The changing landscape of the pharmaceutical statistician

The growth and evolution of pharmaceutical statisticians

Over the past forty years the number of statisticians in the pharmaceutical industry has grown enormously. An excellent, though perhaps somewhat dated, summary of the history of pharmaceutical statisticians may be found in Andy Grieve’s Presidential address to the Royal Statistical Society (2005). In his address, Andy pointed out that membership of PSI (Statisticians in the Pharmaceutical Industry - a predominantly UK organisation), grew from 47 in 1977 to over 1100 in 2004. This growth in numbers has been mirrored in other related disciplines such as statistical programming.

It isn’t purely in numbers that things have changed: the role itself has evolved and continues to do so. From the majority being employed historically in drug discovery we now see that the majority are employed in clinical research; but such a simple description does not adequately describe the variety. Today, pharmaceutical statisticians may be found supporting high throughput screening, microarrays, toxicology, formulation and stability testing, assay validation, genomics, pharmacology, pharmacokinetics, pharmacodynamics, clinical research, manufacturing, post marketing studies, safety surveillance and pharmacoeconomics, to name but a few! Many companies also include a team of statisticians dedicated to researching methodology and innovation.

Navigating the changing regulatory landscape

I think it’s fair to say that one of, if not the main factor, in the evolution of the pharmaceutical statistician role has been the changing regulatory landscape. From the 1960s, the number of laws, regulations and guidelines covering applications for new drug products began to increase globally. Eventually, following initial discussions between Europe, Japan and the US, the International Council for Harmonisation of technical requirements for pharmaceuticals for human use (ICH), was conceived. A quick glance at the ICH website today will provide the reader with a set of guidelines that should be followed, with one in particular close to a clinical trial statistician’s heart: ICH E9 Statistical Principles for Clinical Trials. It would be remiss to think that this is the only guideline covering statistical issues. There are many, ranging from adjusting for baseline covariates to the use of Bayesian statistics. Indeed the UK’s own National Institute for Health and Care Excellence (NICE) provides some excellent guidance regarding network meta-analyses. Guidelines also exist covering the standardisation of data collection, datasets, clinical trial reports and the Integrated Summaries of Effectiveness and Integrated Summaries of Safety.

Computational power and its impact on clinical trials

It may appear that with so many standards, much of the work of a pharmaceutical statistician supporting clinical trials could be automated. This is indeed the case for data such as demographics, etc., but along with the increase in regulations governing drug approval, we have seen an explosion in computing power. This has increased the number and complexity of analysis methods available.

Balancing complexity: The role of bayesian statistics

It wasn’t too long ago that the imputation of missing data or the use of Bayesian designs and analyses would be unthinkable and yet now in early clinical development, adaptive dose finding studies, health technology assessments and increasingly, medical devices and diseases in rare populations and paediatric populations, the use of Bayesian statistics is becoming commonplace. With so much at our fingertips these days, we should not forget the importance of ensuring the basics are covered: plot the data, understand the models and how the data were collected, the methods, the assumptions and the experimental design. In this era of “Big Data”, “Machine Learning” and “Artificial Intelligence” this is particularly true (yes, these may be encountered in pharmaceutical statistics). One of the consequences of having easy access to often quite complex methods, is a “black box” approach to analysis. Whilst this certainly has its merits, users are sometimes unaware of assumptions underpinning the analyses and so do not present their results with the appropriate caveats. One assumption that is often overlooked and not checked, even by the strongest of theoretical statisticians, is that an iterative method such as those used in many Bayesian analyses, has actually converged to a stable solution.

Designing trials with care: The importance of experimental design

The need to cover the basics is not only true of the analysis of trial data, but also its design. Many trials use complex adaptive designs which fundamentally rely on the appropriate randomisation, correct doses, population and endpoints. Sometimes what may be considered a simple stratified randomisation needs to be thought about very carefully. Some companies have no, or very little, experience of crossover designs whereas in others, they are very popular. The basics of experimental design can be forgotten and designs are taken from the shelf simply because they had been used by that department for a long time. I would suggest that there is still a lot that can be learnt from Cox’s Design of Experiments, particularly in those designs that are more complex than a simple 2x2 Crossover.

Strategic decision making in drug development

One will see and hear many presentations discussing the ever increasing cost of developing new drugs. One Contract Research Organisation estimates pivotal Phase III trials cost a median of approximately $41,000 per patient. This increasing cost does not appear to be matched with an increase in the number of drugs being approved. In some companies, statisticians will be involved in the drug development plan from the very beginning, making use of utility analyses and decision theory, helping with strategic decisions.

Exploring the potential of early clinical research and predictive models

In an attempt to reduce the time to market many companies have used adaptive trial designs combining Phase II and Phase III. Sometimes these fail not because of a lack of efficacy or safety issues, but because of rushing into the trial without enough knowledge of the product. Some companies are beginning to take a different approach and focus on more exploratory trials and/or analyses prior to Phase III in order to maximise their knowledge, often developing predictive models during the earlier stages. These can be remarkably complex to derive. Statisticians involved in pre-clinical and early clinical research bring a wealth of knowledge to the development of these models so perhaps we shall see, if not quite a full circle, a return of more pharmaceutical statisticians to the earlier stages of drug development coupled with more thinking time.

About the author

Dr Burke, PhD, CStat, has worked for approximately 30 years as a statistical consultant in the pharmaceutical industry supporting clinical trials from early to late stage development.

Her experience has involved direct clinical trial, regulatory submission and manufacturing support as well as working with statistical research groups. This experience has covered a wide variety of therapeutic areas with trial designs and analyses ranging from the standard to the more novel.