Last week I attended the PAGE conference in Montreux, Switzerland. PAGE conference organizers love beautiful scenery and parties at least as much as they love the scientific program and this year was no different. Montreux is a gem of a Swiss town overlooking Lake Geneva and the Swiss Alps. It is also the final resting place of one my favorite writers, Vladimir Nabokov, as well as his son Dmitri and his wife Vera.
There were several activities related to Bayes and Stan at the conference. Here is a quick summary.
- Generable presented a poster PK/PD Inference in Stan showing the use of Stan for pharmacometrics models using a two compartment model and demonstrating the effectiveness of the NUTS sampler.
Bill Gillespie gave his now customary one-day workshop entitled Advanced Use of Stan, RStan, and Torsten for pharmacometrics applications.
Eunjung Song presented a poster entitled Bayesian estimation of parameters in the pharmacokinetic model. This study developed an alternative TDM package based on RStan, which can be comparable predictability to the previous software packages such as the Abb ott PKS system.
Sebastian Weber presented a poster Supporting drug development as a Bayesian in due time?! This poster demonstrated near linear speed-ups using MPI on per-patient ODE systems.
- Sebastian Wicha presented a poster Handling inter-occasion variability in model implementation for Bayesian forecasting: A comparison of methods and metrics.
- Elvira Erhardt presented a poster Bayesian knowledge integration for an in vitro–in vivo correlation (IVIVC) model
- Felix Held presented a poster Bayesian hierarchical model of oscillatory cortisol response during drug intervention
Paolo Magni presented a poster Evaluation of software tools for Bayesian estimation on population models: an update based on current software versions. In this paper, Paulo and collaborators compared the performance of NONMEM 7.4.1, WinBUGS 1.4.3 (with BlackBox Component Builder 1.5 and WBDiff interface), Stan 2.17, and JAGS 4.3 (with R packages rstan and rjags). We were surprised to see some inconsistent results coming from Stan’s ODE solvers (which had a bug in the version that was being tested). We just received the code they used to fit the model and will run the system again to see if the problem had been resolved.
Ron Keizer from InsightRX gave a talk on Experiences in applied clinical pharmacometrics: challenges, recommendations, and research opportunities. This was particularly interesting to me, as they are focusing on putting pharmacometrics models in the hands of decision makers.
Scott K Pruitt’s talk entitled Clinical Overview of Immunotherapy in Oncology stood out. In his words:
A major challenge remains over how, in Phase 1b/2 studies, to identify promising combinations that have enhanced efficacy over PD-1/PD-L1 blockage alone and therefore warrant further clinical investigation. It is also not clear if the unique paradigm of accelerated approval based on single arm trial data with a subsequent confirmatory pivotal study will be acceptable for such combinations of multiple agents or whether the “combination rule” will require larger multiple arm studies that in turn would delay access of dying cancer patients to potentially effective immunotherapy combination regimens.
Combination trials are challenging for traditional methods of analysis as they effectively reduce the size of the per-treatment sample, but they are natural for Bayesian analysis where we can take advantage of the hierarchical nature of the model to learn both local and average effects.