Stijn W. van Beek (1), Rob ter Heine (1), Rob. E. Aarnoutse (1), Elin M. Svensson (1,2)
(1) Department of Pharmacy, Radboud Institute for Health Sciences, Radboud University Medical Center, Nijmegen, The Netherlands, (2) Department of Pharmaceutical Biosciences, Uppsala University, Uppsala, Sweden
Objectives: Isoniazid (INH) is a drug used in the first-line treatment of tuberculosis (TB). Interindividual variability (IIV) of INH pharmacokinetics (PK) is known to be high [1], this is mainly because of the polymorphic acetylation into acetyl-INH by N-acetyltransferase 2 [2]. In order to achieve optimal treatment, the use of therapeutic drug monitoring (TDM) is advised [3, 4]. It has been suggested that the area under the concentration-time curve from 0 to 24 h (AUC0-24) as measure of exposure is the most relevant to the efficacy of TB drugs [5-7]. Previously, we have shown the potential of a model-based TDM method for rifampicin in treatment of TB [8]. Although various TDM methods for INH are currently in use, no model-based TDM methods for INH have been described as of yet. Our objective is to develop a population pharmacokinetic model and limited sampling strategy (LSS) for INH to be used in model-based TDM.
Methods: We developed one population PK model with a mixture model for clearance and one model without. Model development in NONMEM v7.4 was based on INH PK data from three studies performed in pulmonary TB patients. These studies included: 1) 14 (127 observations) patients from a Dutch TDM setting [9], 2) 96 (868 observations) patients from South Africa and Tanzania [10], and 3) 59 (591 observations) patients from Tanzania [11]. In total, 169 patients and 1586 observations were included in the analysis.
Two LSSs were tested, one with 2, 4 and 6 h sampling and the other with 2 and 4 h sampling. These LSSs were chosen based on practical considerations, available sampling times in the data, and agreement with the LSS developed for rifampicin [8]. The model-predicted AUC0-24 of the two LSSs was compared with the prediction on the full dataset for both models. Bias and precision were assessed using the mean error (ME) and root mean square error (RMSE) [12], both expressed as a percentage of the mean model-predicted AUC0-24 on the full dataset. We set the target for the ME and RMSE at <20% which we regard as reasonable and is commonly used [13].
Results: The PK of INH was best described by a two-compartmental model in addition to five transit absorption compartments and a well-stirred liver compartment. A priori allometric scaling based on fat free mass was applied on the volume and flow parameters using fixed exponents of 1 and 0.75 respectively. For the model including a mixture on clearance, two subgroups (fast and slow) were implemented to incorporate the polymorphic acetylation of INH. The proportion of fast acetylators in the population was estimated at 38.2% and typical clearance was two-fold higher than for the slow acetylators population. In both models, IIV was included on central volume of distribution (18CV%), clearance (with mixture: 34CV%; without mixture: 49CV%), and absorption rate (62CV%) and was assumed to be log-normally distributed. Values of model parameter estimates were in agreement with those reported in previous works. The relative standard error of the estimates as computed by the covariance step was lower than 15% for all parameters of both models except for the proportion of fast acetylators (23%) in the mixture model and the IIV on the central volume of distribution for both with (20%) and without (18%) mixture model. Goodness-of-fit plots and visual predictive checks showed good performance of both models, while individual plots showed that a few individuals still had a poor fit.
Using the mixture model, the 2, 4 and 6 h LSS (ME: 7.3%; RMSE: 21.5%) performed better than the 2 and 4 h LSS (ME: 0.2%; RMSE: 31.3%). Interestingly, when the mixture model was taken out, the 2, 4 and 6 h LSS (ME: 10.5%; RMSE: 22.8%) performed slightly worse, and the 2 and 4 h LSS (ME: 8.3%; RMSE: 28.8%) slightly better compared to when a mixture model was included.
Conclusions: The developed population models described the PK of INH well. However, the predictive performance of the models for the purpose of TDM using LSSs does not yet reach the target. The LSS using fewer samples performed better without the presence of a mixture model. This could indicate that the model was unable to identify the mixtures using this LSS. We expect that the addition of acetyl-INH metabolite data might help to improve the identifiability of the mixture model and predictive performance of the model for TDM purpose. In conclusion, we show a promising model-based TDM method for INH which still has potential for improvement.
References:
[1] Wilkins, J.J., et al., Variability in the population pharmacokinetics of isoniazid in South African tuberculosis patients. Br J Clin Pharmacol, 2011. 72(1): p. 51-62.
[2] Parkin, D.P., et al., Trimodality of isoniazid elimination: phenotype and genotype in patients with tuberculosis. Am J Respir Crit Care Med, 1997. 155(5): p. 1717-22.
[3] Alsultan, A. and C.A. Peloquin, Therapeutic drug monitoring in the treatment of tuberculosis: an update. Drugs, 2014. 74(8): p. 839-54.
[4] Magis-Escurra, C., et al., Therapeutic drug monitoring in the treatment of tuberculosis patients. Pulmonary Pharmacology & Therapeutics, 2012. 25(1): p. 83-86.
[5] Jayaram, R., et al., Pharmacokinetics-Pharmacodynamics of Rifampin in an Aerosol Infection Model of Tuberculosis. Antimicrobial Agents and Chemotherapy, 2003. 47(7): p. 2118-2124.
[6] Jayaram, R., et al., Isoniazid pharmacokinetics-pharmacodynamics in an aerosol infection model of tuberculosis. Antimicrob Agents Chemother, 2004. 48(8): p. 2951-7.
[7] Gumbo, T., et al., Concentration-dependent Mycobacterium tuberculosis killing and prevention of resistance by rifampin. Antimicrob Agents Chemother, 2007. 51(11): p. 3781-8.
[8] van Beek, S.W., et al., Personalized Tuberculosis Treatment Through Model-Informed Dosing of Rifampicin. Clin Pharmacokinet, 2019. (epub ahead of print)
[9] Magis-Escurra, C., et al., Population pharmacokinetics and limited sampling strategy for first-line tuberculosis drugs and moxifloxacin. Int J Antimicrob Agents, 2014. 44(3): p. 229-34.
[10] Boeree, M.J., et al., High-dose rifampicin, moxifloxacin, and SQ109 for treating tuberculosis: a multi-arm, multi-stage randomised controlled trial. Lancet Infect Dis, 2017. 17(1): p. 39-49.
[11] Aarnoutse, R.E., et al., Pharmacokinetics, Tolerability, and Bacteriological Response of Rifampin Administered at 600, 900, and 1,200 Milligrams Daily in Patients with Pulmonary Tuberculosis. Antimicrob Agents Chemother, 2017. 61(11).
[12] Sheiner, L.B. and S.L. Beal, Some suggestions for measuring predictive performance. Journal of Pharmacokinetics and Biopharmaceutics, 1981. 9(4): p. 503-512.
[13] Ting, L.S., E. Villeneuve, and M.H. Ensom, Beyond cyclosporine: a systematic review of limited sampling strategies for other immunosuppressants. Ther Drug Monit, 2006. 28(3): p. 419-30.
Reference: PAGE 28 (2019) Abstr 8942 [www.page-meeting.org/?abstract=8942]
Poster: Drug/Disease Modelling - Infection