Ebola virus disease mathematical models and epidemiological parameters: a systematic review

Abstract

Ebola virus disease poses a recurring risk to human health. We conducted a systematic review (PROSPERO CRD42023393345) of Ebola virus disease transmission models and parameters published from database inception to July 7, 2023, from PubMed and Web of Science. Two people screened each abstract and full text. Papers were extracted with a bespoke Access database, 10% were double extracted. We extracted 1280 parameters and 295 models from 522 papers. Basic reproduction number estimates were highly variable, as were effective reproduction numbers, likely reflecting spatiotemporal variability in interventions. Random-effect estimates were 15·4 days (95% CI 13·2-17·5) for the serial interval, 8·5 days (7·7-9·2) for the incubation period, 9·3 days (8·5-10·1) for the symptom-onset-to-death delay, and 13·0 days (10·4-15·7) for symptom-onset-to-recovery. Common effect estimates were similar, albeit with narrower CIs. Case-fatality ratio estimates were generally high but highly variable, which could reflect heterogeneity in underlying risk factors. Although a substantial body of literature exists on Ebola virus disease models and epidemiological parameter estimates, many of these studies focus on the west African Ebola epidemic and are primarily associated with Zaire Ebola virus, which leaves a key gap in our knowledge regarding other Ebola virus species and outbreak contexts.

Figure 1. PRISMA flowchart illustrating the systematic review process.

Figure 2. Basic reproduction numbers (R₀) by outbreak.

Each panel corresponds to a different outbreak of EVD with the associated EV species in brackets. Points represent central estimates, with symbol shapes corresponding to central value type. Thick coloured shaded lines represent the range of central estimates when R₀ was estimated, for example, across regions or over time. Solid coloured bars represent the uncertainty around the central estimate; this was reported in different formats including standard deviation (in which case the bar represents +/- the standard deviation), 95% highest posterior density interval, range, interquartile range, 95% CrI (credible interval) and 95% CI. The x-axis has been restricted to a maximum of 10 for clarity. All parameters are from articles with a QA score of >= 50% (see appendix pp 21-23 for all R₀ estimates).

Figure 3. Meta-analysis for the mean A) serial interval, B) incubation period, C) time from symptom onset to death, and D) time from symptom onset to recovery.

All included studies have a QA score of >=50%. Parameters used in the meta-analyses are paired mean and standard deviation of the sample or were converted into mean and SD of the sample from the following combinations: mean and standard error, median and interquartile range, or median and range (see appendix pp 9-10). Blue squares are mean estimates from each study with 95% confidence intervals. Diamonds represent the overall mean across studies for the common and random effect models. The random effect model accounts for within-study and between-study variance.

Figure 4. Case Fatality Ratio (CFR) estimates (%) across outbreaks.

Each panel corresponds to a different EVD outbreak, with the associated EV species in brackets. Points represent central estimates, with symbol shapes corresponding to analysis type: adjusted, naïve or unspecified. Thick coloured shaded lines represent the range of central estimates when the CFR was estimated, for example, across regions or over time. Solid coloured bars represent the uncertainty around the central estimates, reported in different formats including 95% CrIs and 95% CIs. All parameters are from articles with QA scores of >= 50% (see appendix pp 39-42).