Comparing the Predictive Performance of Machine Learning Models with a Newly Developed Population Pharmacokinetic Model for Busulfan in Paediatric Patients
- 1Department of Drug Design and Pharmacology, Faculty of Health and Medical Sciences, University of Copenhagen, Universitetsparken 2, 2100 Copenhagen, Denmark
- 2Department of Paediatrics and Adolescent Medicine, Copenhagen University Hospital, Rigshospitalet, Copenhagen, Denmark
- 3Institute of Clinical Medicine, Faculty of Medicine, University of Copenhagen, Copenhagen, Denmark
Abstract
Background and Objectives: Busulfan, used in conditioning regimens prior to haematopoietic stem cell transplantation, has a narrow therapeutic index; therefore, due to pharmacokinetic variability, therapeutic drug monitoring (TDM) is employed to ensure drug exposure levels associated with optimal clinical outcomes. This study aimed to facilitate dose adjustment and potentially reduce the number of samples required for TDM in clinical practice.
Methods: Population pharmacokinetic (popPK) and machine learning (ML) models (Random Forest and XGBoost) were developed using a local, retrospective, extensive-sampling dataset of 67 children (637 observations, Denmark) and externally evaluated using prediction error (PE) on a TDM dataset from another centre with 171 children (1871 observations, Netherlands). Limited sampling schedules were evaluated for both methods.
Results: A one-compartment model was the best data fit, showing time-varying clearance decreases of 10.4% and 13% on Days 2 and 3, respectively, compared to Day 1. The popPK, Random Forest, and XGBoost models performed well on the external dataset for concentration prediction, with median PE ranging from −4.82 to −7.44%, median absolute PE of approximately 15%, and approximately 80% of data predicted within ±30% of the PE. Sampling strategies at 0, 0.5, 1, 4 h or 0, 1, 2, 4 h post-infusion adequately predicted busulfan plasma concentration and exposure using the popPK model.
Conclusions: A new popPK model was developed to support individualised dose adjustments in children using fewer blood samples. Both Random Forest and XGBoost could predict busulfan's plasma concentration with performance like that of the popPK model (population prediction). However, a larger dataset is needed for ML methods to describe a complete pharmacokinetic profile.
Supplementary Information: The online version contains supplementary material available at https://doi.org/10.1007/s40262-026-01702-9.
Related
- Comparing Machine Learning and Population Pharmacokinetic Approaches for Drug Concentration Prediction in Dense and Sparse Data Settings
- A unified population pharmacokinetic model of amikacin in paediatric patients from birth to 18 years for model-informed precision dosing.
- Population pharmacokinetic analyses comparing immediate-release levodopa/carbidopa, controlled-release levodopa/carbidopa, and IPX203. Findings from a phase 1 study in healthy volunteers
- The FINRISK cardiovascular risk calculator: Strong predictive performance in a Finnish population
- Population pharmacokinetic modeling of adalimumab biosimilar SCT630 and Humira® in healthy participants and patients with moderate to severe plaque psoriasis to establish PK similarity
BibTeX
@article{yang2026comparing,
title = {{Comparing the Predictive Performance of Machine Learning Models with a Newly Developed Population Pharmacokinetic Model for Busulfan in Paediatric Patients}},
author = {Tianwu Yang and Maria Thastrup and Tania Nicole Masmas and Hilde Hylland Uhlving and Jacob Nersting and Malene Johanne Petersen and Anna Sofie Buhl Rasmussen and Kjeld Schmiegelow and Trine Meldgaard Lund},
journal = {Clinical Pharmacokinetics},
year = {2026},
month = sep,
doi = {10.1007/s40262-026-01702-9},
url = {https://doi.org/10.1007/s40262-026-01702-9},
note = {CC BY-NC 4.0},
}
RIS
TY - JOUR
TI - Comparing the Predictive Performance of Machine Learning Models with a Newly Developed Population Pharmacokinetic Model for Busulfan in Paediatric Patients
AU - Tianwu Yang
AU - Maria Thastrup
AU - Tania Nicole Masmas
AU - Hilde Hylland Uhlving
AU - Jacob Nersting
AU - Malene Johanne Petersen
AU - Anna Sofie Buhl Rasmussen
AU - Kjeld Schmiegelow
AU - Trine Meldgaard Lund
PY - 2026
DA - 2026/09/22
JO - Clinical Pharmacokinetics
DO - 10.1007/s40262-026-01702-9
UR - https://doi.org/10.1007/s40262-026-01702-9
AB - Background and Objectives: Busulfan, used in conditioning regimens prior to haematopoietic stem cell transplantation, has a narrow therapeutic index; therefore, due to pharmacokinetic variability, therapeutic drug monitoring (TDM) is employed to ensure drug exposure levels associated with optimal clinical outcomes. This study aimed to facilitate dose adjustment and potentially reduce the number of samples required for TDM in clinical practice. Methods: Population pharmacokinetic (popPK) and machine learning (ML) models (Random Forest and XGBoost) were developed using a local, retrospective, extensive-sampling dataset of 67 children (637 observations, Denmark) and externally evaluated using prediction error (PE) on a TDM dataset from another centre with 171 children (1871 observations, Netherlands). Limited sampling schedules were evaluated for both methods. Results: A one-compartment model was the best data fit, showing time-varying clearance decreases of 10.4% and 13% on Days 2 and 3, respectively, compared to Day 1. The popPK, Random Forest, and XGBoost models performed well on the external dataset for concentration prediction, with median PE ranging from −4.82 to −7.44%, median absolute PE of approximately 15%, and approximately 80% of data predicted within ±30% of the PE. Sampling strategies at 0, 0.5, 1, 4 h or 0, 1, 2, 4 h post-infusion adequately predicted busulfan plasma concentration and exposure using the popPK model. Conclusions: A new popPK model was developed to support individualised dose adjustments in children using fewer blood samples. Both Random Forest and XGBoost could predict busulfan's plasma concentration with performance like that of the popPK model (population prediction). However, a larger dataset is needed for ML methods to describe a complete pharmacokinetic profile. Supplementary Information: The online version contains supplementary material available at https://doi.org/10.1007/s40262-026-01702-9.
SN - 0312-5963
SN - 1179-1926
ER -