Development of a prediction model for ebola virus disease: A retrospective study in nzérékoré ebola treatment center, Guinea

Abstract

The 2014 Ebola epidemic has shown the importance of accurate and rapid triage tools for patients with suspected Ebola virus disease (EVD). Our objective was to create a predictive score for EVD. We retrospectively reviewed all suspected cases admitted to the Ebola treatment center (ETC) in Nzérékoré, Guinea, between December 2, 2014, and February 23, 2015. We used a multivariate logistic regression model to identify clinical and epidemiological factors associated with EVD, which were used to create a predictive score. A bootstrap sampling method was applied to our sample to determine characteristics of the score to discriminate EVD. Among the 145 patients included in the study (48% male, median age 29 years), EVD was confirmed in 76 (52%) patients. One hundred and eleven (77%) patients had at least one epidemiological risk factor. Optimal cutoff value of fever to discriminate EVD was 38.5C. After adjustment on presence of a risk factor, temperature higher than 38.5C (odds ratio [OR] = 18.1, 95% confidence interval [CI] = 7.6-42.9), and anorexia (OR = 2.5, 95% CI = 1.1-6.1) were independently associated with EVD. The score had an area under curve of 0.85 (95% CI = 0.78-0.91) for the prediction of laboratory-confirmed EVD. Classification of patients in a high-risk group according to the score had a lower sensitivity (71% versus 86%) but higher specificity (85% versus 41%) than the existing World Health Organization algorithm. This score, which requires external validation, may be used in high-prevalence settings to identify different levels of risk in EVD suspected patients and thus allow a better orientation in different wards of ETC.