# Pharmacokinetic modelling as a tool to assess TB treatment adherence: application to the REMEMBER study

## TL;DR

This study used drug levels in blood samples to assess treatment adherence in a TB prevention trial, finding that poor adherence may have contributed to TB cases.

## Contribution

The study introduces personalized pharmacokinetic thresholds to assess adherence in TB treatment, offering a novel analytical approach.

## Key findings

- Non-adherence was significantly higher in TB cases compared to controls for pyrazinamide metrics.
- Personalized thresholds from simulations were more effective than standard thresholds in detecting non-adherence.
- Poor adherence may have contributed to TB incidence in the REMEMBER study.

## Abstract

The REMEMBER (A5274) study found that the four-drug TB preventive regimen did not reduce mortality compared to isoniazid-only, raising adherence concerns. Using drug measurements and pharmacometrics, we assessed adherence in the four-drug arm by comparing participants who developed TB (cases) to those who did not (controls).

Using a 1:4 matched case-control design, we analysed stored blood samples at weeks 2, 4, and 8 since treatment start. Rifampicin and pyrazinamide were measured, and adherence was assessed using two thresholds: i) lower limit of quantification (LLOQ) and ii) personalised thresholds derived from pharmacokinetic simulations. Population pharmacokinetic models and Monte Carlo simulations were used to predict individualised thresholds assuming 100% adherence. Conditional logistic regression compared non-adherence between cases and controls.

Among 28 cases and 112 controls, the proportion of samples <LLOQ was 52% (cases) versus 45% (controls) for rifampicin and 20% (cases) versus 14% (controls) for pyrazinamide. Non-adherence was significantly higher in cases compared to controls for two pyrazinamide metrics: the week 4 LLOQ (P = 0.050) and the week 2 2.5th percentile personalised threshold (P = 0.023).

Poor adherence may have contributed to TB incidence in REMEMBER. While not definitive, personalised thresholds from model-based simulations remain useful for adherence assessment.