Methods for estimating tuberculosis incidence and mortality by age and sex

To develop methods to disaggregate World Health Orgagnization estimates of tuberculosis (TB) incidence and mortality for each country by sex and age.

For countries where incidence estimates derived from a factor adjustment of notifications and case detection ratio over 0.85, or with <1000 reported TB cases, we disaggregated incidence proportional to notifications. For each other country, a prior was constructed using a hierarchical model of age-stratified prevalence survey data, meta-analysis of sex ratios, and mathematical modelling for children under 15 years. Samples from this prior were used to disaggregate incidence and accepted if incidence exceeded notifications in each age/sex category. Results were inspected and, if implausible, incidence was disaggregated proportional to notifications. Mortality was disaggregated proportional to patterns in vital registration (VR) data in countries with VR data. Where VR data were lacking, a case-fatality ratio (CFR) approach was applied to estimated incidence, with separate CFRs by HIV/ART status, child/adult age groups, and anti-TB treatment status. Uncertainty in all disaggregated country estimates was constructed to be consistent with corresponding overall uncertainty.

We generated disaggregated results for 216 countries. For 125 countries, incidence disaggregation was based on notifications. Of the rest, accepted samples from the prior were considered implausible in 4 countries. For 72 countries, mortality disaggregation was based on VR data; the rest were based on the CFR approach.

While multi-stage, this approach is comparatively simple in overall logic. Disaggregated estimates have relatively larger uncertainty and should be used with caution.