Analysis of time-to-positivity data in tuberculosis treatment studies: Identifying a new limit of quantification

The BACTEC Mycobacteria Growth Indicator Tube (MGIT) machine is the standard globally for detecting viable mycobacteria in patients’ sputum. Samples are observed for no longer than 42 days, at which point the sample is declared “negative” for tuberculosis (TB). This time to detection of bacterial growth, referred to as time-to-positivity (TTP), is increasingly of interest not solely as a diagnostic tool, but as a continuous biomarker wherein change in TTP over time can be used for comparing the bactericidal activity of different TB treatments. However, as a continuous measure, there are oddities in the distribution of TTP values observed, particularly at higher values. We explored whether there is evidence to suggest setting an upper limit of quantification (ULOQM) lower than the diagnostic limit of detection (LOD) using data from several TB-PACTS randomized clinical trials and PanACEA MAMS-TB. Across all trials, less than 7.1% of all weekly samples returned TTP measurements between 25 and 42 days. Further, the relative absolute prediction error (%) was highest in this range. When modeling with ULOQMs of 25 and 30 days, the precision in estimation improved for 23 of 25 regimen-level slopes as compared to models using the diagnostic LOD while also improving the discrimination between regimens based on Bayesian posteriors. While TTP measurements between 25 days and the diagnostic LOD may be important for diagnostic purposes, TTP values in this range may not contribute meaningfully to its use as a quantitative measure, particularly when assessing treatment response, and may lead to under-powered clinical trials.


S1.3 Bayesian Model Specification
This section aims to follow best practices in reporting Bayesian analyses by following the Bayesian Analysis Reporting Guidelines [17].

S1.3.1 Data variables
The dependent variable (y ijk ) is the log 10 (TTP) measured from the sputum sample given by individual i on regimen j at visit k and is modeled as a function of time since randomization (in weeks) (t ijk ).Let Q ijk be an indicator variable that takes the value 1 when y ijk log(ULOQ M ), and 0 otherwise.In other words, Q ijk denotes when a value is above the limit of quantification and therefore cannot be included quantitatively in the model.Instead, these values are handled as "right-censored" [18].The linear mixed effects model is specified as in Eq. 2, with "brms" default priors used for all parameters except 0 .

S1.3.2 Code and software
All analyses were performed using the "brms" package in the R statistical software.All code used to perform these analyses is available at a GitHub repository managed by the first author (https://github.com/sdufault15/ttp-lod).

S1.3.3 Model results and fit
Model results are included in S2 at the regimen-level.Model statistics reflecting fit and convergence at the population-level are included in Tables S3 -S9. Figure S3 contains the posterior predictive checks for the main analysis models.May 6, 2024

Fig S1 .
Fig S1.Histograms showing the distribution of TTP values for each week's sample."Negative" values are recorded as 42 days.

Fig S3 .
Fig S3.Posterior predictive checks comparing the observed response data (y) to data drawn from the posterior of the model (y rep ).

Fig S4 .
Fig S4.Observed time-to-positivity trajectories.Any observations at or above the diagnostic limit of detection (42 days) are recorded as 42 days.A: Regimen-level trends in TTP (lines) and estimated STAND, PaMZard errors (ribbons) as fit by smoothing splines.B: Individual TTP trajectories (light gray) and regimen-level smoothing spline (black).

Fig S5 .
Fig S5.The proportion of samples that return TTP observations below the BACTEC MGIT diagnostic LOD (42 days, green) as well as below various ULOQ M s for each week since randomization for (A) REMox-TB and (B) PanACEA MAMS-TB data.

Fig S6 .
Fig S6.Among regimens with similar or worse bactericidal activity as HRZE, the posterior distributions for the relative comparison of a regimen's slope ( 1j ) against the estimated slope on HRZE ( 1,HRZE ), where a value of 1 indicates equal slopes values (> 1) suggest the regimen has greater bactericidal activity than HRZE.The estimated "confidence" that a regimen has any improvement in bactericidal activity over HRZE (Pr( 1j / 1,HRZE > 1)) as well as the "confidence" that a regimen has more than 10% improvement in bactericidal activity over HRZE (Pr( 1j / 1,HRZE > 1.1)) is indicated for each regimen at each ULOQ M .Annotated are the corresponding values for the posterior probabilities of Pr( 1j / 1,HRZE > ⌧), where ⌧ equals 1 and 1.1, respectively.

Table S2 .
Posterior point estimates (mode) and 95% highest-density credible intervals (HCI) for the regimen-level slopes from the linear models when the TTP diagnostic LOD and different ULOQ M thresholds are applied.