Characterising progression and regression patterns across the spectrum of tuberculosis: a multistate modelling approach

BACKGROUND

The conceptualisation of tuberculosis has undergone a paradigm shift from binary states to a spectrum, resulting in the International Consensus for Early TB (ICE-TB) framework. This study aimed to use data from a prospective, observational cohort study and multistate modelling to address the lack of contemporary data to quantify movement between ICE-TB states.

METHODS

ERASE-TB was a prospective, observational cohort study evaluating novel diagnostic tests for earlier detection of tuberculosis. Household contacts aged at least 10 years in Zimbabwe, Tanzania, and Mozambique were followed up 6-monthly for 12-24 months with comprehensive tuberculosis investigations at each visit. Those not diagnosed with prevalent tuberculosis, with state classification from at least two timepoints were included. ICE-TB states were defined by use of symptomatology, interferon gamma release assays, chest radiographs, and sputum microbiology. A Markov multistate model based on ICE-TB was applied with one initial state (Mycobacterium tuberculosis non-infection), two intermediate states (M tuberculosis infection and non-infectious disease [asymptomatic-symptomatic]), and one absorbing state (infectious disease [asymptomatic-symptomatic]). Transition probabilities were predicted.

FINDINGS

1789 (84·8%) of 2109 recruited household contacts were included. At enrolment, most (1000 [55·9%]) did not have M tuberculosis infection; 674 (37·7%) had M tuberculosis infection, and 115 (6·5%) had non-infectious disease. 34 people developed infectious disease (23 asymptomatic, 11 symptomatic). In the multistate model, the transition probabilities of progressing from M tuberculosis non-infection to M tuberculosis infection and M tuberculosis infection to non-infectious disease were 13% and 3% by month 12. For those in non-infectious disease, the probabilities of regression and progression by month 12 were 85% and 13%, respectively.

INTERPRETATION

This study applied the ICE-TB framework to describe movement between states by use of contemporary, granular, longitudinal data. Although most people remained static over time, the non-infectious state was more dynamic, with most people regressing over time.