Document resource
Background Knowing a pathogen’s lethality is essential for effective preparedness and response. WHO estimates the Case Fatality Ratio (CFR) for avian influenza A(H5N1) at 54%, but this is likely inflated due to under ascertainment of mild cases. Harnessing the flexibility of the Bayesian approach, we produce a revised estimate of CFR for A(H5N1) accounting for the underreporting of both cases and deaths and quantify the uncertainty therein.Methods Our analysis uses World Development Indicators (WDI) from the World Bank, the A(H5N1) WHO confirmed cases and deaths tracker by country (2003–2024), and A(H5N1) line list data from peer-reviewed literature (2003–2015) that includes confirmed, probable, and suspected cases.Since diseases do not recognise borders, our analysis places individual countries within clusters of countries sharing similar WDIs relevant to A(H5N1), which include HIV prevalence, Tuberculosis incidence, GDP, rural population size, availability of healthcare and poverty metrics. We derive clusters by performing Hierarchical Clustering with missing data imputation and dynamic pruning on all countries in the database.To estimate the overall and by-cluster CFR, we fit two Negative Binomial Bayesian Hierarchical Models. The first is fitted to the WHO tracker, the second to the line list. In both models we introduce random effects for the group-level variables: ‘Cluster’, as derived earlier, and ‘Year of case/death recording’. In the line list model, we include age group and sex as individual-level variables.We explicitly modelled key unobserved parameters (such as reporting rate for cases and deaths, infection probability, true CFR) with informative priors from expert opinion and literature. This allowed us to minimise the use of uninformative priors. Model diagnostics include assessing convergence and sensitivity to the choice of priors.Results Our algorithm clustered countries in four groups. The posterior from the first model suggests an overall estimate of 33.4% with 95% Credible Interval 2.2% – 85.7% from confirmed cases. However, the line list and the introduction of age and sex further modifies the posterior’s shape, producing an overall estimate of 21% with 95% Credible Interval 0.0% – 91.5%.Conclusion By modelling underreporting and broadening the case definition to include probable and suspected cases, our analysis suggests much lower fatality compared to WHO’s CFR estimate on lab-confirmed cases. However, credible intervals are wide. Therefore, good preparedness for a potential A(H5N1) pandemic implies adopting scenario planning under our central estimate CFR, as well as for CFRs as high as 92%.