receiver operating characteristic curve, 0.94 and 0.84). Lymphocyte and monocyte counts showed a pattern that follows the level of previous malaria exposure, with lower cell counts in naive vs previously exposed patients, regardless of infection severity. CONCLUSIONS: The value of simple full blood count parameters for classification of P falciparum infection severity and previous exposure is considerable. The accuracy can be increased by integrating individual parameters into a joint machine learning model.