2023 - A Coruña - Spain

PAGE 2023: Drug/Disease Modelling - Other Topics
Yannick Hoffert

An automated multi-model selection algorithm to improve precision dosing of tacrolimus in liver, lung, and bowel transplant recipients

Yannick Hoffert (1); Benedict T. K. Vanlerberghe (2); Dirk Kuypers (3,4); Robin Vos (5,6); Tim Vanuytsel (5,7); Jef Verbeek (5,7); Erwin Dreesen (1)

(1) Department of Pharmaceutical and Pharmacological Sciences, KU Leuven, Leuven, Belgium; (2) Maastricht University Medical Center, Maastricht, The Netherlands; (3) Department of Microbiology, Immunology and Transplantation, KU Leuven, Leuven, Belgium; (4) Department of Nephrology and Renal Transplantation, University Hospitals Leuven, Leuven, Belgium; (5) Department of Chronic Diseases and Metabolism, KU Leuven, Leuven, Belgium; (6) Department of Respiratory Diseases, University Hospitals Leuven, Leuven, Belgium; (7) Department of Gastroenterology and Hepatology, University Hospitals Leuven, Leuven, Belgium

Objectives:

Tacrolimus is a cornerstone immunosuppressant for preventing graft rejection after solid organ transplantation. It is characterized by highly variable pharmacokinetics (PK) and a narrow therapeutic window. Model-informed precision dosing (MIPD) using a population PK model has been suggested to personalize tacrolimus dosing.1 Tacrolimus exposure targets were defined by graft type and the duration of tacrolimus therapy, with trough concentration (TC) targets ranging between 5-25 ng/ml.2 Numerous population PK models have been developed to describe the tacrolimus PK, but their predictive performance in the diverse population of solid organ transplant patients is unknown. In this work, we developed an automated multi-model selection Bayesian forecasting algorithm for MIPD of tacrolimus and compared its predictive performance to that of randomly selected models.

Methods: 

Data of transplant recipients were obtained from retrospective, single-center cohort studies. Three settings were investigated, using a Bayesian forecasting algorithm with a single “historical” TC (TC-1, TC-2, or TC-3; recent to older) to estimate every individual patient’s PK parameter and predict a consecutively collected “future” TC (TC0) under the administered dosing regimen.

An automated multi-model selection Bayesian forecasting algorithm3,4 was compared to Bayesian forecasting with a randomly selected model. The included models were encoded in NONMEM (version 7.5.0) and the multi-model algorithm was implemented in R (version 4.2.2). For each patient and Bayesian forecasting setting, a weight was calculated for every model based on its maximum likelihood (ML) in relation to the sum of MLs of all models. The model with the highest weight was selected for Bayesian forecasting. For each setting, the average prediction error of all n patients was calculated as: ∑(i=1)n[(|Predictioni-Observedi|)/Observedi]×100%. To emulate the random selection of a model for Bayesian forecasting, the average prediction error of all nine single model approaches was calculated. A prediction error below 25% was arbitrarily considered clinically acceptable.

Results: 

An intermediate analysis was performed using data from 15 patients, including five liver, five lung, and five bowel transplant recipients. For liver transplant patients, samples were collected between postoperative day (POD) 6 and 24, with sampling intervals ranging between 5 to 9 days. For lung transplant patients, samples were collected between POD 49 and 215, with 14 to 59 days between consecutive samples. For bowel transplant patients, samples were collected between POD 1538 and 5732, with sampling intervals ranging between 28 to 167.

The model repository so far included nine population PK models:

Model                  (Graft,         POD)

Model 1.5           (liver,           mean 23)

Model 2.6           (liver,           mean 323)

Model 3.7           (liver,           mean 21)

Model 4.8           (lung,           median 86)

Model 5.9           (lung,           mean 36)

Model 6.10          (kidney,       until 365)

Model 7.11          (kidney,       mean 162)

Model 8.12          (kidney,       until 21)

Model 9.13          (heart,         mean 12)

The precision of the predictions was presented as the median [1st quartile; 3rd quartile] prediction error (%):

Setting        Automated multi-model selection         Random model selection

TC-1              14.8% [6.9%; 37.5%]                                 34.0% [9.2%; 67.7%]

TC-2              20.3% [7.0%; 53.0%]                                 46.9% [19.7%; 79.2%]

TC-3              23.2% [8.3%; 52.1%]                                 46.7% [18.7%; 87.1%]

Bayesian forecasting using the automated multi-model selection algorithm resulted in clinically acceptable prediction errors that were systematically lower as compared to randomly picking a model. The prediction error was lowest when using a most recent historical samples (TC-1 < TC-2/TC-3). The prediction error was lowest in the bowel cohort (14.8% [10.2%; 16.2%] when using TC-1), although no population PK model of tacrolimus in bowel transplant patients was included.

Conclusions:

Our results indicate a more precise MIPD of tacrolimus with an automated Bayesian forecasting model selection algorithm compared to Bayesian forecasting with a randomly picked model.



References:

  1. Størset, E. et al. Transplantation 99, 2158–2166 (2015).
  2. Staatz, C.E. et al. Clin Pharmacokinet43, 623–653 (2004)
  3. Kantasiripitak, W. et al. CPT Pharmacometrics Syst Pharmacol 11, 1045–1059 (2022).
  4. Uster, D. W. et al. Clin. Pharmacol. Ther. 109, 175–183 (2021).
  5. Chen, B. et al. Journal of Clinical Pharmacy and Therapeutics 42, 679–688 (2017).
  6. Staatz, C. E., Willis, C., Taylor, P. J., Lynch, S. V. & Tett, S. E. Liver Transplantation 9, 130–137 (2003).
  7. Zhu, L. et al. J Pharm (Cairo) 2014, 713650 (2014).
  8. Chen, W. et al. The Journal of Clinical Pharmacology 62, 1310–1320 (2022).
  9. Cai, X. et al. European Journal of Pharmaceutical Sciences 152, 105448 (2020).
  10. Andreu, F. et al. Clin Pharmacokinet 56, 963–975 (2017).
  11. Vadcharavivad, S. et al. Journal of Clinical Pharmacy and Therapeutics 41, 310–328 (2016).
  12. Han, N. et al. Basic & Clinical Pharmacology & Toxicology 114, 400–406 (2014).
  13. Gong, Y. et al. Eur J Hosp Pharm 27, e12–e18 (2020).
Reference: PAGE 31 (2023) Abstr 10642 [www.page-meeting.org/?abstract=10642]
Poster: Drug/Disease Modelling - Other Topics
Top