He wanted to talk to us about his passion for statistics in health-related areas and providing training for those with a non-statistical background to communicate statistical results effectively. Stephen spent several years working in academia where he delivered undergraduate and postgraduate lectures in statistics and provided statistical consultancy to professionals working in academia, industry and government organizations. He joined Phastar in 2018 and has been involved in phase 1 – phase 3 clinical trials in several different disease areas.
Why it is important to clearly communicate statistical concepts, so data value is understood from the boardroom-level down?
Data science and statistical analysis are hot topics in clinical trials right now. Statisticians play a significant role in ensuring we use study designs that provide robust data, aligned with scientific questions. It is vital the statistics they produce are communicated well so their value can be appreciated at every organizational level.
Unfortunately, many statisticians are not seen to be communicating clearly because of the technicality of language used in the field. This can lead to inefficient and inconsistent messaging to senior leaders who want to see how assets are progressing.
So how do we improve communication and education on statistics throughout the clinical trial system?
Here are two key areas where improvements could be made for the benefit of all stakeholders.
- Clear articulation of challenges to stakeholders
As statisticians we need to take a step away from jargon and try to articulate our challenges in a language our colleagues can interpret and use to convey insights.
Stakeholders and senior leaders want to see answers, and statisticians are expected to deliver these answers. But all statistical methodologies need data. If the data density is poor when it is sent to statisticians for analysis, it can severely limit what can be done.
The desire to gain maximum insight from all data is only natural. If planned analyses are scaled down or stopped due to data, we must clearly manage expectations. This requires clear communication of the challenges faced by the statistical team, the impact on analyses and insights, and the plan for what happens next. Crucially, this communication needs to be understood by key study personnel so consistent messaging can be brought to all stakeholder groups.
Unfortunately, jargon used by statisticians to discuss which methodologies can be used when data is sparse often impedes this communication. This can lead to requests for statistics which do not exist and negative interactions between statisticians and study personnel.
- Using a variety of measurements beyond just P-values
P-values are synonymous with clinical trials. Every day, clinicians and regulators use them to make decisions on which treatments will be used or marketed. Generally, they represent the probability of observing the data you have observed, or, in more extreme cases, if there is no difference between treatment arms.
To determine what is ‘unlikely’, p-values are compared to a threshold pre-specified at the study design stage. The most common threshold in clinical trials is 0.05. This means there is a 1 in 20 chance we will conclude there is evidence of a difference between the arms where none really exists.
The problem is p-values are being used in ways for which they were not designed. For example, reports of trials being ‘successful’ with p-values of 0.049 and ‘failures’ with p-values of 0.051 fail to recognise conclusions do not immediately become true on one side of the divide or false on the other. P-values do not give any insights into clinical importance. They only tell us that the data is statistically different.
Small p-values do not automatically mean you have an important treatment affect. Similarly, larger p-values do not mean a study has found nothing of clinical importance.
Those responsible for interpreting and disseminating data should remember the p-value is a useful statistical measure but it should be interpreted alongside effect size and associated confidence intervals. Only by looking at these measurements together can we truly understand what the data is telling us and provide fully informed guidance on what is best for patients.
What is your conclusion – is now the time to bridge knowledge gaps?
Statisticians and stakeholders need to start talking the same language. Statisticians need to get better at providing unambiguous, non-technical interpretations of their analyses. At the same time, colleagues need to be educated about the common lexicon. This will allow them to convey results to senior leaders while maintaining important caveats on results.
Statistics and the interpretation of statistical results can help us make the right decisions on how to best help patients. This is not about everyone becoming an expert. Instead, it is about bridging knowledge gaps so conversations can be more engaging and effective. By doing so, clinical trial teams can produce fast insights without sacrificing quality.