Deeper RWE insights with AI
The application of AI to swathes of previously un-tapped health data is opening up new potential in the use of real-world evidence for high quality clinical research.
The long-promised potential of Artificial intelligence (AI) and machine learning (ML) to help uncover patterns and insights from real-world evidence (RWE) to inform better clinical and healthcare business decisions is on the cusp of being realized.
Thanks in part to the 21st Century Cures Act in the US, the healthcare business is increasingly moving beyond established analysis of clinical and insurance claims data designed to support the efficacy, value and benefits of drugs in the real-world clinical setting and into new applications that open up entirely new possibilities in RWE.
A useful example is the growing use of AI and ML applied to imaging and genomics datasets in research. The commercial applications are broad too, says Abhimanyu Verma, VP/Lead of Applied Technology Innovation at Novartis. “AI can boost RWE in the commercial side of pharma by allowing us to identify patient subpopulations, find high responders to treatments, forecast therapy switching, enable indication expansion, and identify when prescribing behavior might change.”
The potential is vast, adds Ignacio Medrano, Neurologist and Founder of Savana (an AI-powered deep real-world-evidence company dedicated to accelerating health research) and Mendelian (a genomics platform designed to help doctors diagnose and treat rare diseases earlier). “Using AI with electronic health records, we can understand the natural evolution of diseases, the behaviors of physicians, and the effectiveness and toxicity of drugs. It’s all there, just waiting for us to have a look,” he says. “Omics are key to understanding patterns, but the evidence in the last year is showing us that omics are not enough if they are not complemented by clinical variables, which can only be found in EHR.”
Personalizing medicine with RWE
AI can be an especially powerful tool for boosting RWE in personalized medicine and helping find the right treatment for the right patient at the right time, says Medrano. “Individualized medicine is possible when we can analyze big datasets [because] algorithms allow us to narrow insights at a personal level. That level of extraction of knowledge is simply not possible with classical statistics and manual data collection. We find associations among variables with AI where the human mind would not be able to.”
Profiling patients has traditionally come from knowing the genomic makeup of the individual, says Medrano, “but when we bring in the different factors, such as the patient’s environment and other conditions, we have to go beyond traditional profiling and look at clinical data, medical history and family history,” he adds. “That is where AI has lots of potential.”
Ultimately, AI can help provide a detailed characterization of patients, which informs the matching of treatment and follow-up frequency, all while helping to predict possible reactions. Together, this information enables a tailored approach to healthcare. “It’s just like with the weather or traffic: the more historical data you have, the more you can predict what is going to happen in a particular situation,” says Medrano.
New clinical insights
AI’s potential to help analyze a selection of clinical features that scientists might not have previously considered, such as extracting insights from imaging that humans cannot, shows particular promise. Deep learning and machine learning support medical image processing and analysis for diagnostic and predictive purposes.
AI also offers a novel way of enhancing the work of clinical experts in research, says Asif Jan, previously Global Head of Personalized Healthcare Analytics at Roche, and currently Head of Platform and Solutions at Owkin. “Traditionally, we relied on expert know-how when designing studies. We now have the means to augment this expertise by unleashing capabilities offered by AI/ML/DL. With AI and ML, we can process massive amounts of data, generate, test and visualize hundreds of thousands of combinations, and present the results that we believe best depict the phenomenon we are studying.
“Then, our clinical experts can review the outputs and finalize the best strategy for the drug being studied. This really goes to show how AI/ML could be integrated as an assistive and augmentative capability in drug development programs.”
AI can also enable a deeper understanding of outcomes in clinical research. “Once you have access to all the clinical variables, you can use them to interrogate clinical pathways, deepen the understanding of phenotypes, create risk stratification, or track complex patient journeys. That is what we call ‘deep RWE’,” says Medrano.
“We have always stratified the risk. The difference is that formerly we did that with classical regression, with hundreds of variables and inflexible and outdated databases. Now we can do the same using machine learning, with literally tens of thousands of variables and with new mathematical models like random forests or Kernel networks - and with databases that get automatically updated as clinical practice happens.”
RWE within regulatory frameworks
There is a regulatory application for AI and RWE too, says Jan. “A lot of the work in AI and ML today is done in the R&D domain to support clinical trial design, accelerate trial conduct and to offer additional analyses to support the main trial findings. Naturally, as we use those AI algorithms in R&D, they will end up being used in the regulatory context.”
But do regulators trust RWE? Maybe not as much as we would like. Jan estimates that currently 80% of RWE is used for internal decision-making and 20% or less for external discussions.
These are certainly still early days, says Kelly H. Zou, Head of Medical Analytics and RWE at Viatris. “Although RWE can help understand patients, health conditions, and healthcare resource usage beyond randomized controlled trials, its use in a regulatory capacity is growing and on the cusp of finding cures and expanding labels faster.”
Promoting AI insights in regulatory frameworks first requires trust in these new approaches to RWE, which requires helping regulators understand how AI is used in analyzing it, says Verma, “Since AI/ML is a relatively new approach to analysis in the context of healthcare RWE, explaining how results were achieved, whether to internal decision makers or external regulating bodies, is so much more important now. AI/ML methods are not as well established, so you need to be able to explain how the results were achieved to create confidence in them.”
Pharma must also be alive to biases that are inherent in real-world datasets and be careful to select the right control variables as well as the parameters used to filter the selection of those controls. “Methodological transparency, best practice adoption, data privacy protection, and ethical consideration are all essential,” adds Zou.
RWE’s data dilemma
One of the most significant barriers to realizing the potential to use AI in RWE is the messiness and of real-world data from routine care, compared with clinical trial data obtained in controlled settings , says Jan. “When we look at RWD, there may be a lot of gaps due to data not being collected regularly or not being collected at all. RWD is not collected according to a protocol or guidance.
“The state of RWD is actually a depiction of how things are in the real world – the data are quite messy and not everything is effectively managed or easy to explain. Therefore, we need to be very careful when it comes to interpreting the results from a model, ensuring we have access to experts to cross-check the findings.”
How should pharma address missing data? “It used to be that any missing data was excluded from the study, but if we used the same approach with RWE, we would be left with nothing to analyze,” says Jan.
It turns out that, however, might be able help fill in some of those gaps, he adds. “AI/ML may be able to supplement some of the missing values in a dataset by generating synthetic data that captures underlying disease and patient characteristics.”
Even so, the fact remains that while AI can help fill in some RWD gaps, the available datasets are often far from ideal, says Zou. “Even structured data pose challenges in the application of RWD analytics and AI, besides unstructured data may have inconsistent terms and different formats between sources. There may also be incomplete or messy information. These situations might lead to inaccuracies in the analyses and the convergence of the algorithms.”
However, Medrano is bullish. “Savana applies its scientific methodology throughout its deep Real Word Evidence studies. We ensure high quality data from design, data sources, analytical methods, reproducibility and transparency, results reporting and interpretation to reveal deep Real World Evidence that clinicians and regulators can be confident about.”
While the use of AI to capture, amalgamate, standardize, and analyze RWD is still evolving, it has the potential to support the increased availability of data to improve global health and healthcare now and in the future.
ends
Delivering higher quality RWE and patient insights is a key theme at Reuters Events: Pharma USA 2022.
Be the first to see the agenda and speaker line up! Pre-order your Info Pack now >>>
Pre-order your brochure now and be the first to discover the industry leading speakers who will be giving you practical insights on how to deliver deeper insights and fast-track your approvals.
Past speakers include:
Be the first to see the agenda and speaker line up! Pre-order your Info Pack now >>>
I hope you enjoyed the article - please let me know if you have any feedback, or if you want to get involved with the event.
Kind regards,
Reuters Events is part of Reuters News & Media Ltd, 5 Canada Square, Canary Wharf, London, E14 5AQ. Registered in England and Wales: 2505735.
Pre-order your Pharma USA 2022 info pack now!
We’re coming home to Philadelphia – optimize your engagement strategy for the next generation of patient leaders at Pharma USA 2022 (formerly eyeforpharma Philadelphia) and realize a new era of patient-pharma collaboration