Appropriate comparisons of tuberculosis latency structures with empiric data

Nicolas Menzies and colleagues1 presented a systematic review of the latency structures used in tuberculosis models, which was similar to our 2017 review.2
The main additional contribution of this latest work is the detailed literature search that provides a complete picture of the range of structures in use and highlights the variability in approaches. We strongly agree with the finding that modelled reactivation profiles differ profoundly between studies and that many commonly used structures and parameters fit empirical data poorly. The parameter range findings on goodness of fit validate the results of both studies. However, important differences between the reviews in the two empirical comparison datasets should be highlighted.

First, Menzies and colleagues considered a group of tuberculin skin test (TST)-positive contacts of predominantly (89%) pulmonary index cases from the control arm of a chemoprophylaxis trial,3^,  4 regardless of conversion status. The study was done during the late 1950s in the USA, diagnosis of primary tuberculosis was inconsistent, and around a third of the study population was younger than 10 years and two-thirds were younger than 20 years.3^, 4

Next, Menzies and colleagues fit model parameters to the placebo-assigned cohort of a BCG vaccination trial5 done in the UK in the 1950s. Follow-up with clinical radiological review and TST was undertaken in that trial at intervals of up to 14 months to estimate the timing of disease and infection. All participants were aged 14–19 years, but even within this narrow range, rates of progression were considerably higher in the younger than in the older teenagers.5

We urge care if considering the parameter estimates from Menzies and colleagues’ study when formulating dynamic models of tuberculosis transmission given the substantially different reactivation profiles with age and exposure history. The authors justify their choice of these older studies with the argument that more recent data are uninterpretable because of the widespread use of chemoprophylaxis. However, in our review,2 we fitted to a cohort recruited in a setting of minimal reinfection and with an age distribution that was likely more reflective of newly infected populations, and we were careful to remove individuals receiving preventive treatment from the analysis and present results disaggregated by age group.6

In their discussion, Menzies and colleagues refer to the data from the adult stratum of the study6 we used for parameter estimation and those from another modern study7 of recently infected contacts in Amsterdam that found a virtually identical reactivation profile. These adult data cannot be directly compared with the historical data used by Menzies and colleagues because of the very different age distributions considered and the absence of age stratification in the historical study.

We believe that preference should be given to the modern estimates, particularly those where preventive therapy and loss to follow-up were accounted for,6 because they were derived from populations that could be disaggregated by age, had broader age profiles, had clearly demonstrated recent infection, and were at minimal risk of reinfection.

We declare no competing interests.