Sensitivity Analysis for the Malaria-Attributable Fever Fraction and How Collecting Symptom Information Reduces Sensitivity to Unmeasured Confounding

• Activity Number: 6
• Type: Invited
• Date/Time: Sunday, July 31, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Dylan Small*+ and Jing Cheng and M. Elizabeth Halloran
• Companies: University of Pennsylvania and University of California at San Francisco and Fred Hutchinson Cancer Research Center/University of Washington
• Keywords: attributable fraction ; sensitivity analysis ; causal inference

Abstract

The malaria attributable fever fraction (MAFF) is the fraction of fevers that are caused by (attributable to) malaria. The MAFF is important for measuring the magnitude of the public health burden of malaria and changes in this burden, deciding on sample sizes for intervention trials against malaria and deciding on strategies for treating fevers. Usual inference for the MAFF relies on an assumption that there is no unmeasured confounding of the relationship between malaria parasite densities and fever. This assumption is unlikely to be true in most studies. We develop a method of sensitivity analysis for the MAFF that shows how inferences for the MAFF would change if unmeasured confounding of a specified magnitude is allowed for. We show that the sensitivity to unmeasured confounding can be reduced by collecting information on symptoms that help distinguish malaria caused fevers from non-malaria caused fevers such as the temperature of the fever, the duration of the fever, nail or palmar pallor, an enlarged spleen, the absence of a cough or rash and the presence of a normal chest examination.