Women in life sciences: Miruna Sasu

AI in cancer care

On International Women’s Day, pharmaphorum looks back to an interview with Miruna Sasu, PhD, CEO of COTA, a real-world data company that is advancing generative AI models for cancer care.

Sasu, who has previously worked at Bristol-Myers Squibb and Johnson & Johnson, is responsible for several developments at the company, including a deeper relationship with Google to apply machine learning (ML) and natural language processing (NLP) to EMRs.

Unreliable data is a huge pain point for life sciences companies working to create lifesaving cancer treatments for patients; so too is establishing trust in AI models. And one of COTA’s differentiators is its added medical expertise and human touch to automation, oncologists working in tandem with the AI to ensure that the data is curated in a precise and relevant fashion.

AI in oncological care, avoiding bias

Speaking with Sasu at the end of 2023, pharmaphorum sought her thoughts, as a female leader in the field, on the opportunity for AI in oncology.

“There are a lot of opportunities for AI across healthcare, [but] I think in cancer care, specifically, there's a lot to be learned from what has happened to other patients,” Sasu explained. “Ultimately, my background of what brought me to COTA is a personal story. I think a lot of people within the company have personal stories. They're coming here to work here to really make an impact.

“In terms of AI, we've been using 'AI' for a long time to draw data out of electronic medical records and put it in analysable form. We've been doing that with AI, as well as other technologies, but also with medical oncologists and medical professionals.”

One of COTA’s largest datasets, Vista, has tens of thousands of patients in it, such big datasets enabling research that studies the patient health experience across a continuum of geographies.

“We're specifically in the United States, but […] we have both academic institutions and community institutions in our network,” said Sasu. “When you see a dataset from COTA, you know that there are patients that are stage one cancer, stage two, stage three, stage four across the entire geography. That representativeness is really important.

“For whatever reason, if you don't have access to these types of datasets, or you don't want to use them, then you're biased in what you're trying to do in terms of whether it's in drug discovery or drug development or whether it's in treating patients. If you're an oncologist, you have only what's in your mind and what you've been trained with to see what you can do for these patients. If you're looking at one of these big datasets, you can say, okay, here's my experience and, also, here's what other doctors and what other patients have experienced across the geography, across the world.”

On safety, efficacy, and scalability

The critical purpose of this representativeness is, of course, to develop drugs that are both safe and effective and of benefit to the vast breath of patient diversity. And AI has permitted a process that is accelerated and scalable.

“We want to make sure, number one, that patients are safe, and number two, that it actually works,” explained Sasu. “There's no reason to test a drug if it's not going to be safe and if it's not going to work […] If you can have a view into this prior to actually performing the clinical trial, then you're saving a lot of patients a lot of pain and suffering […] You’re not having them go to the hospital three times a week or whatever it is that you need for that trial. That's the direction now behind why you need these data sets.

“Over the past few years, AI has gotten a lot better,” said Sasu. “What that has allowed us to do is get more of these records into these data sets, and it's also allowed us to get them a lot cleaner, faster. With Chat-GPT and others that have come on the market, what that has done for us is that it's opened an avenue for us to do this on a much larger scale. The Vista product is [COTA’s] largest data product. Now, we have hundreds of thousands of patient records that are conglomerated into one asset, whereas a couple of years ago, that would not have been possible.

“This is a totally different ballgame […] The only thing that we have to do in terms of manual check is at the very end, after the technology is done with that data set, the medical professional goes in and goes, ‘Okay, does this make sense?’ We want to do that, and we're taking that on because we want to make sure that nothing slips through […] AI has helped us scale a lot of the data that we are putting in the hands of these researchers.”

A technological marvel, but no Marvel

As much as there is considerable hype that surrounds AI, in future terms, we’re not quite at the point of such technologies as suggested at in movies, but, as with much imagined within science fiction, time will tell.

“Truly artificial intelligence, like you see in Iron Man, like the Jarvis type of thing, is not here,” explained Sasu. “That's not what we're utilising. Under the scope, under the umbrella of artificial intelligence, there are different mechanisms or different types of intelligence that we can use. One is definitely machine learning. What that means, at a very high level, is that, if we correct an algorithm, then it learns from that correction. That's why it's really important for the medical professional to be in the loop because the algorithm will only make that mistake once. You have now taught it that, hey, no, this is stage two, not stage one. Now, every time it sees that case, it will call it stage two. It's important for the logic to be there by the medical professional. That's machine learning.”

And what of natural language processing?

“This is really important: we've been using NLP for years and years,” said Sasu. “The NLP algorithms, they process paragraphs of language. Those paragraphs of language are very important in oncology because most of the really important outcomes and the things that are happening to the patients are not in the drop downs from the EMRs, but they're actually in the notes that the doctors are putting into the record itself, to document what's going on with the patient […] They're in the paragraphs and these algorithms that read that language, they have to be as good as a medical oncologist essentially reading those.

“The code underlying artificial intelligence as we now think of it, stitches all of that together, and allows for some logic to be applied, so that when the medical professional corrects, it corrects along the entire holistic pathway. The different algorithms are learning from each other.”

Working with Google

The limitations of EHRs discussed, and given that free-text clinical notes and PDF documents remain largely invisible to algorithms that mine structured data fields for key insights into patient care, it is perhaps unsurprising that COTA teamed up with Google, John Snow Labs, and Quantiphi. Together, they are using breakthroughs in ML and NLP to extract unstructured data elements from EHRs to fuel innovation in oncology research and treatment.

“Google has this really cool product called Google Palm that is essentially a general practitioner physician,” enthused Sasu. “It's trained on National Comprehensive Cancer Network (NCCN) guidelines, it's trained on regulations, it's trained on basically every doctor's book that you can imagine. We use that, and then we also train it in oncology, and so our AI is now what I believe to be the first AI oncology specialist. That is what is allowing us to put together products like the Vista product.”

Keeping the human in the loop

Collaboration between companies, and a combined effort of AI and human beings, keeping the human in the technological loop – and the patient as focus – is essential for Sasu, and for COTA.

“The human in the loop concept is something that we can't let go of, at least at COTA,” said Sasu. “There’s a lot of pressure in terms of cost to go the solely technology route, but it should not be allowed to take these medical records and make judgments on what data to pull all by itself. That is not a good way to go. These decisions that are being made on this kind of data in terms of research and in terms of patient treatment can be life or death decisions. It's very heavy stuff. We believe in the human in the loop concept, which is not allowing an algorithm to essentially make assumptions and create logic that is not checked by a person who has had medical training. That's the number one thing.

“The second thing is, when you have this data at your fingertips […] we believe that doctors should be informed. No human, no one human person can see hundreds of thousands of patients in their lifetime. The benefit of having data at your fingertips is that you can see how other patients have reacted to certain things […] a doctor should always be making medical decisions for the patient, with the patient, and with the patient's family.”

And, outside of clinical practice, is it the same for drug development?

“Drug development, it's not quite as direct as the doctor is seeing the patient” explained Sasu. “It's more like we're trying to put together a clinical trial that will allow us to understand the properties of a particular molecule, and to see if it works and if it's safe. Exploratory medicines can be tricky because the first thing you're testing is safety. Again, algorithms and AI should not be relied on solely to make these types of decisions. They can cut bias into who is recruited into a clinical trial. They can help understand which patients to include or to exclude out of a clinical trial. Again, with a human supervisor.”

A personal purpose to a profession in life sciences

Returning to Sasu’s own goals, her personally guided journey to working with COTA, the company’s female CEO concluded:

“This is my life's work. Every time I talk about this, I am so very excited that we're actually here and we're talking about artificial intelligence as a part of decision making for a patient. Personally, I wish that my family had that when I was growing up. My grandmother, unfortunately, passed from non-Hodgkin's lymphoma, and COTA has a really great specialty in haematologic oncology. If we had only equipped her doctor with some of this information, maybe she would still be here.”