Busulfan is used as part of conditioning regimens for allogeneic hematopoietic stem cell transplantation and is characterized by a narrow therapeutic index and notable inter- and intra-individual pharmacokinetic variability. Because exposure correlates with both efficacy and toxicity, therapeutic drug monitoring (TDM) is commonly applied to guide dosing and achieve target systemic exposure, typically expressed as area under the concentration-time curve (AUC). For once-daily (QD) intravenous busulfan there has been no clear consensus on an optimal limited sampling strategy (LSS) suitable for routine TDM and model-informed dosing.
The investigators generated a virtual adult cohort (n = 1000) by Monte Carlo simulation to evaluate LSS performance for QD intravenous busulfan. The simulated dosing regimen was 3.2 mg/kg administered as a 3-hour infusion. Simulations were informed by three published busulfan population pharmacokinetic (PK) models. For each candidate one- and two-sample LSS, Day-1 sample(s) were used to estimate individual PK parameters by maximum a posteriori (MAP) Bayesian estimation; those parameter estimates were then used to predict the cumulative AUC over 4 days.
Three previously published busulfan population PK models were used as the structural basis for Monte Carlo simulations and posterior Bayesian estimation. The source report lists the use of three models but does not detail their individual structural forms or parameter values in the abstract. The evaluation intentionally assessed performance across multiple published models to account for model uncertainty in real-world practice.
The analysis compared multiple one-sample and two-sample LSS designs sampled at different times relative to the start of the 3-hour infusion on Day 1. Design performance was measured by how well Day-1-based parameter estimates could recover the 4-day cumulative AUC and by the downstream impact on target attainment when used to guide dose adjustments via model-informed precision dosing (MIPD).
A MIPD analysis quantified the clinical utility of TDM implemented with each LSS design. Specifically, the analysis evaluated how TDM-driven dose adjustments based on Day-1 samples impacted cumulative target attainment over the 4-day course, comparing the relative benefit of different sampling schedules.
Across all three population PK models, a two-sample early-plus-late schedule provided the most informative data for Bayesian parameter estimation and prediction of cumulative exposure. The recommended two-sample LSS was sampling at 3 hours and 7 hours after start of infusion. When TDM was guided by a well-chosen two-sample LSS and followed by MIPD-guided dose adjustment, cumulative target attainment was approximately doubled compared with no TDM.
A single early 3-hour sample performed markedly worse than the two-sample designs in both AUC recovery and target-attainment benefit. The investigators also identified a single 6-hour sample as a simpler but less robust alternative to the two-sample schedule.
The best-performing designs identified in simulation were validated in an external cohort of adults and adolescents (n = 37). In this cohort, two-sample LSS designs recovered the 4-day cumulative AUC with median absolute percentage error (MAPE) of approximately 5%–8%, whereas a single 3-hour sample produced MAPE of approximately 11%–13%. These external results support the simulation finding that the two-sample 3- and 7-hour schedule provides substantially improved prediction accuracy versus a single early sample.
For once-daily intravenous busulfan administered as a 3-hour infusion in adults and adolescents, the evidence from simulation and limited external validation supports using a two-sample LSS with samples at 3 and 7 hours after infusion start to inform Bayesian estimation and MIPD-guided dose adjustments. Implementing this two-sample approach for Day-1 TDM is associated with improved recovery of cumulative AUC and approximately doubles cumulative target attainment compared with no TDM; it substantially outperforms a single 3-hour sample. A single 6-hour sample may be considered when logistics limit multiple samples, but it is a less robust choice.
The abstract reports the use of three published population PK models, Monte Carlo simulation for a virtual cohort (n = 1000), and validation in an external cohort (n = 37). The abstract does not report detailed model structures, exact sampling times evaluated beyond those highlighted, numeric definitions of target attainment thresholds, formal statistical testing results, or safety and clinical outcome data linked to target attainment. Those details were not reported in the abstract and would require consulting the full text for complete methodology and results.