As the Co-Founder of the AI Applications in Biopharma Summit, my purpose is fueled by providing valuable information about AI applications to the biopharma community so that ultimately, treatments can get to patients faster. In 2020, that purpose has been shifted from in-person events to digital. I have been launching and attending digital events to continue to learn about new innovations and applications of AI in biopharma. I thought I would share what I found to be the key learnings and insights from some recent digital events. Below you will find selected talking points that I found to be thought provoking and valuable insights. Please note that these are summaries and not verbatim comments.
Topics discussed at the AI in Healthcare Leadership Summit in June include:
- AI and Digital Solutions – there is a role for AI and digital health across that entire product lifecycle
- Data Quality – AI is only as good as the training sets
- Data Sharing – a role for AI is to learn across data sets even when they are siloed
- Data Privacy, Access and Consent – engage patients in the conversation
- Real World Evidence – technical issues and the people element are equally important
- Synthetic Control Arms
- Final Thoughts and July 14th webinar information
Amy Abernethy, MD, Ph.D., Principal Deputy Commissioner, FDA
Brandon Allgood, Ph.D., SVP, Data Science & AI, Integral Health
Mathieu Galtier, Ph.D., Chief Product officer, OWKIN
John Halamka, M.D., President, Mayo Clinic Platform
Najat Khan, Ph.D., Chief Data Science Officer, Johnson & Johnson
Naheed Kurji, CEO, Cyclica
Agustin Marquez, Ph.D., VP, Products, nference
Angeli Moeller, Ph.D., VP & Head of Global Data Assets, Pharma Digital Transformation & IT, Bayer Corporation
Maria Luisa Pineda, Ph.D., Co-founder & CEO, Envisagenics
John Reynders, Ph.D., VP, Data Sciences, Genomics and Bioinformatics, Alexion Pharmaceuticals
Dan Vahdat, CEO & Founder, Huma
Joshua T. Vogelstein, Ph.D., Department of Biomedical Engineering, Johns Hopkins University
AI and Digital Solutions
AI and digital solutions are going to revolutionize what we do for medical product development across the entire lifecycle. Starting with the point of discovery, through to the definition of value, and the implications at the point of care. There is a role for AI and digital health across that entire product lifecycle.
There is an explosion of data but is there a more systematic way to make better decisions? AI could make a big difference in this area.
When it comes to AI, it’s a mistake to focus on the sizzle alone and not the steak. Companies must be realistic about what AI can do and what it can’t.
Maria Luisa Pineda
As innovators we need to share different approaches to find ways to get beyond roadblocks to implement AI.
Developing data sources is a key task for everyone. Every source has strengths and weaknesses. You need to be transparent about the data quality and make sure they are well characterized to understand the implications of the data quality.
AI is only as good as the training sets. Healthcare data is mostly bad because of the nature of how it’s recorded. So how do you collect large amounts of curated data? What is Mayo doing and what big questions remain? Mayo is digitizing and deidentifying decades worth of data. Messy issues remain about patient consent. Mayo is working on this issue and will have more to share in the Fall.
Standards are important to be able to compare methods, to ensure that patients and healthcare workers have confidence in the use of ML algorithms and intelligent agents. As we move into clinical trials, safety and patient protection are important and, with the right standards, can possibly improve trial speed. Bias is always going to exist and is not inherent in a model – it is in the application of that model. One of the things that is important is that there is enough meta-data and that the model can give error reports if it’s being used incorrectly. Fairness is an issue that is related to bias that we need to tackle as a society.
We are getting close to an inflection point around cross linkage of data sets and validation across data sets. This is starting to happen in imaging where people are going to see if what they predicted was accurate. The same thing is starting to happen with endpoints. We need to start to build valid data standards now to scale in the future.
The Evidence Accelerator has over 45 groups participating, and this is a data and analysis community. Data holders and analytic teams are coming to the table to share information. This has also become a learning community which is allowing the group to get the methods right, so they are better prepared for future issues.
The transfer of sensitive data is flawed. OWKIN is building a platform to do ML while leaving the data in the hospital. OWKIN’s approach is to be a “data-less” AI company. OWKIN is working on creating a decentralized network for data that will be a central hub where AI can be applied without actually transferring data.
The concept of federated learning techniques allows you to leverage existing data while complying with regulatory and privacy concerns. A role for AI is to learn across data sets even when they are siloed.
Data Privacy, Access and Consent
Patient’s data is owned by the patient. The patient has the option to decide if they want to share their data for clinical care or research purposes. Even sharing anonymized data, you need to be thoughtful about why you are sharing data and who you are sharing it with.
GDPR provides a guardrail, or set of guidance, about using data ethically but culture is even more important than GDPR. Pay attention to culture because sometimes that is the real limiting factor. Engaging patients is key to ensuring that AI can be adopted faster.
Giving the power to the patient regarding what privacy and confidentiality they accept is critical.
Real World Evidence
RWE and data solutions are part of digital health where digital health doesn’t just mean hardware or software but includes data.
There is currently a significant manual component of RWD curation where AI can make a big difference in scale beyond manual curation.
Leveraging real world data is starting to show some interesting results. There is a disconnect between what we call a disease and how we treat a disease. RWD, using explainable AI, is allowing companies to understand the phenotype of a sub-population and how that phenotype is manifesting itself. Then using a number of methods in relating biomarkers to basic biology, going to a target and then to genotype confirmation. So before a discovery program begins, you can know what the responder patient population should look like and what the clinical trials will be and how they relates to the disease. This will hopefully go a long way toward decreasing the failures due to efficacy. This will lead to diagnosing and ultimately treating people biochemically.
What needs to be focused on? Technical issues: need appropriate data to create a complete picture of a patient is needed to predict what will happen to a person. Until we get to this, the utility of real world data will be limited. The people element: talent needs to understand science and data science. This is the biggest key to having one person being able to understand and solve for both clinical and technical issues. This will require change management internally to be able to move beyond pilot programs and move to scale.
Synthetic Control Arms
In the development of synthetic control arms, there is a lot of work to do as this space is still in the developmental stage. The data needs to be fit for purpose, you need to know how to do the analysis, and you need to choose the right use cases, but there is potentially a lot of value to society.
There is no one data source to rule them all and having interoperability as a business strength is key. One of the things COVID-19 has forced is faster learning. We need North Stars, the person who sits in front of you in the clinic, makes everything you are doing more urgent and tangible.
AI in Biopharma (www.ai-inbiopharma.com)
Connecting biopharma’s science and technology is what we do at AI in Biopharma. To continue this important discussion, with a focus on how to translate AI/ML technologies into patient impact, register your interest here: www.ai-inbiopharma.com for an upcoming interactive webinar on July 14th with Niven Narain, CEO, BERG, Andrew A. Radin, CEO, twoXAR, and Harry Glorikian, General Partner, New Ventures Funds, Author, MoneyBall Medicine: Thriving in the New Data-Driven Healthcare Market.
Additional speakers to be announced.