Tempus AI

These companies are built to exhibit resilience amid volatility—and deliver better-than-average growth in the long term.

The Future 50

Future 50

Overall Score

Sector

Get insights into Tempus AI including business profile, stock information, news, and rankings from Fortune.

Tempus AI (TEM): Company Profile, Stock Price, News, Rankings

legacy

tempus ai

Companies

Tempus AI | Fortune

Tempus AI (TEM): Company Profile, Stock Price, News, Rankings | Fortune

Eric Lefkofsky, Founder and CEO, Tempus
Interviewer: Dr. David B. Agus, Ellison Institute for Transformative Medicine


Brainstorm Health 2022 - Patients First: Transforming Care Through Disruptive Technology 

Conferences

Transforming Care Through Disruptive Technology 


Brainstorm Health 2022 - Patients First: Transforming Care Through Disruptive Technology 


alright, we are near the end of the day so we have to be lively, we have to tell jokes and then we have to kind of move around and dance a little just to keep the audience interested. I'm happy to do it. All right. Eric you are ceo of a company now. Well you've been Ceo of many companies but a company now called Tempest Labs, that is one of the largest sequencers of cancer samples in the country in the world. What box does that company go in? Is it a tech company? Is it a biology company? It's a care company. It's a data company. What is? Yeah, so the, it's kind of interesting because the box is a new box. So um you know, historically there were tech companies and there were health care companies and because technology hadn't really permeated healthcare in a big way, there really isn't like a combination of both. And I think Tempest is one of the first true combinations of both. Where we are certainly a tech company at heart, but we're also stooped in health care like we sequence patients we deliver, you know, clinical reports. We are bound by all the same stuff that the lab would be bound by. But it, it is complicated. I mean we have all kinds of people say which one are you and the answer is we're entirely both. So when you started off, I mean obviously we've talked on this stage before about the remarkable story of a personal experience leading to something that literally is changing how we care for cancer. Um, but one of the things that you know I watched you do was start to take medical records and try to convert them into structured data. Taking these you know medical records. You know if you don't know what a medical record is, it's a pdf, a pdf file. And so converting a piece of paper into data. So tell us about that travail. I don't think people really understand the heroics of what you guys did there with that cancer dataset. Thank you. I mean and and so so like we were talking about this before we got out here when you look inside a typical E. H. R. System, it is to quote somebody else, a bag of words. Uh and that's complicated. But we made a decision early on to migrate to these electronic healthcare record systems so fast for all the right reasons that we had to punt the tough stuff which which basically is position progress notes, pathology reports, radiology reports which are basically made up of physicians typing in what's really going on meaning what drugs am I giving the patient what response am I seeing what kind of adverse events are being? I mean, have you noticed? And so all that is just an unstructured text. And when my wife was sick about seven a half years ago and when I uh when we had 77 years and when she was sequenced, I had her sequence that like four or five places. And not only did those people not agree, but it was completely disconnected from her uh from her clinical clinical data surrounding her. Um and I thought this is just crazy like these reports have no idea if she's a male or female, if she's older young, if she has other issues going on. And so we set out the first thing we tried to do and we started Tempest was just see if we could in real time pull that data out of these electronic healthcare record systems basically in that unstructured text of words and put them in some kind of structured format. So we we picked about 100 data fields that we thought were relevant and we began structuring them and ended up building technology. That was really the first technology we built where we could structure those records in about an hour's worth of time because if you can't use technology alone you have to use technology plus some amount of human interaction humans, humans, good old fashioned humans. So today we're almost 2000 people but we have probably around the world three or 400 abstract terms that are, you know, going through these notes and making sure they typically there in pods or teams and somebody in the potter team typically has a medical background that might be a nurse or a postdoc PhD whatever. But somebody has a medical background somewhere and then your training these junior abstract Urz to, you know, go through the technology we built, which now is highly automated but make sure that the fields are accurate. It's not perfect. But what's really interesting is, um, this isn't the kind of data you'd use uh, for a regulatory filing, right? It's not at that level, but what we said to people early on, which I think was a hard concept together. If you, if I structure a million patients data even I don't even though I don't do it perfectly. I'm not really worried about the noise. I'm worried about finding patterns that are statistically significant in and of itself. There's no way the pattern's wrong. And the same concept is what google uses. If you go to google and type in the word jaguar, you get a cat or a car, you don't get the other zillion things, it could be because google basically regresses to the mean. It knows that you probably want a cat or a car. Well, it's listening in your house. Yeah, separate issues. But so we build technology that does that and looks for those patterns that are relevant. We take those patterns and we basically bolt them onto the reports. We provide positions as insights that they're free to follow or not. So creating this large structured data set very powerful. When you could have been just a service company. There are many other players in this field, not to demean them, but they're a service company, you send them tissue. They give you hear the genetics but what you're doing is creating this large structured data set over time. For what purpose? What's the goal? The goal is to make the real long term goal is to make all diagnostics intelligent. I mean really at the end of the day if you think about it, there's only a few ways to bring ai to healthcare, you'd either ask doctors to go to the Ai right? Which which obviously Watson tried or you could try to make these HR systems very intelligent which sooner may try. But these historically have been very hard for a whole bunch of reasons. And when we when we started making genomic tests in cancer smart we realized that we had stumbled into something much bigger which is there's something else doctors interact with all day long when they make decisions. They literally, the key thing they interact with is some kind of new diagnostic results in a laboratory test results, cat scan an M. R. I. A. Blood test and genomic tests. There's new data modalities often how they make critical decisions. If you can make that data modality intelligent meaning it knows everything about the patient. Then you can sneak A. I. Into the U. S. Health care system. And the goal is to make you know bad doctors good doctors and good doctors great doctors and I really think that's the age that we're coming to. Were physicians over the next Maybe five or 10 years, 15 years are going to be unbelievably intelligent. Not because they weren't intelligent before, but because they really were oftentimes like performing their craft with both hands tied behind their back. When you give them enormous amounts of data, they will typically make the right decision. Well, thank you. I mean those insults really make me feel good and I appreciate the criticism. So, you know, one of the frustrations I have with what you do is I take care of cancer patients. And if you come to see me, you know, all of a sudden you're cancer progresses and whatever treatment you're on, I stick a needle in I by stu and you're ready. You're scared to death. Your cancer is growing. And then three weeks later, I get a beautiful email from Tempest with some information in those three weeks it's chaos. So tell me a little bit about how the field could change. Could we ever get to point of care sequencing where I could use the technology and know very quickly how to treat a patient. Yeah, I mean, you can we for sure will get much faster. Our average turnaround today for our solid tumor sequencing is about 10 days and our average turnaround for liquid is about seven. But to a physician it feels about four or five days longer because that's from the time we get the samples from the time you see a patient, you got to get the sample shipment to us. We have two sequences. So you really do want to shrink that down to the US postal service's fault. Well it's actually when you think about our 10 days let's take our 10 days. Um The raw sequencing is about three right so you lose about three days just in sequencing and then the rest of it is just loading the sequences and doing the processing. So the only way you really shrink this down dramatically there's only a few ways one either you do this at point of care. So you're you're saving that week of just the back and forth. Or we create new sequencing technologies that's radically faster. And you guys at oxford Manafort here I think yesterday. But there are all kinds of new technologies that I'm sure will come over time that will allow us to do that much quicker. And companies like Tempest given our scale we're glad we adopt whatever the best technology is. We're agnostic as to which technology we use. We just need to produce a clinically valid report in a in a short amount of time. So you know one of the things that keeps being hammered into people today is that genomics are tremendously important. But so is the study of protein proteomics. The study of your metabolites metabolites. The study of your microbiome the bacteria gi tract microbiome? IX. When will we be into this multi atomic analysis to be able to help us guide to treat diseases like cancer. So I would say this is a this is a guest. But it took us essentially 20 years to kind of get to the point that Genomics matter. I think it's going to take us 10 years to get to the point that transcript almost our name matters. And then I would assume everything you just mentioned beyond that. Epic genomics microbiome IX proteomics will all be quicker but it is like a seven layer cake or whatever the cake would be. Where we had to start for all the right reasons with D. N. A. And now we're like well okay let's go a layer up and the layer up is RNA and then the layer up is proteins and we'll just keep working our way into the immune system and what's going on. But I think we had to start where we started. It's just taking a lot longer for all the reasons that you know you know better than I do Just how slow the U. S. Health care system is to really figure out how to make sense of the data and then turn it into drugs that are actually actually. So when you say drugs I mean we call them drug doble targets I sequence your cancer. Oh my gosh there's an on switch I can turn off with this thing I have in my toolbox of drugs. What percent of patients actually have a drug double target? What percent of patients when you give a result to a doctor's it change their clinical management or care. So it's a it's a it's a it's a it's a huge spectrum. So we find something actionable Over 90% of the time meaning a clinical trial. You could go on or a therapeutic that doesn't really though, Change care the way you would think about it. The change care as and I'm going to durably extend someone's life is probably somewhere between 10 and 20 Now, that might seem small. But if you're a cancer patient, it's a big deal, right? 10% is a big deal, but it's probably 10-20% of time. And and if you when we got it, I've been doing this now for six years. So it's honestly like a ski slope up or down or whichever way this goes where It's that number is growing exponentially. So I would say if we if we if I come back here five or 10 years from now, if it's 15% today, I'll pick the midpoint, it'll be 35%, then it will be 65%. Because just basically for if you're gonna think about it, every drug company is looking for an ever increasing number of biomarkers that can make their drug work better. They don't want to have failed trials. Don't wanna have failed drugs. And those typically those are molecular. So we're finding more and more molecular signatures that give rise to a good response. So by the way I know you're from Chicago and when does weird things but you normally ski down you don't ski up. Um So pharmaceutical companies biotech companies their partners to you they are they're obviously developing the drugs but your relationship with them now is predominantly in the partnership side. Yeah so I mean not being from the space I think um I had a very different view of this. I remember when we started signing up academic medical centers and N. C. I. Cancer centers to begin sequencing with us. The concept that you would licensed de identified data to pharma. It was like toxic. Um And I would say to them I don't understand what would what would you do with the data? You guys don't make drugs wouldn't you want the data to go to the people that make drugs. You want to hide the data from the only people that actually make the drugs and they're like and they kind of look at you like you know it's so I don't understand this whole world our job because we now have one of the largest database health in the world 55 petabytes or something. And growing a couple petabytes a month. But our job is to make this data data available to anyone and everyone who can help a patient. It's it is both appropriately consented meaning patients that said you can you can you can have the data and it's dee identified meaning it isn't identified. So I think the we've done a very good job in this country. It's one of the few places where I think the regulators got it right in terms of how to control identified data for patients and not allow it to go anywhere. How to make sure it's appropriately de identified. Give the definition of that. And then let's let it let's put in the hands of anyone that can use it that actually can help patients. You know, I owe you a lot. I mean, we've had patients together who literally we've changed or you've changed their life and I've just implemented what your data has said. In fact, one patient who had another company do the test and found nothing. You did it because it's actually not that easy to do genomic sensing and cancer. You found something and he stayed alive for six years. I mean, it literally is saving the life. But I have also had the other experience where I feel I potentially harmed somebody through you. Where I've done it, I've identified a target. And then the insurance company goes, we're not going to pay for that drug because it was meant for kidney cancer and they have liver cancer and this person says, do I mortgaged my house my kids future or do I die with dignity now because I can't pay for that drug and I feel guilty as hell when that happens. I mean, so do I. I think I think the I fundamentally believe, look when we started tempest and we started to realize we were going to have one of the largest databases in the world. You have to define the rules around that. Like one of those rules being we will license the de identified data to anyone that could use it for patients. We have patients we have. But we also had to decide what kind of data would do you want to hold back? And we made a decision early on that we had to allow physicians to be the arbiters of that data. Our job was to give you all the data, your job and candidly, it's a tough job. I'm not a doctor, but your job is to figure out how to synthesize it. It's the reason we don't go directly to patients. It's confusing enough for a lot of physicians to make sense of 128 gigs of data, which is typically what's on those reports, let alone the average patient. So, you know, it is hard and I think insurance companies are also having a hard time figuring out. You know, historically, You know, there used to be one subtype of cancer then there were 20. Now there's a couple 100 but more and more, we're beginning to realize that our definitions of cancer are totally archaic and they can't figure that out as fast either. So I think there um everyone's trying to grapple with this new reality of of us defining how a patient might respond more by the molecular composition of the patient, not where the tumor arose or where it is today. So you know at the same time I'm helping people. You're also a company and so you know you've had tremendous growth. Tempest you know, became from nothing to literally 53. How many of the N. C. I. Designated cancer centers are you working with now? So it's probably 52 of the 72 is pretty impressive. I mean that is remarkable and I assume at some point you will enter the public market which you probably can't talk about. But we can all look at your face and see an answer and looking at the water. Um and what will that just say? You do? A tremendous I feel you gotta chords more cash. Are their dream things you want to do that you're limited now or restrained now upon. Is there a new future we don't know about yet. No. I think, look, I think there's no new future. I think we are. Um we're in unchartered territory. Tempest has kind of gone to a place of actually bringing Ai into the clinic that I don't think anyone has ever come close to going to. So you've done it in a usable way. That's cool. Yeah. But we don't know what it means or where it goes. We we we have a fundamental belief that this is going to matter in every disease area. So I think we're always cognizant of where should we go next? We went from cancer to neuroscience and neuropsychology, infectious disease, infectious disease cardiology. And each time you have to pick a new diagnostic like our cardiology products based on an E. K. G. And cancer is based on genomics. So we have to think that through. But I'm convinced that um that are our job is clear which is to go disease every by disease area. Try to make these diagnostic as smart as we can. Work in partnership with physicians to figure out when are they stupid because some of the feedback we get from you which and others which is great as is really dumb. You shouldn't do this, change this, do that. And that's also there has to be a two way street. You know, technology and biology have never been truly married. And if you want to be a company that straddles both worlds, you have to, you have to, you know, create that kind of a relationship. So um we'll have to deploy capital intelligently to go disease everybody. So for the record, I never said what Tempest. That was dumb. I just said Goofy stupid. I never said dumb. Um you know, as we end. I mean we all in life has the opportunity of doing many things and I think everybody in the audience is tremendously accomplished. And one of the things that I've learned in my short existence is you work with people you like, because working with people who you enjoy being with, you get a lot more done. And so it has been a privilege working the two of us together because I like our time together. I enjoy it. I think we push each other and I think getting people from your domain crossing over or where big advances happen. So I am actually privileged that you're in this space and putting your time to our field, which needed your kind of mindset to make change. So thank you for doing that. Thank you. Thanks for having us. And thank you.

Tempus, the Chicago-based technology company driven by A.I. has expanded into infectious diseases, starting with COVID-19.

Disease agnostic: Using A.I. to pivot to and combat COVID-19 

Health

Tempus has expanded into infectious diseases: COVID-19.

Disease agnostic: Using A.I. to pivot to and combat COVID-19

whenever I talk to you, I'm kind of embarrassed, you know, my business, the health care business. Our data are a mess where a lot of pdf are. We're not structured. We collect samples however we want. And you come from the tech world to our world as literally. I'm embarrassed when we talk about data about the state of where we are. How do you kind of approach what we do? I mean, how did you able build Tempest literally from scratch and build structure data from a mess? Yeah. I mean, it's interesting because it is a mess. I mean, you know, and I think part of it is, um I equated to mowing lawns. You know, mowing lawns is not necessarily difficult. You just have to get out there and do the work. And I think in health care. One of the challenges is that cleaning up data, structuring data, harmonizing data is not exciting. It's not sexy, it's not, You know, it's not this thing that you really want to engage in. And so I think there's so many brilliant minds surrounding health care. But they have, you know, lots of reasons chosen to just not kind of clean up their data house. And so when Tempest started, one of the first things we set out to do was just do that hard work to basically go through and structure that clinical data harmonised the data map. The data use informatics to make sense of the data. And what's amazing is just by cleaning up the data that's inside our electronic health care records system, you can produce us as you know, incredible insights. But it's, you know, it's it is monotonous and its horror and tedious and requires both technology and people. We now have something. It's not fun, is what you're saying. There is no fun. It's not fun. We have about 1500 people, a tempest, and you know, we had something like, uh, 50 a month or something, and it's because it takes a lot of people, and it takes a lot of technology to clean this data

[ MUSIC ] So obviously we've heard throughout last year at Fortune Health, and many other meetings, there are a lot of people trying to sequence cancer DNA. How are you different? I mean, sequencing DNA, in our case we sequence both DNA and RNA, but the sequencing has become somewhat commoditized because the equipment line sequencing is so powerful. The real issue is that gathering molecular data is only half the story. If you don't gather clinical data as well, it's like a puzzle with only half the pieces there. When you say clinical data, what do you mean exactly? I mean what's typically referred to as phenotypic therapeutic and outcome response data. Which basically- That is a very cool term, by the way. It is Which basically means, who is this patient? What drugs are they taking? How are they responding? And what's their molecular composition? Cuz if you have all that data in one place, you can begin to answer some very basic questions like, why is this drug only working for 60 % of people and not the other 40 %? Who are the other 40? What medication are they also on? What's their age, are they diabetic, what mutations do they have, what does the unexpression profile look like, and that's the kind of data that should be flowing freely between not just clinicians and researchers but also insurance companies, biotech companies, and the whole system is basically broken as you know Why's that data not at their fingertips? And it's just not. [ BLANK_AUDIO ]

 The Data Needed to Treat Cancer

Leadership

Tempus CEO says how doctors need more data in order to fine-tune cancer treatment.

Tempus AI

Tempus AI in Fortune Rankings

Tempus AI stock chart

Videos