Share this article:
AI and Big Data in Cancer: How to effectively translate technology, data and analytic innovations into clinical practices and patient benefits
The future of medicine is data-driven and AI-enabled, which will impact diagnosis, treatment decisions and patient care. AI and Big Data in Cancer 2020, hosted by Elsevier, is taking place March 29-31st in Boston, and is bringing together leading experts from cancer centers, start-ups, technology, private equity, innovation centers, pharma and government. The aim of the conference is for attendees to learn from others along the digital medicine value-chain to understand how to effectively translate technology, data and analytic innovations into clinical practices and patient benefits. Attendees can network and find partners to move innovations forward. In advance of the conference, we asked some of the key speakers their thoughts on the future of the implementation of AI in oncology care and the barriers that must be overcome in order to succeed.
Craig Mermel, MD, Ph.D., Product Lead for Pathology at Google Health, believes that first we need to overcome the fear that AI will replace doctors and researchers. “AI-drive medicine won’t be the complete replacement of doctors and specialists. That’s what most people fear and while I think it is inevitable that AI is going to make its way into the delivery of care, but that it is largely going to be an augmentation and efficiency tool for physicians to practice in a way that is much more personalized, accurate and scalable. The reason for this is that the recent advances in AI are still very narrow forms of intelligence. We can train AI systems to do very narrow specific tasks, but it isn’t likely that we’ll just replace doctors by training systems like thousands of consisting of thousands of narrow AI- tools; so we simply don’t yet have the technology to replace the general intelligence and synthetic abilities of doctors. It is a uniquely human characteristic. Also, from a human perspective, we want the information these systems to provide to have the backstop and the values of a human decision-maker, and we’ll still need doctors for that. We need to find out how to be sure that AI-Driven medicine augments what people are doing, not replaces it.”
Manoj Saxena, MBA, Executive Chairman, CognitiveScale and Founding Board Member, AI Global, believes AI and Big Data will be as transformational to medicine as the stethoscope was 200 years ago. “All aspects of medicine are going to be impacted, specifically the four major groups of end customers, care providers, the clinician and research community, and healthcare delivery business models…In ten years, I think AI is going to show business efficiency through agility. These are taking boring backoffice processes and cycle management and claims processing and enrollment and really driving efficiencies through it to make these processes more agile. The second is innovation. This is being able to take work that companies like looking at genetic evidence and how you treat and diagnose and prevent diseases and building a whole new type of healthcare platform for disease lifecycle management. There is innovation that is going to happen around member/patient engagement, there is innovation that is going to happen around disease prediction and diagnosis and treatment, there is innovation that is going to happen around business models around how you connect up different participants in an ecosystem, reimagining the existing processes and coming up with a whole new set of innovative engagement models. The ability to completely reimagine and transform a claims process using a combination of AI and big data and block chain is an example of a moonshot. Many will fail but those that succeed will completely revolutionize the industry.”
Ned Sharpless, MD, Director of the National Cancer Institute agrees. “There are a tremendous range of applications for AI. AI/ML is already having a significant impact on clinical medicine as practiced today, by doing things that humans can already do pretty well, but by doing these things in a manner that is cheaper, highly reproduceable and at scale. These approaches are here at this very moment and their use is considerably more prevalent already than has been commonly realized. In the future, though, we believe AI/ML algorithms will be developed for widespread use to do things that no human can do, but that the technology will make possible. There are many places where we expect to see such advances, but I will give one example of interest to the NCI: the development of algorithms that can predict future cancer risk using standard radiographic imaging data. That is, to predict future cancer risk in a way that no radiologist can discern using standard training. The idea is that an algorithm can be trained on large sets of annotated radiographic imaging to predict future cancer risk from so-called “normal” scans. For example, we expect an algorithm could be trained on chest CT scans from large clinical trials like the National Lung Screening Trial (NLST), and based on that training, then identify subsets of patients with a significantly increased risk of developing lung cancer over the next few years, even though these chest CTs would be considered unremarkable by standard radiologic interpretation. That is going to happen. We’re going to see better readings of pancreatic scans, and chest CTs, and other imaging modalities; and in some instances, these algorithms will make accurate predictions about future cancer risk in a way that radiologists just can’t do. If we train these AI imaging analysis algorithms on large enough datasets, they can start to see things that are too subtle for traditional radiologists. This sort of capability will be very important to us for many reasons. Patients with an increased risk of cancer might be followed by their doctors in a different manner, or would be more likely to benefit from a clinical trial of some sort of cancer prevention. Such a technology has immediate application for the NCI.”
There are quite a few barriers that must be overcome for the successful implementation of AI in oncology care and suggestions on how to surmount them will be discussed during the conference. These barriers include access to expertise, a recalcitrant culture, the availability and quality of data, the maturation of the technology, processes like integration into the workflow, reimbursement model, and administrative requirements, and finally external barriers like regulatory requirement, data privacy & governance, and business models. These will be discussed in detail at the conference with suggestions on how to turn them into opportunities from different perspectives including clinical, research and investing.
For Christoph Lengauer, Ph.D., Partner, Third Rock Ventures, the biggest barrier is the data. “It needs to be purposeful data. The second thing is, I think we need to have an application. We do a lot of things because they are cool but very self-fulfilling. From an investor perspective it is so important. We are forgetting in all of this that it has to have some purpose and be value-creating. Just because we can do something faster or quicker; it doesn’t get us anywhere.” David Shaywitz, MD, Ph.D., Founder of Astounding HealthTech advisory services, further adds to this point, “The quality of the data and the need for focus are the biggest barriers for me. When people talk about digital transformation, some folks have the idea that we’ll get lots of data together, stir these up in a big vat, and then magically they will start to speak to us. That seems unlikely. I prefer a use-case drive approach; to define what are the actual problems that you need to solve and then go through the process of trying to solve these specific problems. In the course of doing that, you really learn what the issues are with the data, and you figure out what are the right technologies. Ideally, you would start from the ground up and collect data in a fashion that is amenable to AI – you’d start by structuring it right from the beginning.”
Ned Sharpless, MD, agrees that a big barrier is having suitable datasets. “From the NCI’s point of view, the biggest barrier is having the appropriate datasets and having the appropriate people to work with that data. Both of those are things that I think we can make a lot of progress on with accurate funding. Data that’s better in the sense that it is clean and well-annotated. So, it has features like clinical outcomes matched to radiology that they want to use to train their algorithms. The argument has been made that retrofitting old datasets is not often possible. In some instances, we will need new collections of data that are designed for machine-learning from the get-go. It’s certainly a big investment for us to use AI on existing datasets, but we’ve also heard the message loud and clear that we need to fund new datasets that have AI in mind from the get-go. Another desperate need is the people to work on it. The machine-learning people who are interested in biology – there are so few of them and we really need to train up more.”
Andrew W. Lo, Ph.D., Professor and Director of the Laboratory for Financial Engineering at MIT Sloan School of Management, believes the biggest barrier, from an investor perspective, are new business models. “That, to me, is the one thing that we can be thinking about differently that we aren’t right now. People are already focused on scientific collaborations, new types of biological mechanisms and targets, genomics, transcriptomics, proteomics, and all the other -omics. But the one omics they haven’t focused on is economics, new ways of structuring biopharma businesses and financing them. This challenge also offers tremendous opportunities for applying the tools of modern financial engineering to biomedicine and getting investors to think differently…. By adopting a portfolio approach to biomedical R&D, we can lower the cost of capital, increase the amount of funding, and get new and better therapies to patients faster and cheaper.”
Kathy Giusti, MBA, Founder and Chief Mission Officer of the Multiple Myeloma Research Foundation and Faculty Co-Chair of the Harvard Business School-Kraft Precision Medicine Accelerator, states, “With an intense focus on generating high-quality data, the MMRF has seen firsthand the challenges of using that data to answer clinical questions that the community deals with on a consistent basis. In addition, new business models must be developed that involve the full ecosystem of academia, health systems, industry and patient groups. Those business models should include having AI at the table much earlier helping to frame the data requirements in advance. Furthermore, incentives must be aligned to prioritize use cases that can improve patient outcomes.”
Craig Mermel, MD, Ph.D., believes the biggest barrier is in the process and externalities. “Until we really understand how to fully integrate AI into the existing workflows and build sustainable reimbursement models around it; that’s a barrier that is slowing it progress down and potentially preventing it the technology from reaching the maximum number of patients. Much of the near-term attention needs to go to process and workflow integration. If we can improve on quality and that this technology is safe, and integrates into actual workflows, and improves overall quality of care, I think many of the regulatory externality issues will move forward as well.”
Help figure out how to turn these barriers into opportunities at the AI and Big Data in Cancer in Boston. It is time to shift the conversation from what AI and Big Data can do to what medicine needs. Be sure to attend to hear more from the above great speakers and other leading experts to translate artificial intelligence and data-driven innovations into new clinical care practices for patients.
View the full program and register here: https://www.elsevier.com/events/conferences/ai-and-big-data-in-cancer Follow the conversation on Twitter: https://twitter.com/AICancer2020
The scope of life sciences is as vast as the variety of life on Earth: mathematical biology, developmental biology, molecular and cell biology, parasitology and virology, microbiology and immunology — the list goes on. Elsevier, through its renowned imprints like Academic Press, provides high-quality content in all of these areas that supports learning, teaching, and research. Our books, eBooks, journals, and online tools are cross-disciplinary, allowing academics and professionals to effectively learn about science outside their areas of focus.