The Future Now
Wow! – what an inspiring start to the SCDM Live Global Conference! Some 1,500 data professionals around the world joined to share clinical data management experiences and insights after a year of accelerated, COVID-driven change. There were a number of excellent presentations on AI/ML.
Chris Lintott, Professor of Astrophysics, University of Oxford and Sanjay Bhardwaj, Executive Director and Global Head Clinical Technology Strategy & Operations, AbbVie, delivered a fascinating Keynote titled, “The Next Big Step – The Future for Clinical Data.”
The age of big data is here, and drug development is filled with new possibilities. In a positive and refreshing perspective, Chris and Sanjay look to a future with powerful opportunities in which AI/ML works in tandem with human knowledge and intuition
Big Data Is Here
Clinical development is justifiably regarded as slow to adopt new technologies, given concerns for patient safety and privacy. Astronomy is not constrained in this way and points to a way for clinical development to evolve. With large datasets where content is measured in soon-to-be multi-Petabytes, similar challenges abound regarding acquiring, systematizing, and interpreting large quantities of data.
Astrophysicists are developing new toolkits to overcome technological problems, ranging from the issues of database design and federation to data mining and advanced visualization. Astronomists seek to use big data and huge datasets in new ways and to challenge what’s known. Machine learning provides increasing capability, but as Chris states, the best datasets are the ones that contain the learnings from earlier datasets. AI and ML can only analyze data based on what they have been “taught.”
► Astrophysics found that large datasets even if unclean will outperform small, targeted datasets.
► “Accurate but not correct” suggests that human intervention is necessary, to understand the problem and the solution being sought, and to apply intuition to advance a fit-for-purpose model.
► In one example, ML was used to predict heart disease from x-rays. Standard and portable bedside machines were deployed, using the same algorithms. Each was able to identify heart disease; however, the portable units generated more positive outcomes that were ultimately not correct.
Sanjay acknowledged that DCT is clearly here in a post-COVID world. “The future is now, and we must pivot to that future,” he says. The big data trend in clinical development is driving velocity and variety, while centricity is declining. Having evermore data sources is now routine. Performing line-item analysis is no longer scalable or acceptable.
Sanjay agrees that differentiating between important and unimportant data is a key factor in advancing AI/ML. A typical Phase 3 trial used to have a million data points. Now, with eSource, etc., there can be billions of data points.
More on AI/ML
Some current concerns in AI/ML advancement include:
► Reproducibility – which aspects of data are important, what data is retained? If the wrong data is retained, then whole experiment needs redoing.
► Transparency – Good data is necessary in order to train the algorithms. AI at its core is showing machine examples and asking questions, making training data sets crucial.
► Process change – current DM processes are not scalable and new ones must be fit-for-purpose. Applications like NLP are evolving even as they are incorporated into protocols.
► Privacy – concern amongst patients, providers, sponsors, CROs, and regulators that shared data may be misused and leaked to insurance companies or employers.
Both Chris and Sanjay see a bright future where AI/ML does the heavy lifting for human data science professionals. Centaur Thinking (derived from world chess champion Gary Kasparov who was defeated by IBM’s Deep Blue computer) has shown that combining machine heavy lifting and human insight produces the best results. For example, an Apple watch may detect a raised heartbeat at rest and suggest that you contact a health professional, when the cause was simply watching a sporting event.
AI/ML requires human understanding and intuition to teach the machine and interpret its decisions. To this end we will require new systems that can tell us how the machine makes its conclusions.
DCT is Here to Stay & There is no Secret Sauce
There were several sessions that dealt with changes in decentralized clinical trials, including one hosted by Adaptive Clinical Systems. In the session, “In Conduct of Decentralized Trials: Patient IS the Source,” Adaptive Clinical’s CEO and President Sina Adibi lead a great session that included insights into better data management, trends in globalization and a focus on adding electronic health records, and medical claims data in great case studies including a case study on the challenges and lessons learned from employing EHR in DCT. In this session, leaders from ICON, Science37 and Janssen answered questions around best practices for:
► Data management when moving to a hybrid or virtual setting
► Managing globalization challenges in DCT
► Sponsors and CROs when it comes to leveraging Electronic Health Records and Medical Claims Data
In the presentation, “How to adopt and adapt to the changing eClinical landscape,” the panelists noted several insights into the data management challenges:
► There are many siloed solutions lacking interoperability
► There are ever more sources of data
► EDC holds a decreasing role in terms of the percentage of data in a given trial
► The ability to design a study once, and enter data just once is important
The session, “Decentralized Clinical Trials: What Does it Really Mean?” focused on the importance of front-end preparation for data management. This included recommendations such as pre-planning.
Because of the growing complexity of DCT, pre-planning is essential. Study managers must have a rigorous data management plan and draw a picture of it. Map your data flows! It’s complicated and planning can’t be over-emphasized enough. There are advantages to hybrid, such as faster and easier patient recruiting. A specific example discussed was a large COVID trial, which could not have scaled up as fast without the ability to embrace some degree of DCT.
The DM plan needs to account for a way to capture all of the data, which given the new devices/complexity is not trivial. The speakers from Clinical Ink, Prevail Networks, and Pfizer also shared lessons learned:
► Devote more time for up front planning including all of the stakeholders. There will be new, more involved protocols with DCT, and these need to be reviewed.
► The initial planning should identify vendors/services that need to be brought in. Identify these partners early and involve them in the planning process.
► There needs to be a central database to store and aggregate the data. The system needs to be flexible and must enable early data visibility.
► Regulators are actively working to provide guidance for the new situations and models. They are supportive.
► The panelists recommended that data managers should actually draw/visualize the data flows.
In the session, “Which is the best outsourcing model for Data Science? Have we found the secret sauce?” This session included speakers from Novartis, Cenduit, TATA Consultancy, IQVIA, and GSK. While the presenters all agreed there is no secret sauce, they noted several innovations in clinical trials:
► Data Management is now Data Science. Data scientists must lead and include visualization. With the evolution to data science, data is available in real-time across the board.
► Data Science will evolve to into a leadership role in clinical trial planning and conduct – owning the data inputs through to delivery of the prescriptive outputs for all data users.
► Among the terms used most often by the panelists to describe sought after attributes were “innovation” and “prescriptive information.” It is all about the mindset of change transformation.
► The pendulum of outsourcing will swing from single source providers to multiple and sometimes duplicated resources and rejuvenated in-house GCC (Global Capacity Center) capabilities.
► Regional pockets of data expertise are popping up around the world. India still leads and other regions like the EU now host more highly skilled data professionals.
This year’s virtual conference had outstanding content from some of the best industry leaders. The discussion continued to focus on the transformation of Clinical Data Management to Clinical Data Science and the pragmatic steps necessary for this innovation. We saw sessions covering all the megatrends in CDM from DCTs, RBQM to AI to technology data and communication.