Transmission modeling to infer tuberculosis incidence prevalence and mortality in settings with generalized HIV epidemics

Abstract

Tuberculosis (TB) killed more people globally than any other single pathogen over the past decade. Where surveillance is weak, estimating TB burden estimates uses modeling. In many African countries, increases in HIV prevalence and antiretroviral therapy have driven dynamic TB epidemics, complicating estimation of burden, trends, and potential intervention impact. We therefore develop a novel age-structured TB transmission model incorporating evolving demographic, HIV and antiretroviral therapy effects, and calibrate to TB prevalence and notification data from 12 African countries. We use Bayesian methods to include uncertainty for all TB model parameters, and estimate age-specific annual risks of TB infection, finding up to 16.0%/year in adults, and the proportion of TB incidence from recent (re)infection, finding a mean across countries of 34%. Rapid reduction of the unacceptably high burden of TB in high HIV prevalence settings will require interventions addressing progression as well as transmission.

Fig. 1. Process flow and model diagram.

Blue lines for the TB transmission model in step 4 are processes to which HIV/ART-associated incidence rate ratios IRR(t,s,a,ART) are applied. These time-dependent IRRs (step 2) capture the implications of the AIM model of HIV/ART progression (step 1) for TB risk for each aggregated age, sex, and HIV/ART stratum in the simplified transmission model (step 4). The influence of ART-protection and CD4 risk dependence parameters on simplified transmission model dynamics (step 4) is achieved by emulating the dependence of IRR trajectories on these model parameters (step 3), allowing them to be included in inference (step 5). Red boxes in step 4 represent calibration targets.

Fig. 2. Model outputs compared to empirical data for 12 African countries.

Rows a, f total population (black), people living with HIV (red), people on ART (green) 1980–2019 - points=data, lines=model. Rows b, g: demographic snapshot in 2015 red/left=women, blue/right=men, points=data. Rows c, h: per capita TB prevalence 1980–2019, (line = median, ribbon = 95% credible interval [CrI]), and TB prevalence survey data (point=central estimate; error bar = 95% confidence interval). Rows d, i: incident TB 1980–2019, lines = median/ribbon = 95%CrI for model, points=data, blue = TB incidence, black = notified TB, red = notified TB/HIV, green=notified TB/HIV on ART (data). Rows e, j: incident TB in 2015 by age, lines = median/ribbon = 95%CrI for model, points = data, blue = TB incidence, black = notified TB, red = incident TB/HIV. All medians and CrIs are based runs using n = 300 random samples from the posterior parameter distribution for each country.

Fig. 3. Per capita TB incidence estimated for 12 African countries, 1980–2019.

Line = median, ribbon = 95% credible interval [CrI]. All medians and CrIs are based runs using n = 300 random samples from the posterior parameter distribution for each country.

Fig. 4. Comparison with WHO incidence and mortality estimates for all TB and TB/HIV for 2019.

Line=central estimate, error bar = 95% uncertainty interval [UI]. All model central estimates (defined as medians) and UIs (defined as 95% credible intervals) are based runs using n = 300 random samples from the posterior parameter distribution for each country. Central estimates and UIs for WHO data (x-axis) are as reported.

Fig. 5. Estimates of other epidemiological metrics.

a The proportion of incident TB that is TB/HIV for 1980–2019. b Annual risk of TB infection in 2019 by age. c Proportion of TB transmission from each age group in 2019. d Proportion of all TB incidence in 2019 due to (re)infection within 2 years (error bar = 95% credible interval [CrI]). e Proportion of all TB incidence in 2019 due to (re)infection within 2 years for each age group. All points/lines=medians. All medians and CrIs are based runs using n = 300 random samples from the posterior parameter distribution for each country.