Meet the Expert: Statistics

As part of our ‘Meet the Expert’ series, Strategy Analyst Alex Girling interviews Dr Margaret Jones, SVP Head of Statistics, and Professor Andy Grieve, Senior Drug Development Statistician at Weatherden. 

Briefly describe your background and your role here at Weatherden.

Margaret Jones
I joined Weatherden a year ago after 25 years in the pharmaceutical industry, following an early career in academic oncology research. At Pfizer, I had the privilege of working with Andy on a complex Bayesian adaptive trial, sparking my interest in Bayesian and adaptive methodologies. I later held roles at Lilly, Takeda, and UCB, where I spent a decade advancing Bayesian approaches across drug development, from preclinical to early clinical trials and beyond. 

 
Andy Grieve
I officially retired in April after 49 years in the industry, with a four-year break as a Professor of Biostatistics at King’s College London. My experience spans all areas of pharmaceutical R&D, from drug discovery and preclinical toxicology to all phases of clinical development (Phases I-IV), manufacturing, and pharmacoeconomics. I've been advocating for Bayesian methods in drug development since 1981. Now, at Weatherden, I consult on projects and introduce new ideas. 

For the layperson in a nutshell, can we describe the different types of statistics used in clinical trials and what the differences are?

Andy Grieve 
The traditional approach, known as null hypothesis significance testing, starts with the assumption that a drug has no effect compared to a placebo or standard of care. Data is then collected, and if the results are extreme enough, this assumption is rejected, allowing researchers to claim statistical significance. This method has been the standard since the 1970s and remains widely used today. 

The Bayesian approach, on the other hand, incorporates prior knowledge or beliefs about a drug’s efficacy before the study begins. As new data are collected, these beliefs are updated, creating a posterior distribution, a refined estimate of the treatment effect. This allows for more dynamic decision-making, such as calculating the probability that a drug is better than the standard of care by a certain margin. Rather than testing a fixed hypothesis, Bayesian methods continuously refine understanding through data, shifting from a deductive to an inductive approach. 

Margaret Jones 
Prior beliefs can apply not just to the treatment effect but also to the standard of care. The key question, and often the source of controversy in Bayesian statistics, is how to develop a prior. While some assume it's purely subjective, priors can actually be built from existing data before running a new experiment or clinical trial. There are many analytical methods to do this systematically. 

So this is perhaps a good segue to think about, in the broader context of drug development, how and why is this so important? 

Andy Grieve 
People are naturally resistant to change. In pharma, we often heard, “Why switch to Bayesian methods? The old way has worked for decades.” But at the start of the new millennium, the industry was struggling, yet many clung to outdated methods with rose-tinted nostalgia. Our role is to challenge that mindset, disrupt the status quo, and show there’s a better way forward. 

 
Margaret Jones 
Drug development is a continuous process of building on existing knowledge, from preclinical stages through Phase 1 and beyond. The Bayesian approach naturally aligns with this, allowing for more efficient use of prior information in each new study. 

Moreover, many assumptions in classical analysis align more closely with Bayesian principles than people realize. For example, when most interpret a confidence interval, they’re thinking in terms of a Bayesian credible interval, the probability that the true effect lies within a given range. Once explained, the Bayesian framework often feels more intuitive, as it better reflects how we naturally think about uncertainty and evidence. 

So for the layperson, looking at a clinical trial or interpreting statistics, traditionally they'd focused just on a P value and if it’s significant or not, but your key takeaway in a Bayesian study is more about the credible interval? 

Andy Grieve 
A credible interval is a fixed range where there’s a high probability that the true parameter value lies, distinguishing it from a confidence interval. A confidence interval, instead, means that if we repeated the experiment many times, 95% of those intervals would contain the true value, but for any single interval, we don’t know whether it does or not. 

This distinction is crucial because, in practice, most scientists interpret confidence intervals as if they were Bayesian credible intervals, believing they represent the probability of the true value being within the range, when that’s actually a Bayesian perspective. 

 
Margaret Jones 
Exactly. With a Bayesian approach, you're not just getting a single interval, you're estimating the entire posterior distribution. This allows you to extract detailed probability statements beyond just “better than placebo” or “better than standard of care.” 

Instead, you can assess the probability that the drug exceeds a specific threshold, such as outperforming a competitor or meeting the Target Product Profile (TPP). This provides a much more meaningful and actionable assessment of whether the drug is likely to achieve its intended clinical and commercial goals. 

 
Is Bayesian Statistics gaining more traction and being more widely utilised in a particular area? 
Margaret Jones 
Bayesian statistics are gaining momentum, particularly in early, phase proof-of-concept (PoC) studies. Many companies now use Bayesian methods to incorporate external data, such as placebo or standard of care information, which helps reduce sample sizes while maintaining robust conclusions. 

A key area where this is especially valuable is paediatric extrapolation. By leveraging existing adult data, Bayesian approaches allow for a more efficient assessment of drug effects in paediatric populations. Regulatory agencies are recognizing this potential, with new guidance on Bayesian methods for paediatric studies expected soon. 

 
Andy Grieve 
Bayesian methods have also seen widespread adoption in oncology, particularly in platform, basket, and umbrella trials. These adaptive designs allow for greater flexibility and efficiency in drug development, an area that is rapidly expanding, accelerated in part by shifts in trial methodology during the pandemic. 

A key turning point was the BioNTech & Pfizer COVID,19 vaccine trial, which used a Bayesian design. Its success and regulatory approval demonstrated the practical value of Bayesian approaches, helping to solidify their credibility and drive broader adoption across the industry. 

The recent ICH Draft Guidance on Adaptive Trials does reference Bayesian Methods and we intend to share a more detailed discussion of the guidance in an up-coming post. 

 
So it's something that early-stage biotech needs to be thinking about a lot more. 
Margaret Jones

One of the key advantages of Bayesian and adaptive approaches is the ability to design trials that directly address specific research questions, rather than forcing a program into a rigid, off-the-shelf design. Despite the availability of proven methodologies and regulatory precedents, this flexibility remains underutilized. 

By tailoring trial designs, sponsors can optimize the information gained, ensuring they answer the most clinically and regulatory relevant questions. This leads to more efficient trials, better decision, making, and ultimately, stronger drug development strategies. 

Andy Grieve
A further benefit for the biotech industry is that Bayesian methods provide a greater variety of information and of greater quality which will support them in selling their product to Big Pharma. The ability to produce more positive information utilising Bayesian methods and the adaptive design approach is a big bonus for biotech if they are considering at the end of phase two selling off the asset to Big Pharma. 

 
Margaret Jones
One historical barrier to Bayesian and adaptive approaches has been the time required for simulations and trial setup. These designs demand careful planning and extensive modelling, which often took months to execute. However, with advancements in computing power, what once required months can now be done in days or even hours. 

While thorough documentation and a well-structured simulation plan are still essential for regulatory acceptance, the computational burden is no longer the obstacle it once was. This has made Bayesian and adaptive designs far more practical and accessible, accelerating their adoption across drug development. 

And what about cost? Does it cost much more? 
Margaret Jones
Sometimes it costs less, often you're making a significant reduction in the sample size of the study because you're using external placebo data, for example leveraging external placebo data can significantly cut down the number of patients required in the placebo arm, sometimes by half. This not only lowers costs but is also an ethical advantage, as fewer patients receive placebo treatment. 

In many cases, these efficiencies translate into an overall 25% reduction in study size, making trials more cost, effective. While the upfront investment in simulations and setup remains, the decreasing computational overhead means that, more often than not, these designs can offset costs and streamline development. 

Finally, Margaret and Andy, what are the top advantages of using Bayesian statistics that everyone should be thinking about? 
Andy Grieve
Well, I think the use of all available information, that's crucial. I think we will increasingly be using real world evidence. Information harvesting and utilising it together with the prospective data from a planned clinical trial and combining it using utilising Bayesian methods. I think that's great. I don't think we'll end up with wholly real-world evidence packages to register new drugs, but I think there is evidence they can be combined. I think that's important. I think in the planning of a whole development program, there is greater scope for the use of Bayesian methods. 
That is something that Margaret and I have been working on for the last six months or so within Weatherden and how we can help small biotechs make decisions about their development programmes as a whole, rather than just concentrating on a single clinical trial. And finally, I do think it's because it's a much richer environment to be able to present data and conclusions than compared to the traditional way of doing it. 

 
Margaret Jones    
And I'd add in that I think the availability of data, through sharing initiatives within pharma, for example patient level data on control arms, really helps to give access to much better information. So, the availability of data to build robust priors for both analysis and design is infinitely better than it was 20 years ago. 
That, and Andy's touched on the better understanding of the clinical trial. So, I think there are Bayesian add-ons to a frequentist trial, so you can have a standard clinical trial design, but there are still Bayesian approaches that you can use within that to better understand the properties of the trial, such as the probability of success. 

Curious about how statistics fits into your strategy? Get in touch to learn how our experts can give your drug the best chance of success in clinical trials.