The Future is Now: Leveraging Mobile Tech to Accelerate Scientific Discovery
(audience applauding) - Thanks so much. That's, like, the best introduction ever and it's really great to see all of you guys here today. Reiterating what Cathy said, it's really... I think we've been thinking and excited about getting this group together and all the potential that can come out of it, and I'm really excited for that and hopefully the conversations that happen throughout the day and hopefully in the coming months and years that I think can be really get us to having greater impact and real progress.
So I'll talk a little... A few different topics today. Start off with my area, what Ravi alluded to, the brain behavior, mental health and why I think the potential for mobile technologies, particularly needed and great there. And maybe convince some of you folks to use your expertise and ideas and apply it to our problems. I'll talk about the Intern Study that Cathy mentioned and another newer study, PROMPT, also using mobile technology for a couple different reasons, trying to hopefully relate some of the findings and results that have come out of this study that could be of interest.
But also because we see both of these studies as University of Michigan community resources and if there's data that you think could be used in a different way or you have projects that could use it, we would love that. And particularly for this, if there's mobile technology functionality that has been built, one of the goals of MeTRIC is really to help, find ways to transfer that and to share that. And we really want to do that with these projects in general. And then I'll come back to some of the lessons and pain points that we came across with those studies and how that might inform the potential of this group and what we can do. So starting at a 30,000 foot level, this is from the CDC, over the last 15 years, the leading causes of death in the US.
And you can see that for most of the major causes were as a field of medicine making progress. You're at any given age less likely to die of cancer or HIV or heart disease than you were 15 years ago. The two glaring exceptions are suicide and overdose, behavioral disorders with mental health as a driving factor. It's complicated and multifactorial why we haven't made progress and maybe even going backwards in this area.
But I think one of the important drivers is that from a clinical research perspective, our phenotypic data is really bad. I would've loved to have something like the 1902 version of the EKG in mental health or behavior. Most of the data we work with traditionally is subjective, what people tell us and tell us only when they come in the clinic episodically and that speak particularly in emotional things that we're talking about subject to a lot of recall bias and inconsistency and that, I think the granularity and quality of that data has been a real barrier to progress. From a even more clinical perspective, we are the first line treatments for the common disorders like depression, anxiety, are the same as they were 30 years ago in terms of medications and therapies. They work modestly to moderately well.
But we have very little insight into what's gonna work for what person. And so even when people do get into care, they often cycle through two, three, four, five treatments before they get to the one that works and sometimes don't even get to one that works. In parallel of that, we also have nowhere near the number of clinicians to treat everyone who needs it. And this has gotten exponentially worse during the pandemic.
So most people now who need care can't get it. There's a several month wait list to get in to see a therapist or psychiatrist here at Michigan. And the same all across the country and even worse in other parts of the world. So these are again, major barriers that even if we did have better treatments, it'd be hard to get it to patients who need it. And I think in both these areas that mobile technology can really help in terms of the phenotype data, this is really the first time that there's a potential of capturing objective data in real time in domains that are really relevant to how we think and behave and mental health.
And we can do it not just when people are coming into the office in an artificial environment, but we can do it while they're living their lives and see how things change going through it. Also because mobile technology and particularly smartphones are becoming ubiquitous. I think about 80% of Americans have it now and we interact with our phones like 625 times a day. So it's a major part of our lives and provides a way to get people care outside of the traditional clinic-based system. And there's progress in both these domains over the last few years. This is a review looking at all the different mobile technology phenotypes that have been associated with mood and depression.
And you could see the domains that have come up and most of these findings are robust and multiple studies have identified them. It's limited what we can do or where we can go 'cause most of these studies have been relatively small, a couple dozen subjects and cross-sectional. So we need to advance that. But there's promise there. And similarly for treatment, there's been huge, at least investment and certainly during the pandemic, I think about $10 billion has been invested from the private sector into these remote mental health care companies. I've listed some of them here.
Some of these have been studied and certainly academic versions of mobile mental health treatments in CBT and mindfulness. And generally they find that they're more effective than weightless controls. This a meta-analysis here and particularly when there's a human coach involved almost or more effective than in-person therapy. But there still needs to be a lot of work in identifying which ones are most effective and particularly for a precision perspective, what's gonna work best for what person at what time. And so hopefully as a field we can make those advances in the coming years. And that's part of the motivation for the studies that we've been doing.
So this is the intern health study that Kathy mentioned. And so we started this before, initially not as a mobile technology study, we're, I guess, 17 years ago now, we're enrolling our 17th cohort. So the premise of the study is based on the stress of becoming, being a first year physician. We enroll folks in the spring, in March when they match into a program and don't quite know what they're in for.
They're in a low stress period and happy. And then we can follow them as many folks in the rooms as attendings start to stress them out by having them work 80 hours and sleep inconsistently and for the first time really deal with life and death situation. And we follow them with surveys and get genomic information and mobile technology.
The core of the study is this that before the year starts in that in the spring about tree, four percent of them are depressed, about the same as the general population. But by September, about a quarter of a meet criteria for depression. So a huge increase in in depressive symptoms and anxiety, suicidality and depression's an episodic disorder and people cycle in and out.
About half of the interns have at least one episode of depression during the year. So in the context of mobile technology, it provides a nice platform to study people as they encounter stress and become depressed. And we can follow how things change longitudinally with that. So about, I guess 10 years ago now, we started a mobile technology component of the study.
We'd gone through a few iterations but have an app that each of the interns downloads that pulls in information from various sources from their phones and we provide the interns with mostly Fitbits throughout the years to gather sources of information and more and more recently, Apple watches as well. There's a whole set of different types of data we've been getting from mobile technology, traditional daily metrics of activity, sleep, some more granular things in those domains. And more recently, just in the last year, newer types of data through that are becoming available through (indistinct) ears that get in additional domains. And I'll go through some of them. to go through the results of the study. One of the things we get, as I mentioned, is daily mood.
Each day people enter their mood score and this mostly to demonstrate we can stumble upon things that we never intended to through mobile technology. So Elena Frank, who's back there led the study with Joan Zhao and Brahmaji we found that that looking through all the daily mood scores, there was one day with a lower mood than we've ever seen at any other point in the study. And when we looked closer it was Trump's election and then the next lowest day was the day he was inaugurated. And then as we looked more and more in the data, every day during his... The first couple years of his presidency when something happened that was towards the conservative or republican direction, a supreme court ruling, the mood of the interns would go down. Every time something happened in a more progressive direction, their mood went up.
And this certainly wasn't the main goal of why we were collecting daily mood, but we were able to capture this because we were tracking them through mobile technology. And yeah, just thinking about this, we'll have to see what the mood was like on, I guess November 6th of next year to see the state of the country. But more closer to the core reasons we started the mobile technology arm of the study, we've also found, and a lot of this work is led by Yu Fang who will be on a panel later, a number of mobile factors that have been, that we've found associated with mood and depression and predict changes in those domains.
The first ones and early on in the study, just with a small pilot cohort of about 30 subjects, we found a relationship as you'd expect between total sleep time, how much people are sleeping each day and mood and similarly between activity and mood. But with mobile technology, we're able to get at the direction of the relationship and causality in a way that really hadn't been possible much before. We were able to see that your mood today has somewhat of an effect of how well you sleep and how much you sleep tonight.
But the relationship is much stronger in the other direction and how much you sleep tonight really does have a strong effect on your mood tomorrow. As we went on and gathered more and more subjects, we got into sleep more and this is work with Kathy where we found that total sleep duration was important, but actually as, or even more important is variability on the sleep parameters. How much your sleep varied from night to night.
If you were sleeping seven hours a night, that was much better for subsequent mood and risk for depression than if you were sleeping five hours some nights and nine other nights. And similarly, if the timing of your sleep and when you went to bed and the midpoint of your sleep varied, that also had a great effect and again, in as much or greater than the duration. And again, that's a kind of assessment that would would've been difficult in this field without mobile technology where most of our assessments of sleep in terms of depression and anxiety had been based on self-report. Danny Forger and his group have done a lot of work with our data and have been able to extract from the Fitbit and Apple watch data a mobile marker of circadian phase.
And we see an association between that and your circadian rhythms with mood and mismatches between your circadian rhythms and your sleep patterns that relate to mood and similar more patterns between how long you've been up and your level of fatigue and mood. And I hope there's a lot more we can do now that we have have these circadian markers in this data set and hopefully others. This is a graph on exercise and we know for a long time that there's a correlation between how much people exercise and their depression. People who exercise more, have lower levels of depression, but demonstrating causality there has been difficult because people generally don't adhere well to exercise recommendations.
And there hasn't been a real way of measuring exercise in a large scale without mobile technology. So this is with Yu and Amy Bohnert we emulated a clinical trial as if people were assigned to different exercise levels, exercise recommendations by the the CDC guidelines, low, medium, and high levels each week. And we see a relationship there. The people who are exercising in the high exercise group have lower depression than people in the other exercise groups. With this, we've also been using this in a few other ways of trying to combine the mobile technology with other data types.
The omics that Ravi talked about. In this case, trying to understand a little bit more about genetics and how genetic risk manifests into increased risk of depression and occurrence of depression. And again, trying to get at the personalized or precision angle. In this case, we looked at a polygenic risk score, as I mentioned, all the interns have GWAS data on them and they're about 300 loci that have been associated with depression now. None of those loci have a very big effect.
But overall you can sum across all of those and all the other loci to come up with what's called a polygenic risk score for depression. You can do the same for almost all complex traits and diseases. So when we applied that to this study of exercise and depression, we see that the effects of exercise are different by your genetic background. So people with a low polygenic risk score for depression are much more, or about twice as sensitive to exercise in terms of depression than people with a high polygenic risk score.
So starting to... This is just one study, but the potential of combining genomics with mobile technology to try to get at what people's sensitivities are to different inter interventions and stressors. And this is just to demonstrate most of what I've talked about or with the traditional mobile technology measures we've been getting, just this current cohort, we've been getting a few more domains through SensorKit and soon more from the EARS platform around what we're showing here is ambient light exposure to your phone and watch and how that changes throughout the year and throughout the day. On the right is location data and the different colors or different locations. So you can imagine different sorts of information that we could get.
You know, light is incredibly important to our sleep and mental health, how we go about our days and different locations is clearly important. We're still working and would love input on how to extract signals and important information from this and phone usage and voice data that we'll be getting in. And we haven't really done that yet or associated this with mood, but it's, I think an exciting domain that we can get into more areas that are clearly gonna be relevant to mental health and behavior. So in addition to the digital phenotyping, we've also in the Intern Study used mobile technology to provide mobile care and assess the effects of different interventions.
This is a traditional cognitive behavioral therapy, which is one of the first line therapies for depression. This is through a app called Mood Gym that was administered to about four or five hours of training on this CBT app before internship starts. And you could see that what I'm showing here is suicidality that the interns who had the CBT training in a preventative way had lower incidents of suicidality and the same goes for depression during the year than people who didn't.
And with a relatively large effect size for a relatively minor intervention of a few hours of training. Since then, we've been focusing even more micro or smaller interventions and ways of informing the interns of their own data and other information. Each of the interns get, through the app, a dashboard where they can see their data in displayed in different ways and some information about themselves. Each day they also have a a 50% chance of getting a message, a short text message about the mobile technology data that's been collected on them or a tip derived from therapy.
And through this, we've been doing these micro randomized trials to see the effects of these sorts of messages. This is work that Zhenke Wu led with his grad student and we find that these messages can work and can affect health behaviors and mood, but it depends on the state that the people are in. This is showing for sleep but the same holds for other domains. When people get a message about their sleep, when they're not sleeping enough, when they're sleeping five, six hours a night, the message helps. They sleep a few minutes more, 10, 15 minutes more the next night after they get a message compared to not getting a message.
But if they're sleeping well, if they're sleeping seven, eight hours a night, the messages actually do harm, they sleep worse or they sleep less than they would if they didn't get a message. So highlights the importance of capturing the state that the people are in and the differing effects of this small intervention depending on that state. More recently we've also been experimenting with other ways of getting the information from the mobile technology back to subjects. And this is a competition setup that we did trying to gamify this information.
Each week we pit groups against each other. So you know, this week might be Michigan Internal Medicine is going up against Penn State Surgery on steps and see who can get more steps this week. And so we see that these sort of competitions have have an effect. You know, people when we start off, when you're in a competition, a step competition you'd get about 200 more steps a day than if you're not in a competition.
And a smaller, but still in effect on, on sleep. But the effects of these wear off over time and after a month or two they no longer have much of an increase in steps or sleep with this. So I think there's potential here but work to do in figuring out how to optimize these and how we can nudge people to different behaviors with these sorts of interventions. So I hope that gave you a sense and a flavor of mobile technology work that we've been doing in the Intern Study. And again, would love to hear ideas about what else we can do if some of this data could be used in a different way for thoughts you have or ideas for future cohorts. I'd love to talk and our team would love to talk.
The other study I wanted to talk about is PROMPT, which is co-led by Amy Bohnert came out of Precision Health and is a compliment, I think, to the Intern Study. The interns, we enroll them and follow them from when they're healthy and they then undergo stress and many of them get depressed and anxious. I mentioned earlier on that the wait list for getting into mental health care is ridiculously long. and this is taking advantage of that.
We enroll people who are on the wait list to get in at ambulatory psychiatry and University Health Services when they're at probably their sickest, and then hopefully follow them as they get better. We enroll them on average about six weeks before their appointment. And so have a period before they get their traditional clinic-based care and randomize them to digital treatments and throughout their time in the study we follow them with mobile technology, with a Fitbit and Apple Watch. The interventions that we randomize them to during that pre-clinic period are Headspace, along with Calm, the major meditation apps that are out there and Silver Cloud, which is a another cognitive behavioral therapy app.
And so they also get those sorts of notifications I talked about that we looked at the effects of giving information about the mobile technology back to subjects and also have a slightly different dashboard through my data helps, but getting again, this similar information available to subjects. So we see a pretty good effect of the mobile interventions of Headspace and Silver Cloud. This is their effects from when they're enrolled in the study to the line where they start their clinical care at six weeks and about a 20% decrease in depressive and anxiety symptoms, which is pretty good for this sort of treatment refractory group of patients who've been waiting for care for often months at this time and lesser, but also some effects on substance use and suicidality.
And then their rate of improvement slows down once they get into actual clinic care, but hopefully continues to improve. We also see an effect of the notifications and more of a robust effect than we see of similar notifications in the intern population. People who are getting messages on steps in this population, it seems to be helping and they're taking more steps a day than those who aren't. and maybe because of the increased activity it's affecting their... Their resting heart rate actually is lower after this period.
We're also, again, starting to get some traction towards the precision or personalized effects. As I mentioned, the two interventions that we have are Headspace for mindfulness and Silver Cloud for CBT and we see that people who are coming in with a lower step count and less activity seem to respond better to Headspace in both for both depression and anxiety. Whereas folks who are getting above the 50th percentile for steps have similar, might be a little bit better for Silver Cloud, but we're seeing a differential response based on some of the mobile technology data.
And it's early days, but as we get more and more data and PROMPT and for more domains, hopefully can make even more robust predictions. So I flew through that but hopefully gave you a flavor of what the data is and hopefully the potential. I think the way I presented it was mostly just the results and hopefully some interesting findings, but there were a lot of dead ends that didn't go anywhere and frustratingly slow processes to get there. So piggybacking on what Kathy talked about, that's a lot of the motivation and why I'm excited about this group and hopefully Brahmaji and Sachin can talk more during their talks about their motivations.
But I think this is our third or fourth app company that we worked with to get this information, spent a lot of time struggling with procurement and information assurance and the IRB and trying to get things through and building the data pipelines have been... A lot of people in the room have been involved in that, but a struggle and we're still working on it and especially as new data streams come in. Even the two parallel studies that are getting similar types of data in Intern and PROMPT transferring between them is not simple and more work than it should be. and in talking to a lot of you folks in the room and others that a lot of us were going through the same struggles and we were all kind of reinventing the wheel over and over again and the activation energy to do a mobile technology study has been really high. So I think if we can successfully transfer knowledge and infrastructure in a way that can really lower that activation energy and share best practices and hopefully some resources, I think we can really lower the bar for new people getting in the field. And I think there's incredible potential if we can bring some of the functionality that Brahmaji has put into his studies, into our studies and vice versa, that the potential for impact is exponential and I think make this open for people who are not mobile tech or technology experts.
And if we can add on mobile technology to other studies that aren't primarily mobile technology studies, we can really incorporate with genomics but proteomics and others to really understand the biology as well. And I think the other motivation that I wanted to mention is that it was, I think, serendipity and luck that we were trying to figure out what interventions to do and stumbled upon some of the best people in the world at doing just in time adaptive interventions stumbled into people doing incredible work on circadian phase. And there's probably 10 other areas that we didn't happen to stumble onto that could've taken the study in other directions.
It would be great if through this group we can make those interactions much more intentional and much more frequent. There's incredible experts here at Michigan. This is just a partial list in these domains and adjacent domains to mobile technology. And I think the more we can know what other groups are doing and then make it easy to collaborate and to borrow, I think there's...
It's limit limitless what we can do. And Kathy and Victoria and the mobile technology corps, I think, are starting to do this and Kathy mentioned the affinity groups. There's also a knowledge base that we're building up, repositories, but I think this is just the beginning and we really would love input from you guys on what's needed and how metric we can really provide the sorts of resources and infrastructure that can take your studies to have greater impact and help build collaboration. So I'll stop there, but I'd love to have discussion now and then throughout the day. So thank you all for listening.
(audience applauding) - (indistinct) If you can repeat the question. - I will repeat the question. Okay, sure. Any questions? - [Victoria] What was your biggest surprise or unexpected event that that came up with IHS? Whether it was planning while you were campaigning for funding or actually during the conduct? - Yeah, I think probably a bunch and others should weigh in. Should I repeat the question? Victoria's still with her microphone or? Yes, okay. So what are the the biggest surprises and unexpected things? I think initially and even getting funding for the mobile technology stuff was hard and we lucked into it as selling to NIH, these things as mostly genetic studies and they were happy to fund the genetics and then we with some philanthropy and then just a connection we're able to get the mobile technology as an add-on, but it's become a really important and maybe one of the more important arms of the study.
So I think hopefully that's changing, but that maybe was a surprise. Yeah, I'm trying to think. Nothing else comes to mind, but I think the... I alluded to this a couple times during the talk, but most of the, I think interesting things were not original hypotheses stumbling upon, like variation in sleep or the effects of Donald Trump, were not like expected, but I think capturing broad data, we really are...
There's a lot about mental health that we don't know and probably many of the important things are not things that we're gonna hypothesize. So being able to capture domains and assess them in an unbiased way I think has a lot of potential. Thanks for the good question. - [Moderator] If you have a question, just raise your hand and we'll come to you with a mic.
- [Audience Member 1] Yeah, thanks for your presentation. I really enjoyed it. And my question is, I think one of the biggest challenge in these kinds of studies is to maintain participants involvement to keep according their mood or use of wear sensors.
And I just, I'm curious if you could share any useful tips like that you... It was helpful to maintain their involvement in addition to the gift card. - Yeah, so the question is on engagement and maintaining people's engagement and participation in the study, especially for things that go on for a year and you know, again, I would love ideas and discussions. It's one of the biggest challenges of these sorts of studies. We do in comparing the two studies have better engagement, people filling out their mood score daily in the clinical study, the PROMPT then the Intern Health Study and that might be related to the populations and how busy interns are.
In both studies we do see attrition and attrition as you'd expect, particularly among the mood ratings and people filling them out go down across the year and early on we see attrition in people not wearing their wearables after a month or two and then it stays steady. I think some of the sorts of things... The dashboard, making the dashboard engaging, these competitions might have as big or bigger effects on engagement than actually the behaviors. But I think it's still... We still have a lot to learn about engaging and I think trying to experiment with different things is gonna be a really important part of the field going forward.
So I don't have great answers for you, but I think it's a really good important area going forward. - [Audience Member 2] So in like future studies, in terms targeting more groups that are under studied, such as marginalized populations, people that don't really represent the larger population, (indistinct) - Yeah. So the question is about expanding and new studies to study underrepresented populations. And yeah, I think that's really with PROMPT we're doing that and trying to expand to clinics with more under studied populations, racially, ethnically and in other ways, the demographics of physicians is horrible and not representative.
And so that's a challenge in the Intern Study. We're partnering with some historically Black colleges and universities to expand that part of the population, but the real solution is to change the pipeline of physicians. But I think aside from these two specific studies, I think engaging other populations aside from a lot of studies are done on like college students and expanding to mobile technology, I think the potential, particularly in the treatment arm of things is even greater.
Getting care to populations that can't get care right now and making sure that we're studying and doing things in a way that will really help them is gonna be important. So we need to do that and ideas on how to do that effectively are really welcome. Thanks. Great.
Well thanks. Hopefully we can have... Was there a question over there? Oh, okay. Sorry I didn't. Oh, okay.
- [Audience Member 3] So (indistinct) first I had a question about (indistinct) your subjects. If you wanna see me at the break, just come and see me and we can talk about ways of how to keep your subjects engaged. - Sure. Or if you want to go through some of the main ones now that well, yeah, - [Audience Member 3] Well a lot of it's gonna be project specific.
So there could be ways to build it into your project. So for example, maybe there could be ways to send daily reminders just, "Hey, checking in", so if it's part of your app, so just, "Hey, have a great day." It doesn't have to have any meaningful anything. Just, "Have a great day today, do well today. Call us if you need us, we're available."
If it's not a mobile app, sending a monthly email if is potentially like say a drug study or a device study, quarterly newsletters. It doesn't have to be anything extravagant, just "Hey, this is what we've been up to in the last (indistinct)." Anything that just says "We appreciate you, you may be a number overall, but to us we're a person." So there's a lot of different things that can be done. Varied to no cost, time obviously, but no cost.
So there are a lot of things that can be done and not necessarily me, it may not be my board, but I can help you. I'm too representative but I am willing to help if your board person can't do it. (indistinct) But there are a lot of things that can be done just to kind of, like I said, Even if it's just... if you have a low enough overt, maybe you're calling once a month to say, "Hey, how are you doing?" You know, so anything.
So there's a lot of things that can be done. Maybe you have a huge number of subject, you can't be calling everybody, but like I said, (indistinct) send a bulk email "Hey, we're checking in, don't forget we're available, call us and need us." Really, that would go a long way.
- Great ideas and yeah, and I think that, sorry I won't repeat all of it 'cause it, but I think the engaging with the subjects and newsletters and things like that I think can really help. Just building off that a little bit in PROMPT we track how they're using the wearable devices and if we see they're not using it for a certain number of days or aren't entering their mood, the team checks in with them and assesses if there's technical problems and encourages them. and I think that's been really helpful for engagement. In the Intern Study we do have a newsletter and I think particularly letting people know when their data has been helpful in a finding or really advancing the field makes them feel like it's valuable for them to participate and I think has helped with engagement. But yeah, I think there's a lot of great ideas and I think the more we can do, particularly for the longer term studies, the better for engagement. So thank you.
Great, well thanks, thanks so much. (audience applauding)