Building up a posteriori percentiles for Therapeutic Drug Concentration Monitoring
Aziz Chaouch (1,2), Robert Hooper (3), Chantal Csajka (1,4), Valentin Rousson (2), Yann Thoma (3), Thierry Buclin (1)
(1) Division of Clinical Pharmacology, University Hospital, Lausanne, Switzerland (2) Division of Biostatistics, Institute for Social and Preventive Medicine, University Hospital, Lausanne, Switzerland (3) School of Engineering and Business of the Canton of Vaud, Yverdon, Switzerland (4) School of Pharmaceutical Sciences, Universities of Geneva and Lausanne, Switzerland
Objectives: We propose to calculate a posteriori percentile curves (i.e. percentiles from the posterior predictive distribution of concentrations) for the rendering of Therapeutic Drug Concentration Monitoring results in a patient, for whom past concentration measurements are already available. We illustrate the clinical usefulness of such a posteriori percentiles to:
- Determine the probability that future concentrations lie within a prespecified therapeutic range, under either the current or a modified dosing regimen
- Detect significant changes in drug disposition, e.g. following drug-drug interactions or malfunction of elimination organs
- Identify patient adherence issues
Methods: Considering the population pharmacokinetic model of Voriconazole [1], a set of 1 to 10 simulated trough concentration measurements was generated for a fictive patient receiving 400 mg orally b.i.d., while assuming steady-state. The joint posterior distribution of random effects for this specific patient was obtained using the Sampling Importance Resampling (SIR) algorithm [2,3], while considering the parameter estimates of the model as known. Monte-Carlo simulations were then used to reconstruct the posterior predictive distribution of concentrations over the next dosing interval. This enabled us to assess the expectedness of future concentration measurements in the patient, and the probability of the next trough concentration to lie within the therapeutic range, under different dosing regimens.
Results: We show how the incremental consideration of historical information can reduce the width of the prediction interval for future concentrations in the patient being monitored. We also illustrate how a posteriori percentile curves may detect a future abnormal concentration measurement as the patient gradually becomes his/her own reference, whereas a priori percentiles fail to detect such abnormalities.
Conclusions: When past concentration measurements are available for a patient under monitoring, the rendering of a posteriori percentile curves depicts the likelihood of future concentrations in this patient, under the current or an adapted dosing regimen. Such percentiles constitute an important piece of information that can be graphically communicated to the attending physician, who can then judge whether a measured concentration is both expected and appropriate for his/her patient. This will contribute to better informed treatment decisions, representing a further step towards individualized drug dosage adaptation.
References:
[1] Pascual, A. et al. (2012) Challenging recommended oral and intravenous voriconazole doses for improved efficacy and safety: population pharmacokinetics-based analysis of adult patients with invasive fungal infections. Clin Infect Dis, 55(3):381-390
[2] Rubin, D.B. (1988) Using the SIR algorithm to simulate posterior distributions. In Bayesian Statistics 3, eds, M. H. Bernardo, K. M. DeGroot, D. V. Lindley, and A. F. M. Smith, Cambridge, MA: Oxford University Press, 395-402
[3] Smith, A.F.M. and Gelfand, A.E. (1992) Bayesian statitics without tears: A sampling resampling perspective. The American Statistician, 46(2):84-88
