Statisticians have a crucial role to play in improving the efficiency of clinical trial operations.  As an early-career professional, I’ve developed a keen interest in pursuing practical methods and approaches that can boost productivity and make the day to day work of my colleagues a little easier. In this blog, I’ll discuss the context and benefits of work I produced using PROC MCMC for Bayesian modelling, that I presented at PSI 2019. To realise the promise of Bayesian methods, we also need practical, low-cost tools to allow statisticians to execute the approaches- this  powerful SAS procedure provides one accessible solution. 


I noticed Bayesian modelling becoming more widespread and became intrigued by it as a beneficial alternative to conventional frequentist approaches. Bayesian methods are undoubtedly appealing in a drug development context where patient enrollment is challenging, and clinical data is hard-won.  Bayesian techniques allow us to make better use of historical and accruing data to increase efficiency and maximise knowledge.  There are clear ethical, as well as productivity benefits since fewer patients may need to enrol in the trial, allowing us to gain more accurate results from smaller sample sizes.  I spotted an opportunity to carve out a niche for myself by upskilling in Bayesian approaches, so I took the initiative to undertake research and submit it for presentation at the annual PSI conference. I particularly wanted to produce work that would have a straightforward, real-world application and would serve as a practical guide of how to create and run a Bayesian model.  My abstract was accepted as a poster presentation for the 2019 conference in London.

The approach

The poster, which you can download here, walks you through a set of steps needed to analyse a given standard clinical trial dataset in a Bayesian framework using SAS PROC MCMC.  In the absence of a live project, we created a synthetic dataset and compared two different treatments using a Bayesian analysis method versus a more traditional frequentist approach.  . While it is possible to use Bayesian methods in other SAS procedures such as PROC GENMOD, PROC MCMC gives far greater flexibility and user control over the models and parameters. Indeed, while this method is a little more complicated than alternative PROC approaches, it can efficiently accommodate most Bayesian needs.  Therefore, it is well worth the investment of time for statisticians to get up to speed with the potential of PROC MCMC for this application.

Users of the method do need to have solid background modelling knowledge to apply it correctly.  While I’ve set up the approach to be as simple as possible, it isn’t merely a matter of taking the code and plugging in the numbers. 

However, with a robust basic understanding of Bayesian modelling theory, a statistician can get up and running with the PROC MCMC relatively quickly.  The approach also benefits from being reasonably generalisable. When I later applied the model to some real-life client project work, the overall framework, while requiring some tailoring, was still useful.  Within the poster, we aimed to present as many options within PROC MCMC as possible in the space we had, in order to maximise flexibility for users.  

Promoting practical statistics

First and foremost, I consider myself a practical statistician and am highly motivated to develop pragmatic approaches that can help my industry colleagues in their day to day work.   To this end, I’m currently exploring visualisation techniques to help present data in more useful formats to inform better decision-making. For example, I’ve recently been working on a study where we’ve been producing profile plots that allow us to visualise a particular subject’s timeline and enable exploratory insights. I am hoping to submit work around this topic at a future PSI conference.

If you found this blog useful download the poster Jack Message presented at PSI,  One PROC MCMC away from becoming a Bayesian modelling expert.