Bayesian Forecasting

Dose prediction using Bayesian forecasting has been proven to save lives. Evans et al. demonstrated a larger increase in survival for childhood acute lymphoblastic leukaemia, by individualising methotrexate dose, than any other single drug treatment (Evans, Relling et al. 1998). Despite its striking improvement on 5-year survival, it is hardly used because of the difficulties of access to, and usability of, Bayesian forecasting software. Dose individualisation has also been shown to increase survival after busulfan conditioning by maintaining population concentrations within a narrow safe and effective range (Bleyzac, Souillet et al. 2001; Bolinger, Zangwill et al. 2001; Booth, Rahman et al. 2007; Bartelink, Bredius et al. 2009; Abbasi, Vadnais et al. 2011; Hempel and Trame 2011; Long-Boyle et al. 2015). In addition to the busulfan application, LabPlus and clinical staff needed a tool to help make the best use of concentration and biomarker measurements for dose individualisation. Warfarin dose predictions using NextDose with INR measurements have been shown to be accurate across a wider range of daily doses (especially over 7 mg/day) (Holford, Ma, Tsuji 2018) than other Bayesian algorithms (Saffian et al. 2016). NextDose was developed to meet the clinical need and promote more widespread dose individualisation.

The Need

In January 2012, busulfan analysis and reporting with NextDose was demonstrated to the Clinical Director and lab technicians of LabPlus as well as pharmacists, nurses, haematologists and oncologists associated with the SBCC. The clear opportunity for improved clinical care by using Bayesian forecasting for dosing of busulfan was demonstrated.

These Bayesian forecasting methods have not been available, in a clinically useful form, in Auckland until 2012 and it was decided that NextDose should be adopted initially for busulfan dosing. Methods for other drugs such as methotrexate have been added when suitable models have been developed. Its application to methotrexate dosing may be clinically important because of the major improvement in survival that has been shown elsewhere (Evans, Relling et al. 1998). NextDose will continue to be modified and extended in collaboration with LabPLUS and hospital clinical staff. Data collected will be used to improve the models over time.

Technology

NextDose has been designed around a database to provide security and transportability. This approach allows patient data to be stored remotely (to meet hospital security requirements) or locally within the NextDose application. It is extensible to other medicines, and associated observation and reporting types, and has already been extended to support methotrexate dosing. NextDose consists of three software abstraction layers to provide a clear separation between the user interface, model controllers and the modelling software. This modular approach allows the use of different modelling software via the same intuitive interface, which should make these tools more accessible and useable in a clinical environment.

Evolution

The original model used for busulfan was based upon a modification of the PK model developed by FDA (Booth, Rahman et al. 2007). This modification employed a maturation function to help predict doses in infants. In 2014 the model was updated based on a large data set collected in infants, children and adults (McCune et al. 2014). The model was further refined in 2017 to improve the empirical prediction of clearance changes with time..