Related Experiment Videos
Applying a Pharmacometrics-Enabled Machine Learning Analysis to Predict 2-Month Culture Conversion Using Phase 2a
Huifang You1, Ulrika S H Simonsson1
1Department of Pharmaceutical Biosciences, Uppsala University, Uppsala, Sweden.
Clinical and Translational Science
|August 10, 2026
Summary
Machine learning models can predict tuberculosis culture conversion using early biomarker data from phase 2a trials. Four-week biomarker data offers better prediction than two-week data for phase 2b studies.
Area of Science:
- Pharmacometrics and Machine Learning in Clinical Trials
- Tuberculosis Drug Development and Biomarker Analysis
Background:
- Traditional tuberculosis (TB) phase 2a trials use early bactericidal activity (EBA) over two weeks, with time-to-positivity (TTP) as a biomarker.
- Phase 2b TB studies typically assess culture conversion over eight weeks, using time-to-event endpoints.
- There is a need to optimize early trial phases for predicting later outcomes in TB drug development.
Purpose of the Study:
- To investigate machine learning (ML) models for predicting the time-to-event of 2-month culture conversion in TB patients.
- To evaluate the impact of different phase 2a study lengths (14-day vs. 28-day TTP data) on prediction accuracy.
- To assess the utility of TTP biomarker data from phase 2a for predicting phase 2b culture conversion endpoints.
Main Methods:
- Nonlinear mixed-effects modeling was used to analyze TTP data at baseline and up to 14 or 28 days from the REMoxTB trial.
- Individual TTP predictions were used as features in subsequent machine learning analyses (e.g., C-support vector classification).
- Model selection was guided by statistical metrics and Kaplan-Meier plots to assess prediction of 8-week culture conversion.
Main Results:
- Bi-exponential and exponential decay models effectively described 14-day and 28-day TTP data, respectively.
- Baseline TTP and/or TTP differences were identified as the most important features across all ML models.
- Four-week phase 2a TTP data provided more predictive information for 8-week culture conversion than two-week data.
Conclusions:
- Machine learning models, utilizing phase 2a TTP biomarker data, show potential for predicting phase 2b culture conversion.
- Extending phase 2a TTP data collection to four weeks improves the prediction of 8-week culture conversion compared to two weeks.
- This workflow demonstrates a pharmacometric approach integrating ML and biomarkers to enhance TB clinical trial efficiency.