Early Detection, Part 3: C the Signs

Chuck: Welcome to the Good News About Cancer. I'm Dr. Chuck Ryan.

In each episode of this show, we talk to one of our colleagues about a promising development in cancer. We break down what's new, why it matters, and how it points the way forward.

Bea: One in two patients will experience a missed or delayed cancer diagnosis. And we need more innovative approaches. Like when you go back and you look at a patient's record and you think, ‘Oh, it was really obvious. It was actually really obvious, right?’ The signs were there, but we couldn't see them.

Chuck: In the past few episodes we've been talking about the good news in early cancer detection, where advances in technology are allowing us to find the signs of cancer earlier and earlier, which of course, we hope will allow us to treat it earlier and earlier and lead to better outcomes and save, we hope, thousands of lives.

I n our last episode, we heard about a new blood test called Cancerguard, which screens for over 50 different types of cancer from a single draw of blood. Before that, we heard from one of the researchers whose work in the lab is helping to push the needle on very early detection technology by demonstrating that cancer DNA is in the blood many years before a patient would ever develop symptoms or something that we can see on a CAT scan or a mammogram or something like that.

Well, today we're going to hear about how artificial intelligence can be used in early detection by combing through the medical records of tens of thousands, if

1

not hundreds of thousands of individuals, and finding patterns that may lead to an early diagnosis of cancer.

The company is called C the Signs. The letter C the Signs. They use AI to analyze medical records and flag when a patient may be at risk for an underlying cancer. It's already been implemented in the NHS or the National Health System in the UK, and they hope to soon have it available in the United States.

I met the founder and CEO of this company, Dr. Bea Bakshi, several years ago at a dinner and I was really taken by her story and the message and the work that they were doing. We've kept in touch over the years and I was really delighted to see the progress that this company has made and even more delighted to have an opportunity to sit down with her and talk about what a massive difference this technology can make.

Here's that conversation. __________

Chuck: So Bea Bakshi, founder and CEO of C the Signs. Thank you so much for joining us on the podcast. Really delightful to have you here and to hear about all the really interesting work you're doing, that could shape the detection of cancer for the future.

Bea: Thanks so much, Charles. I'm so happy to be here.

Chuck: I'd like you to start by telling us a little bit about your background and how you got interested in this topic. And then we can talk a little bit about the really very, very interesting work you did at the NHS in the UK, and what that has shown us about how we can improve our ability to detect cancer early, which we know can lead to better outcomes.

Bea: Yeah, absolutely. So, I've been a primary care physician for a number of years, and I also worked nationally in public health policy. And then most recently was in digital health policy across the British Medical Association in the US. And my role there was really looking at the digital transformation of primary care.

And through that I did a lot of work on looking at digital health innovations, creating frameworks for procuring digital health technologies through

2

regulation, data, AI, and quality for both patients and providers, as well as things that we just need for the backbone of digital health, like interoperability, compliance, regulation, and so forth.

And whilst I was doing this, I actually, completely coincidentally, was doing a shift in the ER. Happened to be a night shift and happened to pick up one of the clinical notes of a patient who had come in, kind of floridly jaundiced, and quite unwell at the time.

And it transpired that he had seen his provider over the course of six months back and forth. He felt that his symptoms continued to deteriorate, they continued to progress. But he was consistently reassured and told, actually, there's nothing to worry about. And now he was sitting in front of me in the ER, really unwell, and I had to give him the diagnosis that it's likely he has a primary pancreatic cancer that has metastasized, um, everywhere.

It was a really, obviously, a quite a sad conversation to have. And one of the things that really surprised me about this particular patient was his very stoic response to the diagnosis and saying, ‘I actually don't care about the cancer diagnosis. What I care about is that I did everything right as a patient, but the system failed me. I kept going back, I kept saying I was worried, but the system failed me. And now you're telling me that I've got a cancer that's incurable. And what if, what if something was different?’

And when I went away and looked at the data and tried to see, well, how do patients get diagnosed with cancer? I was really surprised to see that outside of screening programs, which cover maybe three or four cancers nationally, no other cancer actually has a dedicated path to diagnosis.

So patients are diagnosed completely by chance without this formalized route. And that essentially, you know, set in me at least a motion to say, how do we solve this problem? And create a solution that goes really upstream to identify patients across all cancers in the earlier stages.

Chuck: That is a really heartbreaking story and I'm sure you'll agree that this is, every patient's nightmare, that the doctors are missing something terrible, or that a patient has a diagnosis missed, or an opportunity to render curative treatment earlier is missed. So, trying to solve a pretty significant problem.

Were you able to figure out, you know, the proportion of patients who are diagnosed with a cancer, who make contact with the healthcare system ahead of

3

that diagnosis, such as your patient who you met in the emergency room did? How frequent is this problem?

Bea: So actually the data across the UK and US is quite similar. But patients on average will have three to five contacts before they're recognized at risk of cancer. And about 20% of these patients will end up in the ER, like my patient, with a complication of the cancer as the first presentation where their cancer is diagnosed. And we know with those patients, less than 40% will survive to one year.

So it's a very, very common problem. And partly it's because cancer's not one disease. And we often, you know, think it's the same as like diabetes or heart disease, but those conditions are super measurable. But with cancer there's like over a hundred different types of cancers. It's a cradle to grave disease, so it affects anyone at any age, any point in time. And the other thing as well is that in the early stages of cancer, they mimic normal conditions that aren't that serious.

So it can make it very, very difficult for providers in first contact settings where actually they don't see that much cancer on an individual basis – the average provider only sees eight new cases of cancer per year. So when you think about that, even as individual physicians just based on human-led decision making, our own pattern recognition is not getting the volume we need to be good at detecting the cancer independently.

And you know, there's big studies in Europe which have quantified that to say that providers have approximately a 54% sensitivity for cancer, which marries up that one in two patients will experience a missed or delayed cancer diagnosis. And we need more innovative approaches, like when you go back and you look at a patient's record and you look back, you think, ‘Oh, it was really obvious. It was actually really obvious, right? The signs were there, but we couldn't see them.’

Chuck: So, let's talk about the solution you've created. C the Signs is an AI generated platform that helps physicians overcome some of these cognitive gaps we have in diagnosing patients. How did you develop this, and how did you test it?

Bea: So, C the Signs is an AI platform that analyzes clinical data and patient reported data to identify if patients are at risk of cancer. I think what's really important is we're looking for undiagnosed cancer in the population. So it's not about future risk, it's actually about saying that you are at risk today of cancer.

4

We will predict what your tumor origin is, so what cancer type you're at risk of, and then we will navigate into the best care pathway and match you based on the type of cancer that you're at risk of.

And what we did when we first started out is, because both myself and my co-founder are physicians, we took a very evidence-driven approach. So we did a really comprehensive literature review in looking at essentially, studies that had looked at: how do patients present with cancer?

And looking at the presentation where independent, peer-reviewed studies had validated to say: these data points, these factors, these symptoms, these signs, have a predictive value for this cancer type. And we essentially curated all these papers together. And then we built tumor-specific models across all cancer types to say, actually, ‘this is what the breast cancer model would look like.’ And then within that, obviously there is nuance for different types of breast cancers. Same with colorectal cancer, lung cancer, and so forth.

And we implemented this within the National Health Service. So within the National Health Service, they have electronic health record systems similar to the US. And unlike in the US they're not a fragmented record. They're actually a fully comprehensive longitudinal records, where for most patients, there's about 30 years of data to be able to analyze.

So we implemented these models in a real world setting within the NHS, and we first started out in just implementing the models as provider decision support. So as patients would come in, and you can imagine you go to see your provider and you say, ‘Oh, hi, I've got a cough/cold, or I'm feeling a bit fatigue. I've, I've noticed something different, et cetera.’

As soon as the provider launches your records, we conduct a preliminary assessment, and then we'll let the provider know if there's evidence of cancer risk to conduct a more comprehensive assessment. And then as the provider launches our system, through that, they'll input any new signs or symptoms that the patient presents with, and then they'll click, and within kind of 30 seconds, end to end, we'll service an outcome to say: this patient is at risk of cancer, the patient's at risk of pancreatic cancer, and they would benefit from a pancreatic CT scan.

5

And then we would actually create the workflow where all the provider needed to do is say yes, accept. And we would actually refer the patient straight to a pancreatic CT there and then, in less than a minute, end to end.

So we did a lot of studies in evaluating and, and so forth. But one of the things that was really key to the way we implemented this is that for every patient that went through the system, whether we said they were at risk or we said they were not at risk, when we predicted tumor type, et cetera, we would follow these patients through to then capture the outcome to say, ‘Well, what actually happened at the end for this patient? What was the end result?’

And through that – over 7,500 patients had a cancer diagnosis – we had a 99% sensitivity for cancer for those patients going through, and most critically, a 99% negative predictive value. So that was a really exciting study we did. And where we are now is we've had over half a million patients go through our system. Every five minutes in the NHS patient has a risk assessment by C the Signs.

Chuck: Really!
Bea: Every 22 minutes, a new patient is diagnosed with cancer through C the

Signs. Chuck: Wow.

Bea: So we've detected over 65,000 patients so far, and really importantly, over a hundred different cancer types.

Chuck: You mentioned at the outset that primary care physicians using their simple, you know, cognition basically have a 54% sensitivity for the detection of cancer, which is only slightly better than a coin flip, frankly, right? Let's put it that way.

The cell-free DNA, some of the blood-based, pan-cancer screening tests get that number up into the sixties. If you look at your number, the number of patients who have a cancer, where the algorithm would've identified them to be at risk. What does that get us up to now?

Bea: 99%

Chuck: So what you're saying is in the population, for those who have an underlying cancer, 99% of the time your algorithm picked it up.

6

Bea: So our sensitivity is 99%. So with this study and since then, we've seen that if there are a hundred people in the population with cancer, we can find 99 of those patients, essentially, through the algorithm.

Chuck: Does the physician need to be going in and filling out tables for the algorithm, or is it pulling it automatically from the electronic health record?

Bea: So it's fully integrated with the electronic health records. So we can analyze, structured, unstructured, free text data, lab results, things like that, historic data, past medical history, you know, your medication history, your past clinical notes. All of these things actually paint a picture when you analyze them together, to be able to say: in addition to any new signs and symptoms you're experiencing, if there's any evidence of cancer risks. So the algorithm does all of that.

And actually more recently now we've been doing a lot of work where, based on the impact we've had in the provider office, we recognize that, you know, whilst the system is having a great benefit, it still relies on the patient to turn up to the provider office. Whereas many patients, we know, find it hard to get a provider appointment, take time off work, or often we know that patients will sit on their symptoms in the community.

And I think I've seen something between, patients will wait between one to three months, actually, before even, you know, taking action on symptoms they're experiencing at home. So part of what we've really focused on is like, well, how do we meet patients where they are? How do we go further upstream to intercept patients?

So, we've expanded actually in doing what we call population screening. So now even before the patient has attended the provider office, we can scan the electronic health record, identify patients who are at risk, and then we actually message patients directly and we will surface a tailored and personalized questionnaire to each patient to elicit any signs and symptoms that match the cancer types we've identified within their records.

And depending on their answers, if they're at risk – in the UK at least – we are triaging them straight to a diagnostic. So these patients don't even need to go see a provider in those cases, they can go straight to diagnosis. And in some instances, patients actually can just opt in to do a self-assessment for certain cancer types in the UK as well. So direct to consumer. We've been working in the NHS as well where they can assess directly for certain cancer types, and again, be booked straight into a diagnostic.

7

Chuck: Let me pivot a little bit and now say that you're working to implement this in the United States healthcare system. A slightly different system – some would say a vastly different system – but united by the fact that we are now largely on electronic health records. I know earlier this year the Wall Street Journal announced that you had a large financial backer in the United States to bring this here. Tell us a little bit about your plans here.

Bea: Yeah, absolutely. Cancer's a global problem, and we definitely have global ambitions to help as many patients with early detection and survival as possible. It's the second leading cause of death in the US. Very similar, when we think about pathways to diagnosis, I think the University of Chicago published a paper a couple of years back on the fact that 86%, of cancers have no pathway to diagnosis. So only 14% of cancers are diagnosed through screening. So you can imagine just the burden of the problem when it comes to cancer diagnosis.

And I see the use case for Cthe signs in the United States – even more pressing compared to the UK, which is an integrated system, fully connected, fully plugged in – where things are so fragmented that there is so much onus on patients. So, you know, if someone tells a patient you are at risk of breast cancer, that patient needs to make sure that they book the mammogram that they go to attend, that then when they chase the result and they chase the radiologist and then they go back and say, ‘Well, what was my result?’ And they have to chase to find out the reimbursement.

There isn't necessarily the same safety net that exists to make sure that patients move along through the system. So how do we democratize access to early diagnosis to make it available to all 300 million patients in the United States? And not make survivability for cancer almost an insurance lottery. You know, the biggest alignment I think, is the fact that when you catch cancer early, it is significantly cheaper to treat.

Chuck: Right.

Bea: So when we think about it, it's almost looking holistically at cancer to say it needs the investment upstream into early detection infrastructurally. And it's not going to be one company or one solution. It's going to take an army of, you know, public sector, private sector foundations, and most fundamentally policy to drive this forward.

To say: actually this is something we need to invest in as Americans, because it's the right thing for patients and it's the right thing for the economy actually as well when it comes to early diagnosis.

8

So we are really excited, having the backing and support of Khosla Ventures and Vinod Khosla, who really believed in this. So one of the key things we're really excited about is that in the US what we actually plan to do is launch with a direct to consumer app.

So we've worked out how we can actually consent patients, how we can access patient's medical records, and be able to do a real time risk assessment within minutes for patients on their phones. And then work with health systems to navigate and partner with them so that patients can be booked into provider settings for further assessment if they're deemed at risk of cancer.

So that's really, I think, part and parcel in what we're trying to do, which is access reach early and fast when it comes to patients in the United States, and we're super excited about that.

Chuck: So when you go with the direct to consumer app, it would be the health system or the insurance provider for the patient, or their employer perhaps, might offer this as some sort of a benefit or something that they could have. And then they could basically fill out the data from time to time, and that would be brought to the attention of the primary care physician or the primary care system? Is that, is that sort of the idea?

Bea: So it's even more simple than that. Once a patient downloads the app and we consent them, we can actually then constantly just monitor in the background of the electronic health record 24/7 for cancer risk.

Anytime they see their provider, anytime their medical records are updated, we'll get a notification and assess that encounter to see if there's any evidence of cancer risk. So again, going to the idea of: how do we make cancer detection proactive? That we can sit there in the background quietly assessing, and then if the patient's risk changes, we'll contact them. We'll ask them a few questions to evaluate, and then we'll navigate them into a care setting to say, actually you're at risk of this cancer, this is what you'd benefit from.

And really also reframing that narrative. Because if we can get patients earlier – and I say this a lot and it doesn't necessarily make sense often when you say this, but we need to rebrand cancer. We need to create the hopefulness that cancer is survivable cancer can be detected early. Even really rare cancers, we can detect them earlier. And there is a solution, there is a cure. It's just the case that we're always thinking about you know, very downstream opportunities rather than if we move upstream, patients can survive. And we need to say, actually, detection is good. Detection early means survivorship, right?

9

Don’t be scared, actually, present early and, and there's always something that we can do about it.

Chuck: Well, that's one of the things we're trying to do with this podcast in general is rebrand the thinking around cancer to generate hope. And for individuals who do not have cancer, to understand that there is so much that has changed, that fear and that despair, although, it can still be present, obviously, is not as universally present as it once was.

Tell us a little bit about how a listener might get access to this or when they might be expected to be able to get access to C the Signs for their own use.

Bea: Yeah, absolutely. So we're launching a wait list in November, and that will enable patients to sign up and essentially be the first to access the application.

So for more information, please follow us on social media. We're on Instagram, LinkedIn, TikTok, all the channels, as well as just, you can come onto our website at www.cthesigns.com, and please get in touch.

Chuck: Well, thank you Bea Bakshi, CEO and founder of C the Signs, that's the letter C the signs, a cancer detection platform algorithm, that has been integrated into the NHS system in the UK and coming to the United States.

Bea: Thanks so much, Charles, for having me. It's been a pleasure. __________________

Chuck: Well, this was a great conversation with Dr. Bea Bakshi, CEO and founder of C The Signs. They've really done a phenomenal amount of work in the UK; they've diagnosed over 65,000 patients with cancer at a – presumably – much earlier stage than these patients would've been diagnosed otherwise. Also important is the sheer number of different types of cancer that they've been able to diagnose, which I think is no small feat.

We have a lot of early detection for the very common cancers, we talk a lot about pap smears and mammograms and PSA testing and colonoscopy, but many cancers do not have an early test, and they're not well known even by many doctors. And so the opportunity to detect them at an earlier stage offers a real advantage, potentially, to patients.

Well, this wraps up our three episode series on early detection. We did that because we think early detection is really one of the keys to solving the cancer

10

problem, because so many cancers are more treatable and more curable when they're caught early. And we're really delighted to see progress in this area, and we think this is very good news.

And we're going to see more good news along these lines, so pay close attention to your feeds – there's going to be more good news about cancer coming. And thanks for listening to the Good News About Cancer.

I'm Dr. Chuck Ryan at Memorial Sloan Kettering Cancer Center in New York.

The views we express on this show are our own and do not represent the views or opinions of the institutions where we work.

Our production partner is CitizenRacecar. This episode was produced by Anna Van Dine with post-production by Alex Brouwer. Thanks to Lilly for supporting the show.

There's a whole lot more good news to talk about, so make sure you subscribe to this wherever you listen to your podcasts.

We’ll be back again soon with some more good news about cancer.