Background There is an urgent need for biomarkers that predict TB treatment response in clinical practice and research. Despite its poor specificity and sensitivity, sputum microscopy and culture conversion 8 weeks following treatment initiation remains the recommended surrogate for TB treatment response. A blood-based biomarker with the ability to predict TB treatment response will significantly improve research into TB treatment-shortening trials, trials testing new anti-TB therapy and clinical practice where it can potentially aid in early identification of patients at risk of poor treatment outcomes.

Methods We conducted a pilot, nested case-control study to identify potential biomarkers to predict TB treatment response. All participants completed the PredictTB treatment-shortening clinical trial. All available confirmed relapses at the time of this pilot study (17) and one treatment failure participant were included and 54 controls were randomly selected. Multiplex immunoassays were used to measure serum expression of 50 cytokines at baseline, weeks 04, 08 and 16 and 24. In addition, demographic and symptom data, clinical examination parameters and laboratory results were collected.

Results Using baseline and week 8 parameters, we derived a model that discriminated between relapses and controls with an AUC of 0.81, sensitivity of 0.78 and a specificity of 0.85. Parameters that were most useful in discriminating between relapses and controls were changes from baseline to week 8 in TNF-alpha, sIL2R-alpha, IL 12p70, sVEFFR3, sVEGFR1, E-selectin, and MIP-1. In addition to chest pain and diastolic blood pressure; baseline Apo A1, IL-1beta, and Apo C3 also contributed to the model. Our data also validated a previously published treatment response signature.

Conclusion Our results indicate that a multivariable model may be better at predicting TB treatment response compared to current measures. This work is preliminary and will be combined with a larger cohort.

Funders: Predict TB clinical trial- EDCTP2 programme