Pharmacometrics Enhanced Bayesian Borrowing
Objectives
The Pharmacometrics Enhanced Bayesian Borrowing (PEBB) approach uses historical data to build models, projects the outcome of future clinical trials and borrows information from projections to improve the efficiency of target trials (TT: trial to be informed by PEBB). Here we illustrate how PEBB could improve the efficiency of clinical trials.
Methods
The data used to illustrate the PEBB was generated in trials from the development program of empagliflozin in Type 2 Diabetes Mellitus patients. A Phase III trial was selected as the TT. Historical trials (conducted before TT) should contain PK/PD samples and covariate measurements from the same nature as TT. The endpoint to be informed through PEBB was the placebo corrected change in HbA1c from baseline to week 24 in patients receiving either 10 mg, or 25 mg of empagliflozin daily.
Pharmacokinetic/pharmacodynamic modeling was performed using NONMEM 7.4.3 [1]. Simulations and post-processing was performed in R version 3.5.2 [2].
A population pharmacokinetic/pharmacodynamic model was developed based on data from Phase I-II trials. This model included covariates on both the PK and PD models.
The framework follows an objective study design and is applied in 2 stages:
Design Phase: performed before TT data are available. Simulations are performed considering uncertainties (i.e., distribution of covariates observed in historical data, parameter uncertainty, etc.) to obtain the prior of the mean outcome per arm. The prior is robustified and type I error (TIE) and power are explored for a range of potential outcomes of the TT to define the weight of the uninformative component – w – which is a tuning parameter determining how strongly the prior gets dynamically down-weighted in case of prior data conflict. w was chosen to maximise the ESS and power under the constraint of keeping an acceptable TIE .
Analysis Phase: The prior is obtained after completion of TT recruitment but before unblinding, considering the observed distribution of baseline covariates, and robustified using the w from the design phase. Once outcome data are available, the posterior is computed.
This PEBB was applied on 2 scenarios: unchanged TT, and reduced sample size TT.
Results
In case of unchanged TT and no data conflict, borrowing increased the ESS, but the TT was highly powered and the effect of PEEB on power was marginal. In case of prior data conflict (lower drug effect in the observed data) both ESS and power increased with PEBB. The TIE increased only moderately and was controlled at 10%.
In the analysis phase PEBB achieved an increase in ESS.
In case of a hypothetical reduced (underpowered) TT, borrowing significatively increased the power, allowing to reach sufficient power, while TIE increased only moderately and was controlled at 10%.
Conclusions
The PEBB has the potential to increase the power of clinical trials by leveraging the information from previous trials through population pharmacokinetic modelling and simulation
Citations:
[1] Beal SL, S. L. (2009). NONMEM user’s guides. Ellicott City: Icon Development Solutions.
[2] R Core Team, R Foundation for Statistical Computing. (2021). R: A Language and Environment for Statistical Computing.
Author(s)
- Lucie Fayette, Boehringer Ingelheim Pharma GmbH & Co. KG (Presenting Author)
- Alejandro Perez Pitarch, Boehringer Ingelheim Pharma GmbH & Co. KG (CoAuthor)
- Martin Oliver Sailer, Boehringer Ingelheim Pharma GmbH & Co. KG (CoAuthor)