Junjie Ding (1,2,3), Richard M. Hoglund (1,2), Joel Tarning (1,2,3)
1. Mahidol-Oxford Tropical Medicine Research Unit, Mahidol University, Bangkok, Thailand; 2. Centre for Tropical Medicine and Global Health, Nuffield Department of Clinical Medicine, University of Oxford, Oxford, UK; 3. The WorldWide Antimalarial Resistance Network, Oxford, UK
Introduction: Adherence to medication is crucial for expected treatment outcomes. Poor adherence can result in sub-optimal drug exposure, thereby increasing the risk of treatment failure. Furthermore, poor adherence in many disease areas can cause resistance development, resulting in both patient specific and population health risks. Although a number of approaches have been developed to access treatment adherence in patients, many of them are not robust enough to be used as standard methods [1]. Population pharmacokinetic-based approaches [2,3], in which levels of drug and/or its metabolite are measured to assess the adherence, are objective approaches which should be able to identify if a patient has taken the medicine without asking the patient or counting tablets. However, the accuracy of these methods needs to be evaluated before application in a clinical setting.
Objectives: The aim of this project was to evaluate the predictive performance of two population pharmacokinetic-based approaches (the percentile and the Bayesian method) for adherence assessment in a clinical setting of malaria therapeutics.
Methods: A drug with first-order absorption and 1-compartment disposition kinetics was used as the basis for the simulation study and was assumed to be administered once daily for 3 days. Stochastic simulations (n=2,000) were performed for all possible dosing scenarios, from full adherence to complete non-adherence. Two different methods to evaluate adherence were investigate:
- The percentile method, in which different cut-off concentrations (e.g. 5th percentile of predicted concentrations) were calculated based on the simulations from the full adherence scenario. A simulated concentration at time t below the generated cut-off value was considered as non-adherence to the treatment.
- The Bayesian approach, in which the posterior probability was calculated at a given concentration according to conditional probability and the prior equiprobable probability. The most plausible dosing scenario of a given concentration at time t, is the one with the largest posterior probability among all scenarios.
The predictive performance of these methods was evaluated by assessing sensitivity, specificity, Youden’s index, as well as by evaluating the receiver operating characteristic (ROC) curves. In addition, the impact on predictive performance of different magnitudes of inter-subject variability, and more complicated structure models were investigated. Finally, the best performing method was used to assess the adherence to an antimalarial drug in a programmatic setting in Africa.
Results: For the one-compartment disposition model, the highest Youdex’s index as well as ROC AUC was observed at Tmax and were reduced with increasing time. The predictive performance of the Bayesian approach was similar to the percentile method when investigating the adherence of the last dose (assuming full adherence for the first two doses), but the percentile method was superior when investigating the adherence at all other dosing scenarios. Different magnitudes of inter-subject variability had an impact on the predictive performance of both approaches, resulting in lower Youdex’s index and ROC AUC in the event of higher variability. For the percentile method it was found that a higher percentile cut-off value (e.g., 20-30%) was preferred when the inter-subject variability was high (60%). Using a two-compartment disposition model resulted in the same findings and conclusions as that found with a one-compartment model. When applying the percentile method to a clinical trial of distributing seasonal antimalarial prophylactic therapy and collecting one drug concentration, the method predicted full adherence in less than 20% of the children.
Conclusions: The two population pharmacokinetic-based approaches investigated here, showed satisfactory predictive performances, but the percentile method was superior when investigating more than one missing dose. The percentile method was also successfully applied in a clinical trial setting of an antimalarial drug.
References:
[1] Osterberg, L. & Blaschke, T. Adherence to medication. N. Engl. J. Med. 353, 487–497 (2005).
[2] Zhang, C. et al. Population pharmacokinetic model for adherence evaluation using lamivudine concentration monitoring. Ther. Drug Monit. 34, 481–484 (2012).
[3] Barriere, O., Li, J. & Nekka, F. A Bayesian approach for the estimation of patient compliance based on the last sampling information. J. Pharmacokinet. Pharmacodyn. 38, 333–351 (2011).
Reference: PAGE 28 (2019) Abstr 9025 [www.page-meeting.org/?abstract=9025]
Poster: Methodology - Other topics