Background: Pyrazinamide (PZA) is a key drug for both drug-sensitive and drug-resistant tuberculosis (TB). Patients co-infected with TB and human immunodeficiency virus (HIV) are more likely to have low blood levels of PZA, associated with inferior outcomes. Therapeutic drug monitoring (TDM) with sparse blood sampling is recommended for high-risk groups, including HIV/TB patients, but the accuracy is uncertain. We performed a pharmacokinetic (PK) simulation study to estimate the diagnostic accuracy of TDM for PZA among HIV/TB patients.
Methods: We recently performed a population PK study among HIV/TB patients in Botswana, identifying a 1-compartment model with first-order elimination. In the current work, we performed an intensive PK simulation (n= 10,000 patients) to determine the accuracy of sparse blood sampling in identifying HIV/TB patients with low PZA blood levels, as defined by the AUC in a dosing interval (AUC0-24) predictive of successful outcome (363 mg*hr/l). PZA dosing followed WHO guidelines with weight-based dosing bands. In secondary analysis, we examined the peak concentration (Cmax) target predictive of 2-month sputum conversion (58 mg/l). To determine the accuracy of sparse sampling (2- and 6-hours), we performed receiver-operating-characteristic (ROC) analysis, with bootstrapping (n=1,000) for 95% confidence intervals (CI), and defined accuracy as the area under the ROC curve.
Results: In this simulation PK study of PZA among HIV/TB patients, the PZA AUC0-24 fell below the target in 29% of patients, while in 71% of patients the PZA Cmax was below the target. For the AUC0-24 target, the area under the ROC curve was 0.69 (95% CI 0.680.70) for a single 2-hour sample, increasing to 0.75 (95% CI 0.74-0.76) for 2- and 6-hour samples. For the Cmax target, diagnostic accuracy was similar for a 2-hour sample (0.87, 95% CI 0.860.87) and 2- and 6-hour samples (0.88, 95% CI 0.880.89).
Conclusion: We observed modest diagnostic accuracy of TDM for identifying in silico HIV/TB patients with low PZA AUC0-24, and higher accuracy for low Cmax. By identifying diagnostic performance characteristics of sparse sampling strategies, including optimal cut-offs, the ROC framework can support wider implementation of TDM in high-risk TB populations.