Reevaluating progression and pathways following Mycobacterium tuberculosis infection within the spectrum of tuberculosis

Significance Understanding of the risk of progression to tuberculosis (TB) after infection with Mycobacterium tuberculosis (Mtb) has traditionally relied on a binary distinction between infection and infectious, symptomatic disease. However, this advanced disease state is only one of many across a spectrum of disease presentations. We utilized mathematical modeling informed by an extensive systematic review of TB natural history to reevaluate progression and pathways following Mtb infection. We show the impact of different disease thresholds and highlight heterogeneous pathways through the spectrum of disease. These results update our understanding of progression risks and timelines in line with the spectrum of TB to guide more effective prevention, detection, and treatment efforts and avert morbidity and transmission to end TB.

Intervals between tuberculin testing are uncertain and inconsistent across the population ("the majority of the population has been submitted to systematic tuberculin tests… at intervals of one or two years") and not all "converters" had prior evidence of a negative tuberculin reaction.Rubinstein H, Kotschnowa I. Beginn und entwicklung der lungentuberkulose beim erwachsenen.Acta Medica URSS.1940;3(3):250-65.No description of the study methodology underlying reported time between exposure and disease onset could be found.

Data from the longitudinal study of TB natural history conducted by the National Tuberculosis
Institute in Bangalore, India, are included as evidence of progression following Mtb infection (1,2).
Authors of this study, conducted between 1961 and 1968, state that "no organized antituberculosis treatment was available to the people of the area during the entire study period" (1).However, they also report that " [d]uring the second survey and the early part of the third, with a view to obtaining better cooperation, one month's supply of isoniazid tablets was issued to persons in whom pulmonary tuberculosis was diagnosed" (1).This limited course of isoniazid would have had little effect on disease progression and may have contributed to the emergence of drug resistant strains of Mtb.
Authors' acknowledgement of this limited provision implies that investigators had knowledge of and access to isoniazid, yet it appears no efforts were made to provide an effective regimen of the drug to study participants outside the scope of improving "cooperation".We have decided to include data from this research, which would now be considered highly unethical, because we were unable to identify any other data sources that could provide comparable data, and we feel not using the data would make the burden placed on study participants even more onerous.

II. Data Adjustments
After extracting data from three included studies describing progression from Mtb infection to TB, we adjusted those data to reflect uncertainty in the time of tuberculin conversion and disease onset, recognising that neither infection nor disease onset occur at the point those developments are detected in these studies.
For studies reporting progression from TST conversion to minimal disease, we first reduced the number of reported cases by 25% to acknowledge that some pathological anomalies detected by chest x-ray may not have been attributable to TB (3).
For all studies, to estimate time of infection, we sampled from a uniform distribution over the interval between the last negative tuberculin test and the first positive tuberculin test.
For all studies, to estimate time of disease onset, we sampled between the last disease negative screening and the first disease positive screening.The distribution from which we sampled was informed by data from Poulsen (1957) (4) on the proportion of incident cases of TB over time following infection, shown in Supplemental Table 2. Various distributions were examined, and a Cauchy distribution (location=2.0600957,shape=0.8282028) was selected as the best fit to the data.This distribution was aligned with the previously sampled time of tuberculin conversion and truncated to sample within the interval between the last negative disease screen and the first positive disease screen.For Model A, the calibrated incidence of minimal disease was considered reasonable, but the shape of the calibrated incidence of subclinical disease deviated from the shape of the data for this calibration target.For Model B, the shape of the calibrated incidence of subclinical disease was improved, relative to Model A. Trace plots for models C and D suggest these model structures are not consistent with empirical data.Therefore Model B was selected as the most parsimonious for the main model analysis.

IV. Model Calibration
The equations used to describe the main model are as follows:

A V. Sensitivity Analysis
We conducted sensitivity analyses to explore uncertainty in the disease states represented by NTI data due to the lack of reporting of any clinical signs or symptoms.We examined three scenarios with different degrees of clinical representation among cases, classifying cases as either 0%, 50%, or 100% clinical.Calibrations are shown in Supplemental Figure 8 and Supplemental Table 6.

Main model 0% clinical 50% clinical 100% clinical
Supplemental Figure 8: Sensitivity analysis calibrations for different classifications of NTI data The model structure relating Mtb infection to TB states was developed through an iterative process.In total, we examined 4 potential model structures defining progression from Mtb infection in different ways (Supplemental Figure1).Model A allowed only progression from infection to minimal disease.Model B allowed direct progression from infection to subclinical disease in addition to progression from infection to minimal disease.Model C allowed progression to subclinical disease, progression to minimal disease, and indirect progression to minimal disease via an intermediate state.Model D allowed direct progression from infection to clinical disease, in addition to progression from infection to minimal disease and progression from infection to subclinical disease.Supplemental Table5: Model development calibrations.Error bars show calibration targets, and lines and shading show median and 95% uncertainty intervals for calibrated model.
+  + ) *    =  *  − ( + ) *  +  *    =  *  +  *  − ( + ) *  +  *    =  *  − ( + ) *  where I, M, S, and C refer to infection, minimal, subclinical, and clinical states, respectively, and remaining terms represent transition rates with infclr from infection to cleared/recovered, infmin from infection to minimal, minrec from minimal to cleared/recovered, minsub from minimal to subclinical, submin from subclinical to minimal, infsub from infection to subclinical, subclin from subclinical to clinical, clinsub from clinical to subclinical, and clinmort from clinical to death.Appropriate sub-structures from this model were used in the calibration process to align with the data informing calibration targets for different transitions (3).Model structures used to calculate incidence of minimal, subclinical, and clinical disease are shown in Supplemental Figure 2.

Table 2 :
Poulsen (1957) cumulative proportion of incident cases of TB by years since infection perPoulsen (1957) disease at the appropriate time while also reflecting loss to follow-up as reported by each study.We note that no loss to follow-up was reported in manuscripts describing the National Tuberculosis Institute study in Bangalore, India, which risks underestimating progression to subclinical disease in later years following infection.Sensitivity analyses (not shown) indicate that model estimates are robust to minor changes to individual data sources, so potential underestimation of long-term progression in this study is expected to have little impact on posterior transition parameters.Raw and adjusted data are shown in Supplemental Table3and unweighted adjusted incidence is shown in Supplemental Table4.

Table 4 :
Adjusted data showing years since infection, number of events reported, population at risk, and unweighted incidence of minimal or subclinical TB for each study