Attributable fraction estimates and case definitions for malaria in endemic

TLDR

In malaria‑endemic areas, asymptomatic parasitaemia is common and parasite detection in febrile patients does not always indicate clinical malaria; comparing parasite prevalence in fever cases to community controls can estimate the attributable fraction, but this method is unreliable in high‑transmission settings where few parasite‑free individuals exist and non‑malarial fevers suppress low parasitaemia, biasing estimates, and a widely used case definition requires fever plus parasite density above a cutoff. The authors explored alternative estimation techniques using 1989‑1991 data from a highly endemic Tanzanian area where over 80 % of young children were parasitaemic. Logistic regression models that treat fever risk as a continuous function of parasite density provide more precise, bias‑resistant estimates of malaria‑attributable fever, enable calculation of the probability that any episode is malaria‑attributable, and can guide the selection of parasite‑density cutoffs by estimating sensitivity and specificity.

Abstract

Abstract Asymptomatic carriage of malaria parasites occurs frequently in endemic areas and the detection of parasites in a blood film from a febrile individual does not necessarily indicate clinical malaria. In areas of low and moderate endemicity the parasite prevalence in fever cases can be compared with that in community controls to estimate the fraction of cases which are attributable to malaria. In areas of very high transmission such estimates of the attributable fraction may be imprecise because very few individuals are without parasites. Furthermore, non‐malarial fevers appear to suppress low levels of parasitaemia resulting in biased estimates of the attributable fraction. Alternative estimation techniques were therefore explored using data collected during 1989‐1991 from a highly endemic area of Tanzania, where over 80 per cent of young children are parasitaemic. Logistic regression methods which model fever risk as a continuous function of parasite density give more precise estimates than simple analyses of parasite prevalence and overcome problems of bias caused by the effects of non‐malarial fevers. Such models can be used to estimate the probability that any individual episode is malaria‐attributable and can be extended to allow for covariates. A case definition for symptomatic malaria that is used widely in endemic areas requires fever together with a parasite density above a specific cutoff. The choice of a cutoff value can be assisted by using the probabilities derived from the logistic model to estimate the sensitivity and specificity of the case definition.