Bayesian forecasting versus clinical practice for intravenous busulfan dose adjustment in paediatric stem cell transplantations

Aims: Haematopoietic stem cell transplantation (HSCT) is a vital therapy that is used to cure high-risk or relapsed malignant and nonmalignant conditions. Busulfan is an important chemotherapeutic agent used in HSCT, particularly in children with cancer. It has been demonstrated that optimal exposure over four days reduces the incidence of toxicities and decreases the risk of transplant failure. Current local clinical practice uses intensive sampling on each of the four days of dosing to calculate cumulative exposue over the dosing period. This study aimed to evaluate the predictive performance of two Bayesian forecasting programs under various sampling scenarios to estimate an individual's busulfan exposure to individulise doses and attain optimal busulfan exposure.

Methods: Pediatric oncology patients underwent repeated blood sampling seven times daily (samples taken at pre-dose trough, 3h after beginning infusion, 3:15h, 4h, 5h, 6h and 8h) to determine the patient's overall drug exposure throughout the conditoning regimen at the Lady Cilento Children's Hospital, Brisbane, Australia from 2012 to 2016. Patient characteristics, busulfan concentrations and sampling times were collected to estimate the daily area under the concentration-time curve (AUC) utilising two Bayesian forecasting programs (NextDose®, InsightRX®) and comparing these to the true measured AUC. Various sampling scenarios were tested using the programs including a full profile over all four days compared to reduced sampling strategies.

Results: Thirty-two children, median age 5.5 years, contributed 720 busulfan concentrations resulting in 93 true AUC calculations via the trapezoidal rule. The mean true daily AUC was 20.1 (8.83-41.81) mcg∙hr∙mL-1. One child had a true AUC observation on day 1 only, seven children on the first 2 days, five children on days 1-3 and twelve children on all 4 days of the regimen. Of those remaining, one had a true AUC on day 3, four children had 2 days of true AUCs and two children had 3 days of true AUCs. The estimated AUC povided by the two Bayesian forecasting programs resulted overall in low impression and high accuracy compared to the true AUC under scenario 1-3 (see Table 1) over all 4 days of busulfan dosing.

Table 1: Predictive performance of two Bayesian forecasting programs estimating daily busulfan AUCs

Conclusion: While Bayesian forecasting programs have been utilised in other countries, the Queensland paediatric HSCT unit has not yet adopted this practice. Balancing reduction in sampling and predictive ability of the Bayesian forecasting tools during further investigations may offer great potential for use of these programs in local clinical practice.