Innovative modelling for predicting TB treatment outcomes in global cohorts

Post Date: 

Our overarching hypothesis is that clinical and laboratory risk factors for the adversity be treatment outcomes of failure, recurrence and mortality are heterogeneous and time-dependent. Precision medicine approaches supported by Machine learning techniques that overcome the limitations of traditional approaches and account for changes over time of on treatment data may yield improved tools to identify at-risk individuals. In addition, these methods establish a framework to evaluate the additional predictive value of combining novel biomarkers with other clinical and laboratory variables.

Aim 1: To develop and validate parsimonious models using A) Baseline and B) longitudinal features predictive of individual adverse TB treatment outcomes.

Aim 1A: To develop and validate a parsimonious model to predict individual adverse TB treatment outcomes using Baseline clinical features. Hypothesis: models employing machine learning algorithms will predict adverse outcomes with better discrimination and calibration than traditional models.

Aim1B: To produce and validate a model applicable at any point during a TB treatment to predict adverse outcomes using longitudinal clinical features. Hypothesis: consideration of time updated Trends in clinical and laboratory features will improve the performance of models to predict adverse outcomes.