data.bythebay.io: Panel: Living Data
Recording: data.bythebay.io: Panel: Living Data
Welcome to the closing panel of the Life Sciences and UX Day of Data by the Bay. It was a long week. We had 175 speakers, which makes us in the Guinness Book of Records the longest and the deepest open source technology conference on data in multiple verticals, but not the biggest because we don't want to be like giant conferences. We want everybody to be able to meet everybody else. And we had this really good community feeling, and I think we succeeded in that. So today we're the two verticals. The conference is structured as a matrix. We have verticals and we have horizontals
So the horizontals are algorithms and pipelines and machine learning techniques which kind of tie together all the fields that are applicable everywhere, such as deep learning and Spark for data engineering, but verticals differ. And so today we actually ventured into completely new territory for us, which is life sciences and UX, and we kind of bravely and boldly drove into the unknown, and we met a lot of new people. So most of them actually asked, you know, why didn't we know about this conference earlier? And they promised to bring next year to come themselves and bring all of, you know, hundreds of their colleagues. So we really plan to be, you know, in a completely full venue next year, but since everything is recorded, we hope that the world will spread. So we're excited about this. So today we have a great panel. We have folks who actually brought a lot of people, who are not able to bridge across UX and lab sciences. And before I let folks introduce themselves, I just want to mention a topic which came up that a lot of decisions in healthcare are taken upon data, right? So all these companies which want to build products for doctors and patients, they eventually need to present data to these people who are not data scientists
And then these folks make decisions which concern life and death and maybe care and comfort and pain based on a little rectangle, right? It may be a watch, it may be a laptop at most, you know, it's something like this. But essentially, you know, the way you present information on the rectangle with colors will affect life and death of a lot of people. So I think it's really putting a lot of interesting challenges in front of us. And I kind of want to see how we can tackle it. So the way it's going to work, we'll let all the panelists introduce themselves and focus on the areas of interest. And then about halfway into the panel, we'll ask you guys to ask questions. So let's proceed. I'll ask probably Jeff to start introductions from that end
Jeff Lerman, Hi, my name is Jeff Lerman. I work at Chiagen. I used to work at Ingenuity Systems before it was acquired by Chiagen. I'm a biomedical ontologist. And I've been doing that for about eight years now. Background in the biological sciences before that. And I think I'll stop there and let John take over. Cool
Thank you. Thanks, Jeff. I'm John St. John. I am a co-founder and bioinformaticist at Driver Group. We are a company that processes tumor samples from patients with cancer. We try to find out what the mutations are inside of the patient's tumor, what drugs that patient might benefit from most. And then as we match patients up with drugs, we try to identify where there are needs for new drugs in the market
So situations where patients don't currently have a better option than either chemotherapy or radiation. And then we actually push those drugs through early phase clinical trials ourselves. So a little bit of everything in the cancer therapeutic space. So yeah, thank you. My name is Hunter Whitney. I'm a user experience design consultant. And I also write about the topic a little bit and evangelize a little bit about it. I love the life sciences
I have a little biology background. And I've consulted for companies like Genentech, actually the Monterey Bay Aquarium. I love biology. And I'm very interested in, I think, like she was saying about making visual communication clearer so that people can make better informed decisions when there's so much complex decision making that's involved these days. Daniella Perlroth, Hi, I'm Daniella Perlroth. And I lead the data science team at Lyra Health, which is a behavioral healthcare technology startup. We sell a solution to large employers that elicits needs and preferences of people in the mental health and substance abuse and emotional health space and then matches them to treatment plan and then a provider solution that's best for them. We also then track and monitor in terms of outcomes and feed that back to improve our system
So I'm Priya Joseph. I've done multiple life sciences projects, next-gen sequencing and translating big data into medicine. You guys saw the chairman of NUMEDi do the talk this morning. So I generally deal with data. Okay, thank you. So maybe I'll pose this original question to all of you. So which data do you find most useful for healthcare practitioners and patients and doctors? And how do you make it consumable by end users? And what is difficult in that space? Like which problems can you overcome with better representation of data? Daniella Perlroth, So I think the challenge really is to understand probabilities and statistics, for example. I think human beings in general are not good at it
And within that there are people who are even worse. And these are, as you point out, life and death decisions. So how do you take things like a simple and what's seemingly obvious is like a mammogram result, a positive or negative, and actually put that into the right context so the person understands how likely is it that I've got cancer, I've got breast cancer if I've got a positive mammogram. And I've actually written about this and sort of done some sketching to show it in more natural numbers, natural frequencies versus probabilities and statistics. That and the thing also is that by showing it to both patient and doctor at the same time, that helps both of them. Because a lot of times physicians, I think, don't, are not necessarily that good at probabilities either. And so they see it in a simpler way. They actually understand better what they're talking about than they would have if they had not gotten that help
The data we find most useful in terms of provider matching, which is one way of interpreting your question, is data on what providers actually do, not what they say they do. So we pull that from many different sources. I think some of the best sources on the pharmacy side are the data there is a bit better in terms of what people actually do. The claims data is a great source as well. I think also finding the way when providers speak about their practice, that can be a clue as well. And the key there is to try to discern from what they're saying or how they describe it, how relevant it is for kind of the patient you're trying to profile and match on the other side. So something that we do when we're designing our reports, which so far have gotten some pretty good feedback from patients and doctors, is we start by figuring out why are these people ordering these tests? You know, what is the take home here, right? That needs to go up front, because if you start throwing chart junk at people and information, they quickly get lost. We've noticed that doctors and patients, for example, don't tend to be much different in their level of sophistication when it comes to understanding some of the latest research that's out
You know, it's really hard to stay on top of everything. So figuring out what it is that people want to get out of the report and then making it just dead simple to get at that data. It's really challenging, though, for different kinds, obviously. So one of the things that we focus on a lot is supporting molecular diagnostics. And, you know, one of the sort of basically the most basic piece of that is being able to present information about what is known for particular genetic variants with respect to their effect on disease and disease risk. And in order to do that, it would be nice to know what reports have been published in that vein. And we spend an enormous amount of time locating those articles in the literature and pulling data out of them. And so far, that's essentially a manual, at least at the last stages, it's a manual process
And at the earlier stages, it's just a very difficult process. So what would help would be a more standard, you know, public database that contain that kind of information. So I'm just speaking specifically of the kind of publication where someone has reported on at any level of detail, this variant seems to correlate with this disease or seems to have an effect on this disease, possibly with, you know, with additional detail. So those things are surprisingly difficult to find and, you know, of great interest data. So that's a very important thing. And I think that's a very important thing. to think about the fact that we're going to be able to do that. And I think that's a very important thing to think about the fact that we're going to be able to do that
And I think that's a very important thing to think about the fact that we're going to be able to do that. And I think that's a very important thing. to think about the fact that we're going to be able to do that. And I think that's a very important thing to think about the fact that we're going to be able to do that. And I think that's a very important thing to think about the fact that we're going to be able to do that. And I think that's a very important thing to think about the fact that we're going to be able to do that. And I think that's a very important thing to think about the fact that we're going to do that. And I think that's a very important thing to think about the fact that we're going to be able to do that
And I think that's a very important thing to think about the fact that we're going to do that. So, I want to kind of tell a little story which is different kind of UX and maybe hear what you think. So, when I started in computer science, my first project was a PhD project was actually medical informatics. I worked with Kurt Langlotz, who is a radiologist, and he is one of the first graduates of MD-PhD program from Stanford. And so his thesis was using Lisp. He basically did the early natural language processing on medical decision making. And so his problem was an vagina surgery decision, which was basically based on the patient preference for pain and risk. Right? And essentially the equation was very simple
So you have certain probability of operation going successfully, which mostly will succeed, but you know, some small percentage will fail. And also there is pain. So basically you can live many years with pain. Right? Or you can subject yourself to a surgery, very low risk of failure, in which case you may die. But if it succeeds, then you will live without pain. Right? And so the equation is A times B is greater than C times D, which is very simple. But apparently the doctors could not understand this. When they looked at the equation, it didn't speak to them
So what Kurt did in his thesis using Lisp and, you know, early NLP, he unrolled the formula. So basically it produced from A times B, so it received the utilities from the patient and probabilities from practice. And then it basically produced the big chunk of tech, which I just said earlier. Right? Given that your probability of success looks like this, and your preference for pain looks like this, we advocate that. Right? And apparently it was a major revelation. The doctors were very happy with this. Right? So, and obviously to somebody in computer science who is kind of liking math, it was very big news. Right? So I wonder, in your dealings with healthcare at all levels, what did you find most surprising as quantitatively minded people? Right? What is kind of most unintuitive and which requires a UX like this? Because basically this is a UX solution, which requires, you know, natural language processing and textual presentation of information in a different mode
I think it depends on the problem. Exactly. I think there's many ways to answer that question. So I can maybe give you a few examples of mine. First of all, I'm a physician. I stopped practice about five or six years ago. So I know medical decision making, and I've done many different kinds of decision models and decision making as well. And so I really actually thought that doctors made decisions based on weighing A and B's and C's and D's
And then I took a job with a health economist about four and a half years ago, who did, worked with the universe of Medicare claims data at that time. And when I first started that, I thought, oh, no, no, no, physicians make decisions based on the best interest of the patient and all that. And I got to tell you, every time I've said that, I was proven wrong because they follow payment policies and incentives very closely. So and of course, that's at the margin. So anytime you change actually an incentive, so you're combating so many things, you're not just combating their calculation of that patient for what's best for them, which is oftentimes really hard for them to, you know, auto compute in their head, you're also combating the incentives of the system that make them at a subconscious level, maybe even more likely to do A versus B, because A, which happens to be the stent for the patient with chronic angina, even though it won't extend their life, very, very lucrative to them. So the problem you described, for example, is one that I've looked at personally myself. And you can go and you can find the best studies in the literature that will be trials on the chronic anginal pain patient, which show no improvement in survival. And you can look at Medicare's utilization of stents and PTCA and they don't make a bump, not like a bump in utilization
And so in that situation, the incentives are just overdriving a lot of what's coming out in the literature or the data or whatever. So I would just mention that I think anything you can do on the to make a physician present to a patient a more rational but also palatable decision, they actually really do want to make the right decisions. But again, they're often combating, you know, a system that's messed up, fee for service, not fee for value. And so and just simply, you know, cognitive biases and misunderstandings and not being up to date. So the easier you make it for them, which is a UX thing that I'll let you talk about in a second, and make them have the conversation with the patient, because the patient on the consumer side is also driving some of this too. They keep coming back to the doctor and they want something. And that's really also a very motivating incentive for a physician to do more. So anyway, I think it's a great question
And we all probably have a story or perspective. I do, and maybe slightly different. I mean, my perspective is that I really would like to see the patients being presented with clear and simple stories of the trade-offs and let them make the decision more. So it's more a conversational mode. And, you know, the A and B and C and D for each person is going to be different. They're not weighted. A and B and C and D are not always weighted. I may care more about B than the person next to me
And maybe I'll show you later, but there is something I use in some presentations. And it was a kind of a decision matrix visualization that talks about the benefits and problems of two different types of lap band surgery. And, you know, one worked was more effective. And this one was, you know, you got lost more weight. And this one was faster, et cetera, There was a column, two columns. And then the last one was like the chances of being fatal. And then they actually showed, you know, showing one out of 10, you know, those little ideagram types of things. And that actually made the difference for a lot of people
Even if it's a small amount, they saw that one person. They said, I don't care if it's one out of 10 or 20 or 50. I don't want to be that one person. So I'll go for the less, you know, presumably effective thing. And you know what? That's fine. And I think that's the thing. The patient should make the decision and an informed decision. Let the doctor more be a facilitator of that sort of conversation
I think. Sure. So one of the things that really surprised me early on, it kind of relates to the whole doctors being a little less sophisticated than you might expect up front. And when we were presenting, you know, basically planning out our reports, we started looking at some reports from other institutions, like just standard diagnostic reports. And a lot of other institutions were handing out basically lists of mutations. So, you know, here's your KRAS status. Here's your EGFR status. And, you know, what we found out was that a lot of physicians had maybe one or two genes that they looked for
But, you know, they didn't memorize even the 20 drugs that were currently available under FDA approved, you know, recommendations for specific sets of mutations. So, you know, just not assuming anything in terms of what people understand, what, you know, what's easy for people. And really just putting all the information in front of people in a really intuitive way, I think, is something that we honed in on very quickly. Yeah, I was going to say, you know, pay for performance and, you know, personalized medicine. And hopefully that's the direction we are heading. So, another question I had is coming from earlier presentations. Today, so there are, you know, life and death conditions, but there are kind of a lot of fuzzy situations where, you know, hypertension may be or may not be there. And so there was a company presenting a bracelet which monitors patients in the hospital
And then basically it has to present this data in some way to the doctor. And there is a fuzzy area where the graph can go one way or another. So, how do you deal with uncertainty of this kind? All right. And kind of, I mean, you need to prognosticate a bit. You need to kind of, you know, kind of extrapolate maybe and see where it's going to go. So, kind of talking about, you know, not clear-cut decisions and kind of using data to look at them and to make decisions with uncertainty. I have sort of two thoughts on that. One is when I look at, I had my whole genome sequence last year
I mean, a closing thought, I want to mention something about that. And, but, you know, when you have a hard number, sometimes you think that it's very fixed. It doesn't, it just seems like that's an absolute, 23% chance of, of X or Y. Well, really that's very hard to say for any number of reasons. So, that's one sort of point, I think, maybe to sort of explain that more and not just to make, to make the number not an absolute number, but a range kind of thing. The other thing which I think is from the diabetes community, which I think is very interesting, and you look at a single number, hopefully, for blood sugar, right? It doesn't tell you a lot, but if you, if what you really want to know is the directionality of it. So, you have a number plus, like, a directional arrow that's going up or down or double arrow going up or down. So, it's things that sort of indicate where that number is in sort of a pathway kind of a scenario
So, it gives context. So, surrounding, I guess the answer is surrounding the, any single number with context, so that they have a sense of what it means and where it's going. Yeah, I think context is more, I think each situation is pretty different. So, glucose has to be taken one way and, you know, another situation could be quite different, like a genetic test. And I think the thing that comes to mind with genetic testing, or even glucose for that matter, is that you had a belief about some sort of event before, some, maybe a prior. So, one number that comes into it, you know, in isolation or alone is not that helpful. There's very few, like, one numbers that, like, really, like, just drive us to completely throw out everything we knew before. And I think that also just gets to context
And then I also think there's, like, a sort of probability and severity thing that if it's, you know, that you, in healthcare, you often have time and you have history. And, and so we shouldn't read anything into two, into one data point when, you know, most people have, most situations in healthcare, you have time and history. And so, I would just throw those two things out there. Maybe I'll follow up with you, because mental healthcare is especially a lot of great areas, right? So, the, the, even, you know, the diagnostic manual is changing over the years. So, how, how do you address these conditions? How do you tackle all these uncertainties? Oh, in mental healthcare, you mean? Well, mental healthcare, the uncertainty, you know, I started in infectious, infectious disease, which actually within the realm of internal medicine has quite a bit of uncertainty, but you actually, you know, you essentially have a state of there's an infection or not of an infection boils, everything boils down to that's not quite that clean. In mental healthcare, it's actually much more ambiguous because it's a continuum. Most of these conditions around, you know, your emotional dysregulation or your mental or personality is a continuum. And so, deciding when it becomes, when it's, it's leading to a, you know, a lot of disability or pain in an individual, as often to that individual, not to you or I
So, it becomes even, the medical model, people in mental health often say, it doesn't quite apply in mental and behavioral health. And I, I think that's true. And one of the things I think about it though, is that mental health issues are so pervasive. We all know people who have them, we may, ourselves may have them, would be incredibly common. And yet, it doesn't, doesn't tend to kill most of us. We walk around and have time and they're quite chronic or subacute. But the amount of relief that can be provided, and we have effective treatments, and yet so many people don't get them. So, it's more of like in, in, you know, utility type speak, you're not necessarily, you're not always, you know, death isn't necessarily the outcome
But the quality of life is dramatically different. And can, you know, and that's, I guess, why we're here is really around quality of life. So, but you can lead a whole life with mental illness the same years, and it's just such a tragedy, because it's, so I think the ambiguity is huge. I think that one of the things is that just from a non-judgmental way, there's ways that you can put people into expectations around how much help they can get, and then serve up those treatments that, you know, would be most likely to benefit that sort of mental phenotype. And we have good ways of doing that, and we should just use them more and make those, there's a huge access problem in mental and behavioral health, huge access at the provider level. So, you know, in 10 years from now, I see no other solutions but technology to really be able to address that. And the good news is, is most of the people who are out there don't need a hospital or a health system or any of the things that are in the traditional medical health system infrastructure. They can benefit with a lot of cognitive or behavioral or emotional kind of help, even in their own home, that you can bring with technology
So I think it's just a huge area for impact in the future. Thank you. Does anybody in the audience have a question for the panel? Maybe I'll ask another question and you guys can think of your own questions. So, one thing which was, again, was mentioned in the talk earlier, caught my attention, was, you know, how much impact was, you know, Angelina Jolie's decision to manage her own breast cancer probability on public context. So there were basically multiple media articles about, you know, her pretty radical or decisions and that caused a spike in genomic testing. And so I'm wondering, you know, it can be a good and a bad thing in the way that how much power do you give the patient to, you know, make decisions? How do you manage it? Because what is the option of the very radical treatments, which, which may be at the edge of the spectrum, right? And I also, and I know that John, John's driver company basically is, is the premises to give the patient the portal to, to give, to collect all the data and let them make decisions. So I think, I think we have folks who, and the field is going this way, but what, what if these decisions involve fairly radical options? And what if the patient it feels compelled to take this options because they're, you know, they're scared and they have, you know, family history of disease and so forth? How, how do you kind of decide, how do you provide this context to, to patients? Do they need to control it? What if they make decisions which are not advisable? Do you follow up from your point of view? So, yeah, I mean, that's, that's a really interesting example. And the short answer is that driver group doesn't go there
So we strictly deal with patients who already have cancer, who have been diagnosed. It's, you know, clear that they have cancer. Typically the patients that come to us are even late stage cancer, because with early stage cancer, typically surgery is the best course of action once you can do surgery. So, you know, that's sort of within the realm of what most oncologists and hospitals can deal with very nicely. When, when you start getting to, as you said, these treatments that can potentially affect, you know, people's health when they make these decisions, what you don't want to have happen are people making decisions, not being informed of the right answers and the right sort of appreciation for, you know, the probabilities and whether or not, you know, like, I actually don't know the answer. Like, I don't have a good opinion about the whole breast cancer issue and BRCA testing. I, I don't feel like I understand those numbers well enough. Yeah
Could I, maybe I can ask a follow-up question for, um, so to, to many oncologists or in the oncology community, a radical strategy might be like not giving treatment. And so I was just wondering how you deal with that, like a patient who actually doesn't want to be treated or wants to stop treatment or wants palliative or hospice care is. Yeah. I mean, that's, that's a really interesting area. I think the way that you would help people make that kind of decision is by really presenting them with all the data. You know, what is, what is the final, the, the overall outcome data? I mean, at driver, you know, we're, we're offering this report that tells people what they could do, right? Um, is nothing on it? Is do nothing on your report? It's not. Um, but I would hope that these conversations are happening between patients and their oncologists. I mean, those kinds of conversations, I'd imagine you want to have with someone who you have a stronger relationship with
Um, typically by the time patients come to us, they've already had some of these conversations with their, with their doctor. Sometimes they haven't, and that's awkward when, you know, they ask us what the probabilities are of, of different situations given their, uh, diagnosis. But, um, but generally speaking, it's, it's something that happens between the patients and their doctors. It's a tough area. I mean, so we, uh, we insulated ourselves from that kind of question in a slightly different way, in that, uh, we're providing, we're, we're not interacting directly with patients or physicians, but with, uh, the services companies that are in turn interacting with, with this. And so, uh, we are providing, um, the best available, um, interpretation, uh, using, you know, a combination of, of, uh, observations in the literature and, um, other inferences that, uh, are reasonable to draw. And then we're working within a framework of, uh, of existing guidelines. And so, there are guidelines, uh, by the, uh, ACMG, the American College of Medical Geneticists, um, and, uh, and NCCN for, uh, how to interpret risk, uh, given what data are available
And so, um, it's helpful to fall back on that kind of a standard, but it doesn't really answer the question of, you know, we, we always wish we had more certainty. It falls in the patient's realm. So, we still have to provide as much context as possible. I think we have a question from the other. Let's introduce yourself and ask a question. Oh, thanks. Uh, my name's, uh, Matt. Uh, I've been helping out, uh, Alexei, um, sort of put all these wonderful talks together
Um, I wondered, so we've heard today a lot about, um, how, kind of medicine is like pretty broken and maybe not so correlated with, uh, people's, uh, best outcomes. And we've heard about these kind of big, exciting, uh, kind of, uh, solutions in, um, diabetes and heart disease and these kind of big, big problems. And so, I was wondering, since we've got a bunch of people that might be able to really fix a lot of these things, what would you recommend is the easiest or the most important problem? So, specifically, if I might define sort of four or maybe five kind of terms. So, health economists talk about kind of quality adjusted life years, sort of how many years you can extend someone's life, how good those years are. So, if it's the number of people that it affects times the, if you implement it, maybe it's a UX thing, the proportion of those people that make a different decision times how profound that kind of change is. Um, and then also, um, multiplied by how easy the problem is to solve for say one data scientist or a small team. And then finally, uh, how neglected that problem is. So, how likely it is that nobody else is going to just fix this anyway, if that makes any sense
So, what's the, what, what would you recommend that people work on? So, that's, that's a really great question. And, um, so I'll, I'll take a stab at it first, which is not going to be a, you know, an optimal answer for you. But if I had to advise you or others to focus in on a certain area, I would, um, start by asking or encouraging you to look at children. Um, because I think, and in particular for children, there's explosions of, you know, diabetes, for sure mental health, for sure developmental, behavioral. If you talk about developmental, behavioral, this is autism, autism spectrum, um, these sorts of things. Um, there's something going on in, I think, childhood or in children in our country. And I think like all countries, um, I'm not sure now is any different than before, but I just think that it's a very vulnerable population that improvements can span the whole lifespan. And it's an area that's relatively neglected in terms of, um, current research and providers
Um, and so I would look at some of the neurodevelopmental, some of the children. I think there's like a great opportunity for technology and data and gaming and all that stuff in those areas. Actually, I would just concur. And I, I, I've heard this vision of every child being sequenced at birth, or at least roughly speaking, it's a first world thing, I suppose. But that you really, you start off and knowing a little, you know, the, the, the physician knows about this patient from the get-go, um, you know, what medications might be, uh, good or bad for this particular individual. And starting from, as I, I think from the get-go is really, it makes a lot of sense in making those interventions, because it seems like a lot of the medical interventions that currently happen are after the fact type of situations, right? We've already, we broke them, they've broken themselves, and then we have to go and fix them. That seems like a much more costly and less optimal problem to solve than trying to prevent some of those things from the get-go, and at least being aware of possible ways to prevent problems from occurring down the road. I mean, I think that there's, um, you know, some, some basic ethical issues around, um, you know, universal sequencing, um, with, uh, you know, there are potentially, uh, great benefits, but there are also all kinds of, of risks
Um, and, you know, we, we face this sort of thing, not, not our company, but in general, we face this sort of thing, uh, now when genetic tests, our tests are ordered and, uh, you know, risk factors, uh, come back that are unrelated to the disease that the person is presenting with, uh, is it, um, is it a good idea to, to provide that information to the patient? And, there's, you know, there's guidelines around that sort of thing as a specific example, but, um, can imagine, you know, many others as, as well. So, uh, it's something to be, it, it's an area of, uh, a lot of potential, but, um, I think requires some care. Yeah, I was just going to say, you know, there was, uh, a note on how, you know, people were offended by the CRISPR, you know, meetings, you know, so privacy is a concern for sure, right? I mean, uh, but I think it's a great time to be alive. I mean, Nanopore was used in the Zika virus sequencing, so to be interdisciplinary, this is such a great time. So, you know, if you're a physician, you know, learn about technology, be more, you know, pathways aware and, you know, open data, right? I mean, so go for that single pane of glass view of, you know, every patient is kind of how I think. I have another question. Can you just introduce yourself and ask a question? Oh, sure. I'm Catherine Ahern
I am, uh, with Clear Story Data, which is not in the health space vertical, but I personally, and I think everyone interested in public health, um, should be interested in misaligned incentives or publication bias or the other kinds of things that affect physician decision making. And I sort of, when, uh, you at Lyra Health introduced yourself, I sort of heard that maybe you're using visualization to counteract some of those, uh, misaligned incentives. Uh, did I hear that? Did I make that up and project that onto you or is that what you were saying? Um, so I think, um, you know, I, I think you, you maybe made that up, but I'll, I'll, I'll go with it for a minute. We should totally do that. Um, so I think, you know, our, our approach and it's one reason, um, we're hoping to find a business model that realigns incentives or else it's sort of not, not, we're not going to have big impact. So we're trying to do innovation on the business model side. That has nothing to do with the, um, uh, what, what you just stated. I think our most, um, interesting learnings on the, um, user, it would be more user experience than data visualization at this point have to do, um, with in mental health
There's this, I'll give you an example. Um, as you go through a user experience, like for, for us, we're, we're essentially eliciting information about how the person's feeling or what their stated needs are and then what they're kind of their unstated, um, needs might be to, to try to help them and identify what they could benefit from the most. And when we do that, it, it, this is not totally data visualization, but again, it's user experience, how you present the flows matters so much. And one of the things we learned, which we're hoping to, um, publish on, I don't think I'm going to ruin that by saying anything here, but like is in the people in the mental health space always talk about priming, but nobody's ever measured the effect of prime. I don't know if you know what priming is. It's like you show somebody sadness, depressed down in the dumps and then you do some like, like the studies, the epidemiology of emotion. Yeah. Yeah
Yeah. And, but they always talk about that in their discussion sections, um, in the traditional academic literature, but no one's ever measured the effect. And the Facebook study. And then, yes, exactly. But the effect is enormous. It's, it's, it's, it easily doubles the rate of depression if you use a standardized instrument. Can you explain what this study is for those who didn't, didn't see it? Well, I think this is an example. I'll, I'll do my best and you can tell me if, of how traditional research approaches it very in the context, in the silo of what they usually study and what the experts all do or say
And they often have paper-based things or whatever, but when you move over to technology, you change the paradigm so dramatically that you can take an instrument that diagnoses depression, like the PHQ-9, which is endorsed by all of the, all of the quality agencies. You know, every health plan is being measured by quality on whether they've taken this PHQ-9. And they're reporting it. Exactly. But if you take and you put that on an internet, internet-based environment or a mobile app, and you ask a patient to do that in the privacy of their home at nighttime, the, um, the rates of diagnosis of depression go up based on the old yardstick for that instrument, which was calibrated in a primary care physician's office with people all around you. And it's totally different. So you get two things. One is the effect size is totally different
The old measurement sticks don't actually measure. Um, and I don't know if the health plans know this now because as they're moving to these non-paper-based ways and trying to get people to do all those surveys before they come into the doctor, it actually just makes people look sicker because when you take it at home. Anyway, but the second thing, the priming effect is just in your emotional or your cognitive, if you show people a bunch of negative words and then you ask them how they feel, they say they feel bad. And like, you can say, oh no, no, no, I'm not suggestible, but it's enormous how suggestible people are. So we deal with that all the time because we don't want to over-diagnose. We don't want to under-diagnose. But if we ask questions a certain way, we change like the results. And so I think this just gets to, um, and we can do it and we can test that really fast and rapid in, on an internet-based environment where you can do tons of testing, right? Um, so those are some of the things we're grappling with at the, I think more traditional literature and research and health system hasn't adequately, like, reported or talked about and how we take those learnings over here with technology and then go and fit them back and try to have impact in the context of the existing system is completely an open question in my mind
Also just talk to behavioral economists and advertisers. Thank you. Thanks. I think we have another question. Please introduce yourself. Hey, I'm Patrick, uh, in the healthcare data and analytics space. So one of the topics you all have mentioned is that patients' view of the outcome can be subjective. You know, uh, life expectancy is objective, but total spending or quality of life can be subjective
So how do you identify best practices, whether through a predictive model or through BI, when what you're optimizing can be subjective? Great question. And I don't know the answer to this. So we work on this. So who are you? I don't know the answer. So the UX person, I would say like, so my first question usually is like, well, who you, what are you optimizing for? Who are you optimizing for? What's the use case? And I mean, usually I like to break it down. So it's not a generic question. And it's a good question, but I mean, specifically, what are you trying to optimize? Yeah. So, well, it could, it could vary, but say it's, um, take one, a diabetic patient in, uh, in a care management setting deciding on like what, uh, care plan that patient should receive
Um, so that's what you're trying to predict is, is what's the optimal care plan to, or what's the care plan that gets to the optimal outcome? The problem is when you're building that model or doing that UI, the, or doing that, uh, BI, the optimal outcome can vary depending on what that patient kind of personally desires. Mm-hmm. Yeah. That's what you do, isn't it? Yeah, I get it. Yeah. It's like, okay, so let me tell you how we approach this. Um, so, we try to break down the problem to the degree that, so let me, let me answer it in two ways. First of all, our first goal is to get people engaged and then get access to care
So we, you know, we're optimizing for, did they close the loop on, on care? And then, um, when you recognize, when, when that's your first goal, now there's later goals, later goals, if you can get them into care happen to be, is the care working? Let's have that be the objective measure. Like, is there a change in meaningful outcomes? And then last would be for a purchaser when you want to have a business strategy. Is that, is that improving your customer's, um, business in some way, either for us through productivity or spend more on behavioral health. So total medical spend goes down. That's downstream. I first need to do a chain of value before that and just get them into care. So I want to connect to care. So that's our first optimization problem
And then after that, we can talk about, did, did the outcomes that matter change? And then the, those outcomes that matter change lead to the end, um, productivity and all that stuff. Now, how do we get them to care? So this is, like deeply understanding the decision-making at the consumer level, the patient or client level, we call them. And I have to say it surprised me. And I think this is where some of the breakdown and the conversations around data and then maximizing some objective, like you just said happened because, um, you can give all that information to them and you can put it in a great data visualizations and they're still going to make somewhat of an emotional decision. And it has to do with like things like, so one thing that we've found that tends to be emotional decision is ease, ease of the solution you offer them. So it doesn't make any sense because you think, no, this person is like, I'm telling you, this person is higher quality and they're more relevant to the specific need you have. You know, you can optimize for that. And then what you find is that provider, that solution is too far away
It's not easy to go to and they don't like overcome that first hump, which is just get, them access to, to care, right? Get them connected to care. So what we've had to do is go back and be like, okay, it seems like this consumer in this situation for that objective measure really cares about like ease of onboarding to the mental health system. So it's got to be nearby or, or, or it's got to be like low burden on them, that solution. And, um, it has to be right away. They can't wait four weeks. They're just not going to be still motivated maybe at that time. So it's like the timing, the ease of access and the cost, if it's free versus a $900, like it's this, those things first matter to close that first gap. And they matter more than like what the objective data says about how great that provider is at, you know, prescribing, um, depression treatments or what kind of evidence based therapies that therapist is using
And so when I first started this, this company, I really wanted to say we were only about quality and we were going to get the best outcomes because, you know, you can follow tons. There's, there are actually a lot of research studies in mental health that say what's most likely to lead to the best outcomes. But if they don't even take the first step and connect or close to care based on, you know, whether the care was cheap, easy to access and convenient for them, you don't have anything. So I think you have to really under, so I think, I hope that was helpful, but. Yeah, it sounds like engagement is essentially what you were optimizing. I mean, I think it's, engagement and hearings at first. And when you can crack that, you can go further and say, was that then helpful in terms of objective measures of effectiveness? I mean, you can, you can do, do the long way, but if you can't, if you have this like sort of chain, this logic or causal chain, like first I got to move them to close the loop on care, right? Then I got to do, and then, so let's just tackle that first problem. And when you start to dig into that first problem, you find that it's not about communicating
We've got, we've got lots of, lots of, um, information and data and user testing now about communicating quality information to consumers. And it's really hard. But once you start getting into like, what we initially thought was a soft stuff, like, you know, who's, they're ready now, and they're two miles away, and they're, you know, 20 bucks, your copay, it changes the conversation enormously. So I think it's not always the right answer to go with those things. And so basically, what you then end up is this sort of like, you know, multi-dimensional optimization across these space of different things that you're trying. And it'd be great to give them like, here's some slider bars, can you say what's important to you? But it just doesn't like, you know, the ideas that we have that, that come from the data science team, we'd have a totally different project than that comes from our user experience team, which I'm sure is not as bright. But I think, um, it's a complex problem, and, and it answers somewhere in the middle, I hope. Maybe if I just add, I mean, I think it's breaking the problem, I mean, the first step, I think, is true, is the person, they're not robots, and they're not like decision-making logic machines, they're going to do these things based on whether they want to do it or not, basically
And if they don't show up, then all the great logic and visualizations in the world are not going to change that equation. So I still say, from my perspective, understanding the, the users, the, you know, where they are, understanding what they're, what's important to them, the A or B or C that's important to them, is part of that to get the whole ball rolling. It also seems like there's probably an element, um, and I don't know how, how well this can be taken into account, of understanding, what you, what you really want to be optimizing, I, I, I think, is the person's happiness with the outcome once the process is complete. Uh, and people are really bad at predicting that for themselves, let alone others. Uh, so, I mean, you would like to have some data from people who have achieved, who have gotten to those end points or gotten, gotten to those outcomes and, you know, how, how happy are they. And if, I mean, if you could optimize against those kinds of, those kinds of things and tell people, we think this is, um, a really, this, this is a good, um, way to go because people who went that way were, uh, the most satisfied and, and here's by how much. I mean, that, that would be really nice. It, it, it doesn't, it doesn't directly address the, issue, the sort of pathway issues of, uh, how do you get over these sort of trans, uh, essentially transient issues of, uh, of, of treatment and treatment course and, and access
Uh, but it does provide people, it, it might provide people with some objective, um, something that they can, they can use to, to make those decisions. So, I, I, I don't know if that's, um, part of, in any sense available or something that people are thinking about. Yeah, patient reported outcomes is, I think, what you're speaking to, which is, people are talking about a lot. So, that's helpful. Thank you. Any other questions? So, maybe then we'll move to closing statements and, you know, based on what we've discussed, uh, you guys can basically, uh, I would say, uh, the audience is very interested in, in helping. We do not have a lot of open source folks who want to be engaged in this field, right? So, so, uh, a lot of technologists feel that, uh, the field is, is transforming and multiple startups are being, uh, founded by a lot of technical people. But, you know, like it's kind of, you know, cavalier attack on the tree of science, right? We, we don't really know what's on the tree
It's a lot of biomedical data, which is very difficult and, uh, you really need to go to school to, to understand or partner with people who work there. So, what would be kind of your insight for those folks who want to, uh, disrupt this field and what's your experience so far? What, what's kind of the, the best thing you can do to, to change the state of healthcare of data, you know, and your advice to kind of, you know, budding entrepreneurs in the space? And that's for everybody in random order. Speaker 1 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 00 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 001 00