Validation of a Bayesian Ebola Diagnostic Model Using Data from the Tenth Epidemic of 2018 to 2020 in the Democratic Republic of Congo

Abstract

Background/Objectives: Diagnosing Ebola virus disease (EVD) based solely on clinical signs remains challenging in settings where its manifestations overlap with other endemic infections. This study aimed to validate a Bayesian clinical diagnostic model to support decision-making for suspected EVD cases in resource-limited settings without access to laboratory confirmation. Methods: We conducted a retrospective analysis using data from the 10th Ebola outbreak in the Democratic Republic of the Congo (2018–2020). Clinical information from 450 suspected and laboratory-confirmed EVD cases was evaluated by seven epidemiological surveillance experts. Data were analyzed using descriptive statistics, binary logistic regression, discriminant analysis, and receiver operating characteristic (ROC) curves to assess diagnostic performance. Results: At the predefined 0.80 threshold, the SBM showed a sensitivity of 82.4%, specificity of 33.7%, PPV of 80.5%, and NPV of 36.5%. Balanced accuracy was 58.0%, with LR+ of 1.24 and LR− of approximately 0.52. Thus, although the model identified a relatively high proportion of RT-PCR confirmed cases, its ability to discriminate non-EVD cases was limited. At the symptom level, vomiting (p = 0.003), anorexia (p = 0.041), abdominal pain (p = 0.024), and muscle pain (p = 0.006) were independently linked to confirmed cases. Conclusions: This Bayesian model shows potential as a decision support tool for early identification of EVD in settings lacking laboratory capacity. Its use could improve case detection and support outbreak response in resource-limited contexts. The expert-derived SBM showed relatively high sensitivity but limited specificity in this selected outbreak cohort. These findings support further recalibration and independent validation rather than use of the model as a stand-alone diagnostic tool.