Transcription:
Patti Harman (00:07):
Welcome to the Dig In Podcast. I'm Patti Harman, editor in chief of Digital Insurance. Digital insurance has been covering the adoption of AI and other large language models throughout the insurance industry. For some time, we've been examining the risks as well as the rewards. One of AI's greatest benefits is the ability to synthesize large amounts of data to provide insights for decision-making across various lines of insurance, ranging from P&C to workers' comp, specialty lines, life and health.
Dr. Bradley Heltemes (01:18):
Thank you, Patti. Pleasure to be here.
Patti Harman (01:20):
So technology and AI are playing a more prominent role in the life insurance sector and as a cancer survivor, I am particularly grateful for the part new technology played in my diagnosis, and we know that prevention can play an important role in many types of cancer. But how is AI changing the diagnostic process today for cancer and other health risks? And Dr. Brad, I'll ask you to answer that first.
Dr. Bradley Heltemes (01:50):
Okay. Well, thanks Patti. And I guess first off, I would say when we're talking about cancer in particular, it's good to emphasize that cancer is after all a genetic disease or really it's rather hundreds of different genetic diseases. And as we start identifying cancer more by its genetic patterns, rather than that it's just an abnormal appearance of cells that are growing irregularly. You start looking at these genetic patterns and the metabolites of the cancers themselves that drive the cancer formation and the growth of the cancer, and can therefore identify targets for treatment that are specific for those alterations. So you're going to the source of the problem rather than just more broadly affecting it based on the appearance of the cells or a particular organ. But the problem is that the interplay between all these multiple factors is it's really very complex. It's really too much for us to sort through Well, and that's where the AI comes in. It really benefits from the use of AI to better understand and organize these many facets of the genetic patterns and other biomarkers that are associated with each cancer. So that's where AI starts. And then there's a lot of specific examples to that as well.
Patti Harman (03:19):
Okay. Dr. Tim, your thoughts on that?
Dr. Tim Meagher (03:22):
Yeah, well, I just go from that. Apart from cancer, Patti, AI is great with large amounts of data, and it's really that ability that is central to its application in medicine. And so the more data we can throw at it, data of different types, be it images, be it words, be it genetic sequences, the more it has traction, the more it gains value. So I could cite you many different examples. I'll just give you a couple because it's really image analysis that's one of, I would say one of the top three applications at present. And so we are discovering that AI's ability to analyze images is as good, if not better than human beings, because it can detect things that are not perceptible to the human eye. And if you think about that, then you just apply that to images in general that we generate in medicine, be it CT, be it magnetic resonance.
(04:22):
So the whole field of what we call radio mix, which is radiology. These days you add omics to any word you want to, and you sound very intelligent at a cocktail party, you see. So radio mix is really anything that's an image that generates huge amounts of data. AI can actually extract information that we were not able to do. We have extraordinary examples of x-rays. For example, being able to look at a chest x-ray, the old garden variety chest x-ray, and predict whether somebody will develop diabetes or heart disease, and who would've thought of that in the past? That's quite extraordinary. Similarly, we can look at the back of the eye at the retina, which is previously done by using an ophthalmoscope. You've probably had that examination done, but now retinal images are able to do an awful lot more. They can actually predict the likelihood of developing Alzheimer's disease or developing diabetes. So once again, not something that the human eye could ever have managed, but the AI can. I think that's potentially huge in terms of early detection, early diagnosis, along the lines of anything you can do earlier and detect earlier, you have a reasonably good chance of having an impact on this outcome. So those would be my examples, Patti.
Patti Harman (05:40):
Wow, that's just fascinating that you could use it to get across so many different areas to examine lots of different types of information. How is the use of AI improving cancer outcomes then, and what impact does it have on screening or an early diagnosis? I think Dr. Tim, you kind of talked about that a little bit, but are there concerns about over-diagnosis or even misdiagnosing a condition? Then Dr. Brad, I'll ask you.
Dr. Bradley Heltemes (06:14):
Well, I'll answer that latter question first because yeah, you're exactly right. There is an important risk of over-diagnosis of indolent tumors. We don't want to identify tumors that are really never going to be a problem for that individual. Potentially, when we see this with prostate cancer, for example, or thyroid cancers where we don't need to identify that it causes undue stress, maybe even unnecessary treatment. However, and then there's always the false positive results that can occur with the management as well. But that's the same thing as with any screening modalities that we currently have. And as Tim was alluding to, there's a lot of potential with the use of AI that we would have less of, that we'd have less of this false positive results, and we'd be able to better identify those individuals who have the non-indolent tumors that are important. So that's why it's so important that the training is done on appropriate and well-coordinated data sets.
(07:16):
But again, I think it has the potential to decrease both the false positives and the false negative results with the use of these AI assisted imaging processes. And it's not just the imaging type of things that Tim mentioned, but you talked about prevention. And AI plays an important role prevention too. You can use AI to interrogate large amounts of data to really obtain a personalized assessment of one's risk. So instead of just saying, you're 50 years old, so your cancer risk is going up, therefore you should get screening or you're 40 years old for mammography or something like that. But you could use this to better personalize who is at greater risk or who is at less risk for that matter. And then even use that to help select the proper screening modalities using polygenetic risk scores, for example, using personal data as well as the fact that you can also use these ideas of personalized screening as they're doing with AI to develop anti-cancer early detection type tests. So tests using circulating tumor DNA from cancers or other markers of a tumor that can be picked up on a liquid biopsy, for example. But you identify those individuals who are at greater risk and who might benefit from this process as well as the AI that helps to develop these types of tests in the first place, that can be useful.
Patti Harman (08:53):
Are there certain types of cancers then where you're finding that the use of AI and related technologies are particularly effective?
Dr. Bradley Heltemes (09:02):
Yes, certainly. I would say there's two sides to this coin. One is that for common cancers for breast cancer, lung cancer, colon cancer, for example, where we have a lot of data to analyze, and even small gains that can be identified with the use of artificial intelligence or other modalities can have an important public health impact. Lung cancer is the leading cause of cancer death. If you can reduce even a 10% reduction in lung cancer, deaths has a big impact overall. But the other side of the coin is that you can also start identifying patterns and potential treatments and potential risk factors for less common cancers. Ones that maybe don't have, you don't get studied quite as rigorously, don't have the data sets, but because of the use machine learning processes, you can help have a potential impact. This is better more for on the individual basis. So if you're an individual with an uncommon cancer, you may be able to identify synergisms with that type of cancer or the genetic pattern of your specific cancer and relate that to other cancers and treatments that might've worked for that cancer might therefore work for the cancer that you have that would've never been identified otherwise.
Patti Harman (10:31):
It's just amazing the way that technology is changing, how you can synthesize and collect information and then helping how to read it and use it in new and different ways. How is the use of AI affecting cancer survival rates then, Dr. Brad?
Dr. Bradley Heltemes (10:49):
Yeah. Well, we've seen cancer mortality rates declining really for the last three decades. And that's been mostly because of preventative measures. We started identifying, Hey, smoking's not so good for you. Maybe we should try to get people to quit smoking, try to become more active, less alcohol, other factors that might be impactful. But going forward, I think there's so much potential for improvements in treatment and also identifying those preventative measures that will be useful on an individual basis that we anticipate that there should be important survival outcomes. These take years to show up the changes to have that effect, but already we're seeing dramatic improvements for specific cancers or the incremental type of improvements in cancer mortality that have continued to persist even as things like the decline in smoking has slowed down some.
(11:56):
And then we have lots of treatment options that are becoming available that can be developed through the use of AI such as vaccines that actually target that specific cancer. So that you've heard of the mRNA technology that's used to create vaccines for like covid, the same vaccines strategy. The technology was actually being previously already utilized to try to develop vaccines that targeted your own cancer, not so much to prevent the cancer starting, but rather to allow the body to recognize those cancer cells as being foreign material. And so you can develop these mRNA vaccines. And again, that the identification of the targets for these vaccines is augmented by the use of artificial intelligence.
Patti Harman (12:55):
We're going to take a short break now and we'll be back in just a few minutes. Welcome back to the digging podcast for chatting with Dr. Brad Heltemes, vice president of medical director of R&D at MunichRe Life U.S., and then Dr. Tim Maher, vice president and medical director of Munich Re for Canada for life. So with more data regarding diagnosis and patient outcomes, how is that affecting insurance and does it make coverage more or less accessible or provide underwriters maybe with more accurate information? Dr. Tim, I'll ask you that one first.
Dr. Tim Meagher (13:37):
Thanks, Patti. It's a huge question, and the answer is with more data regarding diagnosis and patient outcomes, things are going to change. I think that inevitably the overall impact will be positive. The more accurate we can be about diagnosis and the better understanding we have about patient outcomes, and the more accurately we can assess risk, which is what life insurers do, and that's the kind of work that Brad and I do. Our job is to keep up to date with medical progress and make sure that is adapted into the risk assessment process processes that we use. So as I kind of look broadly at the arrival of artificial intelligence in healthcare, the question is when does that spill over into the life insurance side? And I think that the first answer to that is it has to get well implemented in healthcare first. And I would say this, we are at early days still in that respect, I think we're at the point of perhaps seeing mainly efficiencies in healthcare delivery will be the first stroke, and then I think improved diagnostics will be next.
(14:55):
So at that point, once that is widely understood and that knowledge is available in healthcare, it spills over into life insurance. So to get to your question, more diagnostic material, more accurate diagnosis, that will have an effect. So then I just look at it at a second level. The beauty of AI, as I mentioned already, is the ability to analyze vast amounts of data. And the second beauty is that the data is coming from different directions. It's coming from, I already mentioned radios. It's coming from genetic sequencing that Brad has alluded to, but it's coming from lots of places. It's coming from lab tests, it's coming from wellness devices, it's coming from everything that makes up what we call an electronic health record. So that is multi variety data of very different types. Now, the promise of AI is its ability to take all of those different types, put them into the mix and come up with answers.
(15:59):
And that's what we call multi, I'm forgotten the term, I'll come back to it in a sec. But the ability to process very disparate types of data. Then we get into the whole realm of possibly coming up with novel associations that we weren't aware of before. If you have this on an x-ray and this on a lab test, and this is your demographic profile, well hey, presto, we think you're at risk for something that we never thought you were at risk before. And that's the potential, and I'm under underlying the word potential here, but this is the promise. This is not yet the reality. This is the potential of AI to really improved diagnostic ability. And by virtue of that improved outcomes.
Patti Harman (16:48):
And I think about all of the patients who are suffering different types of symptoms and having trouble finding a diagnosis for it, it would seem to me that AI would play an important role and at least helping to maybe pinpoint what some of those situations could be. Dr. Brad, what are you seeing from your perspective then in terms of coverage and making it more or less accessible to provide underwriters with more information?
Dr. Bradley Heltemes (17:20):
I think as Tim mentioned, I think the more accurate information is going to improve the precision of the underwriting that's present. And it is still in earlier stages, but we're already seeing that. And as far as potentially offering coverage for individuals that we couldn't previously, a couple examples would be, again, in the cancer realm, we know that certain cancers, if you identify if there's any minimal residual disease, measurable residual disease, another term use that can be done by using circulating tumor cells analysis. If that is negative, those individuals have a much lower rate of recurrence rates. So if somebody with a cancer diagnosis might be able to get insurance sooner than they would currently be able to do so, or similarly, AI has been used to better classify the hematologic malignancies, the blood tumors, the leukemias, lymphomas and things like that. A very complex group of types of processes because again, all these different genetic alterations that may be present that have led to those, the changes. So it's very complex and AI has been able to be able to better classify those types of malignancies. And again, we can identify subsets with certain patterns that do much better and therefore could be insurable. Whereas as a group, we would've in the past perhaps not been able to allow insurance because of the uncertainty of their prognosis.
Patti Harman (19:05):
So it really is having a major effect in terms of underwriting for life insurance. Then does it provide you, I guess it provides you just a lot more in terms of information and answers questions that before either would've taken much longer or would've still been unknown. Is that what you're seeing then, Dr. Brad?
Dr. Bradley Heltemes (19:29):
Right. I think that's a good part of it is that you have information that allows you to more accurately identify one's prognosis, and when you have more certainty, you can be more confident in your underwriting decisions.
Patti Harman (19:46):
Dr. Tim, anything else you want to add in terms of using AI and how it's affecting underwriting in the life insurance space?
Dr. Tim Meagher (19:54):
Yeah, I think that AI is already affecting underwriting, but it's at a different level than what we've been talking about so far, which is really AI and healthcare. And so I think AI has already, if you look at the last five years, the amount, the progress of AI in the actual process of issuing life insurance policies has changed quite dramatically. There's a much higher degree of automation than ever existed, which makes the life of underwriters, the people who are the first line risk assessors much more productive because some of the more kind of repetitive pieces are being automated. And we already see the ability of AI to look at different types of personal health data that's not totally medical, if you will, and its ability to make predictions about the future already. But I think it's the arrival of the richness of purely medical information that we're talking about here.
(20:53):
That's when you start to really change things a lot more dramatically, in my view. So I think, as I mentioned already, this is still early days for a lot of what we're speaking about. There's huge excitement about it, but it's early days and will it deliver its promise is a question. Personally, I'm pretty optimistic about it. I think it's really exciting, but I think we have to be somewhat cautious. So yes, I think it is going to be a real plus in the realm of underwriting life insurance. I think it's going to enrich underwriter's lives. It's going to make it a lot more interesting. And I think the end result is going to be really good for life insurance applicants because it will allow a more accurate assessment of risk at a much more individual level than we've been able to do before. We tend to group demographics like 20 year olds are such and 30 year olds are such.
(21:48):
But ideally we would do it way more in a way more granular level. This is what Brad was speaking to about earlier about screening at a more kind of focused and relevant level. So I think that's the hope. And then lastly, I would say maybe jumping ahead here a little bit, but we very, we talk about health, most life insurance applicants are healthy. And if I were to ask you, Patti, how would you define health? You might say, well, absence of disease, maybe that's what health is. But if you think about it, health can be much more granular than that. I think one of the hopes about AI is that it would allow us to define health in a much different way that would've huge impacts for life insurance.
Patti Harman (22:42):
Yes, I can see where it would. Have you seen any trends or changes that have surprised you concerning the use of AI and cancer diagnosis? And I'm wondering if there's anything that really excites you or even concerns you and Dr. Brad, I'll ask you that question.
Dr. Bradley Heltemes (23:01):
Alright. Yeah, well, in a way, multiple questions there. I see something new almost daily that surprises me in a way because there's so much going on. And a few things, for example, I kind of touched on this previously, but the detection in tumors of potentially treatable oncogenic drivers, the drivers of cancer. And when you identify those, you raise the potential for what we often call tumor agnostic treatment. So instead of saying you've got a lung cancer, so this is the treatment you get, it's that you have a cancer that expresses EGFR and is a LK negative and PD L one inhibitor positive finding or something. It's the pattern of the cancers. It doesn't matter where in the body it is. The treatment would be for that specific pattern of change. And again, that allows us for a much broader look at treatment. And that's kind of that along with the ability to identify whether or not somebody would benefit from treatment, either pre-treatment like pre-surgical treatment or after surgical treatment.
(24:29):
These adjuvant therapies based on other markers is also exciting because if you can identify those people right now, if you have a certain stage of breast cancer, for example, you get either neoadjuvant therapy or you get adjuvant chemotherapy afterwards, a lot of those people would've been cured even if they hadn't gotten that chemotherapy. But you've given 'em the chemotherapy that it will reduce the likelihood of recurrence for a certain subset of people. If you can identify that subset of people who would actually benefit from it, and likewise those people who would not benefit from it, you can spare a lot of people unnecessary and not pleasant treatment. So that's that kind of thing excites me as well as the use of, we're actually starting to develop vaccines against the development of cancer. So not just against the cancer itself, but actually targeting in particular, this has been in breast cancer, the earliest changes that occur that might lead to that breast cancer can be targeted by a vaccine.
(25:36):
And then noting that certain drugs like GLP one agents actually might help prevent cancer and certain individuals. So lots of exciting things. The concerning side is the other part of it. And there, on the one hand I do think we have a concern is actually too much reliance on AI potentially. And the concern there is primarily you think about this, people go out and you do a chat GPT to tell me what I should be doing for this. Well, there's these hallucinations that occur are real, they give you a response that seems very legitimate and very accurate, very convincing, but that is not always the case. So you need to really make sure you understand what you're assessing. And once if people start recognizing that this isn't giving me the answers I want, it kind of erodes their trust in artificial intelligence. And therefore that's the other side of the coin. And that's that I think especially practicing physicians are really busy with things and they're used to doing things in a certain pattern and it takes a while to start trusting the outcomes, especially when they're coming out of this so-called black box with AI where it's recognizing patterns that the human eyes can't see. But that takes a while to overcome that. So you have to have that trust that these results have meaning. And so we don't want to erode that trust by having things that are not as effective.
Patti Harman (27:15):
Right. Well, and AI continues to evolve it. There seems to be a new iteration coming out every couple of months. And so then I would think that that could also impact how it is used and maybe even the results for it. What kind of impact do you think AI will continue to have in the life insurance industry? And Dr. Tim, I'll ask you that first.
Dr. Tim Meagher (27:38):
Yeah, I think I mentioned already, I think it's likely positive. The degree is difficult to assess. It depends where you situate yourself. Are you a rank optimist or kind of a cynic? And you pick your spot on that line between those two and I, of all the things I've seen in the last 30, 40 years in medicine, I think this is one of the most exciting. So I'm on the optimistic end. I actually think that this technology, the maximum potential of this technology would be to uncover new, what I call epistemic knowledge or foundational knowledge, which is what everything is built on, our ability to diagnose and anticipate, predict and so on. And we have a foundation of knowledge that we use today. And I think if it becomes much more sophisticated in the future, it just allows everything by extension to improve diagnosis, treatment and so on.
(28:45):
And so the things, I'm flipping back to the previous question about things that excite and bother me here, Patti, but I look at the ability of AI to predict new protein structures and therefore allow new drug development. I think that's colossal. I think the ability of AI to finally help us map the entire human cell sets 37 trillion human cells. We actually don't know how they talk to each other or why they're all in the position that they are or how they move around when they get unhealthy. And so once you start to understand those things, that's what I call foundational knowledge. Then you blow the lid off what I think what's possible. Now that's a wish. Okay. So you have to be very cautious about how, I don't know. I hope that that's what will happen. So I think that epistemic progress will be determinant here.
(29:46):
We will make gains, there's no doubt about what we're already seeing. But the real big progress is what I'm mentioning there. I think when our understanding of life and so on really improves. And then just to add one to Brad's concerns, I think one of the problems about AI is non-represented data sets. And I think we should mention that if you build a data set based on 1,000,050 year old white males, can you use the information from that to underwrite, I dunno, an Indonesian octogenarian or somebody from the Romani traveling class, or you have to understand the origin of the data and you have to make it widely representative of where you're going to apply the information. And that is a real concern at the moment. I personally think we'll overcome that one, but at present that is another problem.
Patti Harman (30:53):
Dr. Brad, what kind of impact do you think that AI is going to continue to have in the life insurance industry?
Dr. Bradley Heltemes (30:59):
Yeah, the life insurance industry, I mean there's certainly the medical aspects. So the medical underwriting component as we've been talking about a lot, but the other side is just simply how we're going to use it in our underwriting process. And this is where I think one of the big things there is this whole, the idea of being able to utilize machine learning to tap medical information, especially like electronic health records and synthesize one's risk profiles over a period of time. So you're looking, again, as Tim said, it's what represents healthy. It's representing not just a single point in time when that person applies for insurance, but rather what's their health status been over time and how does that correlate with potential outcomes? And of course, that can be utilized for that individual's benefit to improve those outcomes either by finding better ways to manage their risks or to know when to be testing for screening for other conditions.
(32:09):
And I think the other thing that we have to watch for here is that on the cancer side of things again, is that the use of cancer genomics to better identify the targets for treatment and for diagnosing the cancers, could eventually also lead to significant change in the classification of cancers. So we have to stay on top of that. We have to understand that what we currently call leukemia might not just be a leukemia anymore, it's going to have a much broader category. Or again, it may be tumor agnostic. It may not even be that it's a specific organ related. And we need to understand that, especially for things like critical illness products that we underwrite in Canada, but for any other insurance product as well.
Patti Harman (33:06):
Are there any areas that you're watching that you think are still untapped in the life insurance space when it comes to using AI and Dr. Tim? I'll ask you that first.
Dr. Tim Meagher (33:17):
Yeah, I don't know about that. Listen, it's early days in terms of us really using an electronic health record to its maximum benefit. And we use electronic health records, a kind of a really important piece of information when we're assessing risk. Once it becomes a much more richer resource of information, then I think that would be a major change. I think we have our eye on the horizon. I think we know we have spotted most of the things we think are coming up. So yeah, untapped. Yes. I think we would probably answer this question differently in two or three years time, but right now, mainly untapped, but with huge promise.
Patti Harman (34:10):
Dr. Brad, what are you seeing? Anything that you think that might be untapped in the life insurance space area?
Dr. Bradley Heltemes (34:17):
I think just mostly as I just mentioned too, I think again, it's how we can be using it in certain aspects of underwriting. And we're all looking at that. It's more matter of the speed of implementation and how much we can get appropriate, again, appropriate data sets and appropriate analysis of these data sources. So that correlates with what we need for insurance underwriting.
Patti Harman (34:49):
We've covered a lot over the last 45 minutes or so. Is there anything that I haven't asked you that you think our audience should know or understand about utilizing AI in the healthcare or the life insurance industries? Brad, I'll ask you that first.
Dr. Bradley Heltemes (35:05):
Yeah, we did cover quite a bit, I think, didn't we? Did. I guess from a healthcare perspective, the one thing that I found intriguing is the idea. Medical studies take a long time. It takes a long time to develop drugs. It takes a long time to assess the outcomes for treatments and for assessing risks. But AI can actually be used to aid in designing studies such that these studies are more efficient in how they use our resources. And then interpreting those results such that you can get adequate and well correlated results in shorter periods of time. So I think that's one area that I'm watching at least, because I think that has great potential for accelerating certain processes if we can get the buy-in is the one and if we can afford it. So I think that's the other thing we have to watch a little bit is a lot of these things are not inexpensive cost money, and you have to have that and also will that with potential for disruptions in research fundings and things like that that can leave some degree of uncertainty whether or not we could move forward at the pace we have been and achieve that optimism that we both Tim and I share highly.
Patti Harman (36:34):
Dr. Tim, any closing thoughts or something that you think our audience should know maybe that we haven't discussed yet?
Dr. Tim Meagher (36:41):
Probably not. Patti. I would just say the promise is huge. I think we are going to have to be patient. The timelines are long here. This is not a quick thing. This is going to take multiple years. There are substantial hurdles and we are clearly in a very hyped up period at the moment. And I think our job is to be try and be as realistic as possible about, well, the promise is the promise. It's there, but realistic on what actually is happening and how we can apply that in a sensible way, adapt it in a sensible way to life insurance.
Patti Harman (37:19):
Well, thank you so much Tim and Brad for sharing your insights with our audience. Thank you for listening to the Dig in podcast. I produced this episode with audio production by when Jeanmarie. Special thanks this week to Dr. Bradley Heltemes, and Dr. Tim Maher for joining us. Please rate us, review us, and subscribe to our content at www.dig-in.com/subscribe From Digital Insurance, I'm Patti Harman and thank you for listening.