The Cytel team made its annual trip to the PSI (Statisticians in the Pharmaceutical Industry) conference 2nd to 5th June. Taking place in London, UK, the theme of this year's meeting was Data-driven decision-making in medical research. As ever, the discussions both within the official conference agenda and during the networking breaks were engaging and productive.
In this blog, we share some of the particular highlights from the sessions that our team attended. We look forward to participating again in 2020 when the conference will return to Europe.
Survival analysis for AdVerse events with VarYing follow-up times” (SAVVY) project
Ursula Garczarek, Associate Principal, Strategic Consulting at Cytel singled out Tim Friede’s presentation 'Rationale and some results from the “Survival analysis for Adverse events with Varying follow-up times' (SAVVY) project from the session Statistical Issues in Drug Safety Labelling' as a particular highlight.
The analysis of adverse events (AE) is an essential part of the assessment of safety in the evaluation of new therapies in clinical trials and stakeholders agree that improvements in the evaluation of a drug's safety would have widespread benefits. Despite advances in statistical methodologies, statistical AE analyses are often too simplistic and therefore potentially misleading. The SAVVY project aims to improve the analyses of adverse event data in clinical trials through the use of survival techniques appropriately dealing with censoring, competing risks (events before AE occurrence) and varying follow-up times. It investigates the potential impact that improved AE analyses might have on the conclusion of the safety assessment compared to the current standard of incidence proportions or incidence densities. At the PSI conference, Friede presented the rationale, statistical concept and some empirical results of the SAVVY project, an empirical meta-analytic study that includes randomized controlled clinical trials by several sponsor companies,
Ursula chose to attend this presentation because she recognized that information on safety data is currently sub-optimally reported, leading to several potential problems post-approval- including delays, re-analyses, or even wrong conclusions. She was intrigued to learn from Tim Friede how to overcome this situation and commented that the SAVVY project is doing “very structured and impressive work” in this area. She gained some valuable guidance from the presentation, including to use the Aalen—Johnson estimator for the probability of experiencing an AE, and to use the Nelson-Aalen estimator for the cumulative hazard function.
Decision Making Special Interest Group
For Florence Le Maulf, Associate Director of Biostatistics at Cytel, the most absorbing session was the 'Decision Making Special Interest Group.' As the overarching theme of the conference this year was "Data-driven decision making in medical research," the topic of decision-making also permeated throughout many other sessions and posters. Within this specific session, Florence found the talk by Paul Frewer from AstraZeneca particularly engaging, describing an approach for performing continuous decision making within a clinical trial. The presentation described an adaptation of Go/No-Go methods discussed in Frewer et al. 's earlier paper on Decision-making in early clinical drug development (Pharm Stat, 2016), using the approaches continuously in a clinical trial. The approach uses a predictive power calculation to assess the chance of observing a given rate or better. The predictive power can be recalculated after each patient's outcome is available, and if the predictive power falls below a pre-agreed value, then the arm/study may be stopped. Florence found the talk thought-provoking as it clearly illustrated the role statisticians can play in helping clinical teams make the right decisions about drugs in a timely fashion for drug development. From this, and other sessions on this theme during the conference, Florence was struck by the advancements in the decision-making process made by the industry over the last few years. The rewards of these improvements are significant- helping to advance the development of the most promising medicines and making them available to patients sooner.
Implementing Estimands in Trials: Detailed Clinical Objectives
Christelle Pommie, Associate Director of Biostatistics at Cytel, highlighted the Estimands in practice session, chaired by Cytel’s Associate Principal Paul Terrill. She was particularly impressed by the session’s first presentation “Implementing Estimands in Trials: Detailed Clinical Objectives" that covered the overall concept of estimands, and contextualized their importance as well as the need for stakeholders to clearly define clinical objectives. The presentation sparked further research for Christelle, who will be attending Cytel’s webinar on Estimands, not just a statistical issue, and accessing the recommended training sessions on the ICH website.
Regulatory Town Hall
Yannis Jemiai, Senior Vice President of Strategic Consulting, Software Solutions noted the Regulatory & HTA Town Hall featuring Ralf Bender (IQWIG), Rose Lovett (NICE), Kit Roes (UMC Utrecht), and James Matcham (AstraZeneca) as a critical session which he attended.
The session discussed the use of real-world data, estimands, and external control arms in regulatory and Health Technology Assessment settings. Yannis came away from the session with several key learning points. First of all, he was struck by the differences of opinion among the participants about terminology currently in use to define emerging trends. There appeared to be a consensus dislike among the session participants for the term 'synthetic control arm' with external control emerging as the preferred descriptor. A need was acknowledged to extend the estimands discussion beyond the statistical community to engage with other stakeholders who need to know about the topic. From the session, Yannis also gained a better understanding of how estimands can be used in the HTA setting and how RWD can help inform the benefit-risk assessment. While the term 'big data' is less prevalent in discussions nowadays, it's common to hear commentary about the challenges of using high-density data such as data from wearables.
A comparison study of methods to control the influence of priors derived from historical studies on Bayesian posterior inference
Finally, our own Andrea Hita and Rajat Mukherjee jointly presented on the topic 'A comparison study of methods to control the influence of priors derived from historical studies on Bayesian posterior inference". In rare diseases, historical controls can help to reduce enrolment burden. In a Bayesian framework, borrowing from historical data is equivalent to considering informative priors. These priors can be derived as meta-analytic predictive (MAP) priors or using patient-level data. In either case, the influence of the prior on the posterior analysis should be controlled such that simulated operating characteristics of the Bayesian design still conform to regulatory standards. Andrea and Rajat presented a numerical study comparing two existing methods to control the influence of the informative prior on the posterior analysis and created, and shared some animations to compare the performance of the two methods. These animations will form the basis of a future blog.
We hope to see you next year in Barcelona for the 2020 meeting!
Would you like to learn more about the estimands topic? Click the image below and register to join us on Tuesday 25th June at 10am EDT for our webinar 'Estimands, not just a statistical issue' with Paul Terrill. Can't make the time? Register anyway and you will receive the slides and recording after the webinar.