Hello and welcome. I'm Amy Cohen, a member of the clinical expert team with SMI Advisor, and I'll be moderating today's event entitled Treatment and Risk Prediction, Clinical Informatics for Care of Serious Mental Illness. SMI Advisor, also known as the Clinical Support System for Serious Mental Illness, is an APA and SAMHSA initiative devoted to helping clinicians implement evidence-based care for those living with serious mental illness. Working with experts from across the SMI clinician community, our interdisciplinary effort has been designed to help you get the answers you need to care for your patient. Now, I'd like to introduce you to the faculty for today's webinar, Dr. Alexander Young. Dr. Young is a psychiatrist, health services researcher, and professor of psychiatry at UCLA, and serves as a core member of SMI Advisor's clinical expert team. He is board-certified in psychiatry and clinical informatics. He has focused on improving and studying care for people with mental illness, emphasizing implementation science, psychopharmacology, measurement-based care, and health informatics. Dr. Young has developed quality measurement for mental health and worked to implement state-of-the-art treatments for people with mental illness, including leading clozapine programs at public health clinics. Dr. Young, thank you for leading today's webinar. Thank you very much, Dr. Cohen. It's a pleasure to be here, and I'm glad people can join us today. So, I don't have any disclosures for the talk. So, I'm going to start with some slides. Today, we're going to be talking about clinical informatics and about how particularly they can be used in people with serious mental illness, how they have been used in various kinds of ways to hopefully get people thinking about how clinical informatics might be useful in their practice and their daily lives. So, here's a quote from a number of years ago. This is from an individual in the IT community who was having some interactions with oncologists for care and getting frustrated, and basically made the statement, there's no reason an oncologist should be a human being. And the quote here, what it really means is that he thought that when getting care for oncology, there's a lot of information that's being digested and considered and a lot of data from different sources and complex treatment decisions being made. And he didn't feel like the oncologist was particularly adding a lot to the decision-making, and that it perhaps could be done by the patient with the appropriate data or perhaps supported by computers and information and informatics systems. And I think this is sort of an interesting note for clinicians and raises the question, well, what about, so that's sure oncology, but then what about radiology, pathology, which are a couple of the specialties that have been proposed as where informatics may have a tremendous impact in the future. What about dermatology, ophthalmology, psychiatry, for example? Are these fields that computers could either play a substantial role in, larger role over time in the delivery of services, or maybe could they completely replace the clinician? And this also feeds into the field of concerns about job loss due to automation. So I think most people have probably been hearing about this. There are, of course, job losses over automation that have happened over the past 30, 40 years in our country, mostly around reductions in factory jobs and labor-based jobs, farming, factory, things like that, that are increasingly done by machine as opposed to people. And certainly this has been a very disruptive process nationally. A lot of people have lost jobs that have been reasonably high paying in the past, and that their middle-class livelihood is threatened by this process, which increases productivity, but can be very disruptive. And the concern is now that there's a new wave of this that is happening through automation, through computer automation. And people have examined this, and there have been a number of different estimates, but here's one, basically, that about half of US employment is at risk due to automation, meaning that of the current job pools, that the job can be sufficiently specified to be performed by a computer. And they've also estimated these for medical fields, people working in the medical field. These are some of the numbers from the work. Fortunately, I think most people on the call can rest fairly easily, although there have been some sort of chatbots and interesting efforts to have computers answer questions that people put to them often by just reflecting the question or coming up with something related, that the actual chances of psychologists, psychiatrists, counselors, psychiatric aides, so forth, by being replaced by computers are actually reasonably small. That the jobs that are more at risk tend to be accounting jobs, receptionist jobs, things like that. Though it doesn't mean that computers can't play a larger role. And in a lot of these fields, to a varying degree, I think you'll see that computers will be increasingly taking over some of the decision-making and information supply and management. These are things that people have wanted to actually do. Clinicians are interested in having EMRs and things like that, do that to a greater extent to provide help with complex decision-making, which has been somewhat slower to develop than people had thought. But that doesn't seem like for some of these specialties that it's necessarily a near-term threat to the job, per se. What is the reality of informatics? I gave a talk at UCLA, and this was the email that was sent out before the talk, announcing my talk. And so it's not a very flattering picture of me. So this is an example of what happens in the real world. I think we're all accustomed to the fact that the computers and things just don't work as expected, and that bugs are the rule. And this is, of course, something that we have to always be concerned about. So what about informatics and care of people with serious mental illness? I'm going to talk about two different areas. One of the areas is, of course, clinical informatics, computer systems providing treatment or supporting treatment. There have been many studies now of computerized efforts to provide treatment to a certain degree, and by computer I include things here like apps and web systems and various kinds of paths or interfaces to computers. I'm going to present one, which is an informatic system that we developed and used to provide services for weight for people with serious mental illness and studied in a controlled trial. Again, this is just one instance, one example. One thing you should also observe when you're watching this and the other informatics approaches is pay attention to how much the human is involved in the process. There's emerging literature on the effectiveness of these programs being connected to how much they're fully autonomous as opposed to semi-autonomous or involving people in certain ways in the process, and I think you'll see that in this study. Even though it is an informatics approach, it is not devoid of human contact and involvement in the treatment process, and I think that actually is a part of its effectiveness. I'm also going to talk about risk prediction. This is one of the more cutting-edge areas. People talk about predictive analytics, machine learning, that sort of thing, and I'm going to talk about how this is actually being used to predict risk for suicide and general medical outcomes, and then we should think about how this can, again, be applied to our care of people with serious mental illness. Then I'm going to talk a little about clinical informatics in general, sort of fellowships and certification training opportunities. This is a relatively new field. It is definitely a field now. For example, informatics has a board specialization, American Board of Medical Specialties for Physicians, has a medical informatics certification, and there's a number of other certifications and fellowships that are available, so it's clearly something that people can think about. So, why care about clinical informatics? We're good at providing services and treatments for folks without computers, and in fact, most of us work most of our day at service with patients without direct computer assist, and so why be concerned about this? The second question is, why serious mental illness? I would say for this, people on the call presumably are sensitive to this, but the population with serious mental illness has a very high rate of obesity and smoking and low rates of primary care and prevention. This is a population that dies many years prematurely, mostly due to cardiovascular illness and cancer, sort of things that kill other people, but these folks are not taking advantage of the preventive and treatment services that many other people use and are therefore paying the price. The intersection of these two is, I think, a good example for application of clinical informatics. We have, of course, in-person wait programs for people with serious mental illness. There have been many research studies of these. There are many different types of in-person wait programs that are clearly effective in randomized controlled trials. They've particularly been effective in moderate-sized or small controlled trials when patients volunteer for the study, meaning that they're motivated to receive wait services in general and are paid to actually attend the intervention, which is, of course, a motivating factor and helps to address some of the barriers to attendance and participation. Then these wait programs on the basis of these randomized controlled trials are now found in treatment guidelines and therefore should be used. One thing that folks that have tried to implement these becomes fairly clear is that they require major clinician time and patient travel. If you're running a group for people with weight, for example, maybe have like nine people, ten people in the group on average, something like that, probably three-quarters of your clinic is overweight and half of it may be obese. If you run the numbers, the hours of clinician time to run groups like these on a regular basis is enormous. Then also, there are issues of patient travel to weekly groups. There are many settings or most settings where patients, it's not convenient to come to the clinic on a weekly basis. I think for those reasons, in part, effectiveness and usual practice of wait services has been limited and they're rarely actually used. The rate of delivery of this service, even though it is an evidence-based practice and found in guidelines, is very low. This is eWellness, the program that I'm presenting, which is a weight program interactive web-based system to self-manage weight in conjunction with peer coaching. This is the first interactive web system of peer coaching to help people with mental illness self-manage their weight. What you see a picture here is a picture of a computer interface been designed for people with serious mental illness. The design of the research that I'm going to show you is that we recruited people in usual care from outpatient clinics who are overweight. This wasn't a volunteer-based study. People were invited to participate if they're overweight. Of this group, 81% had internet access on their own. A third of them had mobile internet access or they had internet access at home or other places. This was a randomized control trial. We randomized patients to one of three different treatment arms and they were assigned to the arm. They could receive it if they wish for six months. One was the eWellness, so this computerized system using clinic kiosks or internet access elsewhere to receive a computerized curriculum and interactive personalized program for diet and exercise management, which I will show you. They also received peer coaching of a reasonably low intensity, like a once-a-week contact by phone. The second group received in-person groups of the same curriculum, so the same curriculum as the web-based program delivered, but delivered in person by clinical staff. The third received usual care. They had access to weight services. There are weight services in the community, for example, a variety of things. They could certainly make use of those, though the rates of use were relatively low. We maintained this and they were assessed at three to six months. A little about eWellness, this clinical informatics system, what does it do? It educates users on diet and exercise. It sets personalized goals, helps the person choose from amongst various goals that are appropriate to where they are in the curriculum. It helps them set individualized goals in an automated fashion. It tracks weight and exercise progress and it helps users balance what they eat, how much they're eating, and how physically active they are. We tried to make use of components that are effective in behavioral interventions in general. Things again like goal setting, of course education, but providing homework assignments, the kind of things that effective cognitive behavioral therapies have to have. Then finally, we also had a dashboard for the peer coaches so the peers could keep track of how the patients were using this, were doing, where they were in the curriculum, and help them with problems to provide encouragement and motivation around specific areas of interest over time with regard to diet exercise. People could access these from kiosks or homes. We placed kiosks at clinical locations so patients could, whenever they wanted, sit down and interact with the kiosks. These are audio-assisted touchscreen kiosks. It's similar to what you see at the airport, but it has audio with it so people wear headphones. It allows them a self-paced experience and it's a touchscreen interface so it has no mouse or keyboard. It's designed to be very facile to use, user-friendly, visually engaging, includes videos and that kind of thing. Patients made use of this. They could also do it at home. The same sort of interface can be used with a standard personal computer, and many people did via the internet and using a secure login and so forth, and for their own personalized account. It operates on a home PC. The way these touchscreens work is that they basically would touch the screen to provide the mouse click, so it can be done at home with a regular mouse as well. We provided pedometer weight tracking. We gave people pedometers. It's a little old school these days. Of course, people often use their smart wristbands and so forth, but these pedometers actually work quite well as well. These are the kind of pedometers you can just throw in your pocket or your purse or whatever and take with you. They don't need to be worn. They track movement by accelerometer. We had people tracking their activity on a daily basis as a way of helping them understand how they're doing and provide motivation, you know, sort of like an Apple Watch on a budget. You can see the step count here for an individual on a daily basis. We set goals to people, which was basically about a 10 percent increase over whatever they were doing, trying to encourage them to do a little more on a daily basis. People talk a lot about 10,000 steps a day. 10,000 steps a day is certainly a perfectly reasonable goal, but it's also possible to get activity with lower levels of steps and where people could do more if they're so incentivized. The content was delivered by chapters. These are just an example. Every chapter has a flow to it. Introduction has learning content and then has questions to assess people's learning to see whether the content has made it in. Of course, most people with serious mental illness have cognitive deficits to some degree that can impair learning disabilities in a sense. This is an effective way of allowing people to repeat content, which is something that folks with serious mental illness often do on their own anyway. It works that into the program as they would like that. Then there's also a review of the chapter content with some handouts and that kind of thing. You can see from the pictures that this would try to make a lot of use of graphics and help with learning and that kind of thing. This system was developed specifically for people with serious mental illness. There are computerized web-based curricula for weight, particularly there are now. They were developed around the same time that we developed this. For the general population, those are available. You can pay for them. They're online. I won't put in a plug for specific ones, but just to say that there's a number of them. This one though, we have an understanding that people with serious mental illness, because of cognitive and literacy issues in particular and somewhat less computer experience on average, though still substantial for most people, is a population that has specific needs. There's been a whole research area on understanding the computer interaction needs of people with serious mental illness. We specifically designed the system for this population to make it usable and effective with them. It has the same content, but I think you'll see it's designed to require not to basically be easy to use and ready to learn from without requiring the cognitive mapping and some of the cognitive skills that people may be challenged with. This is an example of a screen. It says, today portions are being served to restaurants that are bigger than they should be. This is called portion distortion and then shows a picture of what I think is a typical hamburger size these days. There's also audio with this and a video. I won't play that here. You can see the screen layout. It provides ways of moving around sequentially through slides, repeating slides, stopping people, adjusting volume, that sort of thing. This is an example of goal setting. People offered three possible goals that are relevant to the topic of this education, which is food size at restaurants. The goals could be eating meat portions no larger than a deck of cards, eating sizes of fruit no larger than a tennis ball, or packing half a restaurant food in a to-go box before starting the meal. These are ideas that people can use to adapt to the fact that these days at most restaurants, when you're offered a meal, you're usually offered a portion size that is an order of magnitude or several times larger than what is actually a healthy portion. Again, that's because we have a capitalistic society. People get what they want. It's very effective providing people what they want, and that's what people want, but it doesn't mean it's good for them. In terms of the printout, this is at the end of each module. They would get a printout which personalized for the specific progress around weight and activity, and then a number of specific goals that would summarize the content of the chapter that they learned and provide their specific goal and how they might implement it over the next week between their sessions. People were encouraged to do at least one module a week. Though it's self-paced, they could do more or less according to their schedule and what they wanted to do. As I said, I think a very important part of this curriculum are the peer wellness coaches. So there's a substantial body of literature showing that the effectiveness of computerized cognitive behavioral interventions is substantially less if they have no human contact and that the programs that are the most effective tend to have some human contact during or as part of or integrated with the program, which shouldn't be terribly surprising. Though I guess people have a fondness to the robot idea that the robot is somehow going to take over everything. There are issues, I think, and this comes down to a certain part of engagement and motivation that human contact and human connection is, of course, one of the things that motivates people and engages people and that they like. We had peer wellness coaches who supported the curriculum. The idea being that a lot of the content, which is a road to a certain extent, will be delivered by the computer, but that will be supported by peer coaches who would have contact with folks over time. Who were these peer coaches? They had lived experience themselves with serious mental illness, they were peers. And they provided weekly phone coaching and encouragement using a strength-based curriculum. They were trained in motivational interviewing strategies. We developed a program to hire, train, and supervise the peers. And this was with Dr. Cohen, who's the announcer for today, so she's very aware of this. And this was, we put training in the e-wellness manual. The coaching sessions were scripted. There was a lot of work developing by the team to develop a e-wellness manual with specific content. So one of the issues of peers and consumer providers who are increasingly being implemented into mental health settings is what are the work activities of these people? And what this curriculum did is it provides a structured approach. People can, the peers are able to have a job description and activities that they can engage in that are meaningful and supportive of people. There's, of course, room for, as there is always, room for interpersonal interaction and supportive encouragement, drawing on people's own experiences and so forth, but there's also a context and an overall approach that involves a structured management of weight. And this is about, and there was intensive supervision. Peers are often lay people who have little or no experience working in healthcare systems, and so, or with, in dealing with the various kind of issues around confidentiality boundaries and so forth that clinicians are trained in. And so there was an hour a week individual supervision with a master therapist that was diminishing over time as folks demonstrated their competencies and their ability to engage and deliver the curriculum, though certainly supervision ongoing at all times, as with all people who are functioning clinically is critical. So what did we find? We enrolled 276 patients from usual care. The peers were successful in providing the coaching, the motivation, and the social connection. Patients liked the e-wellness and the peer services. They liked both of them. It's hard to disentangle the two. They were able, the people did track their BMI, they used the pedometers, they set goals, they did the curriculum. And as you can see, the e-wellness curriculum, the mean number of sessions completed, the average person completed about half of all the sessions, which is pretty good. And the average number of calls from a coach was eight, which is, I do think this is a very important part of the intervention, and it's just remarkable because these are relatively brief calls. That's not a lot of time, but it makes a difference. And then we also, in the in-person group, the average number of sessions completed was lower, but there were also many more people in the in-person group who completed nothing, or very little, because probably of barriers to participation. And as we usually see in usual care, with regard to wait, there's very little use of wait services. So what did we find at six months? The computerized peer program, a third completed all the sessions, and 31%. In the in-person group, nobody completed all the sessions, which is, I think, again, a marker that it's more challenging to make use of in-person services for all kinds of reasons. And indeed, there are certainly people that don't want to participate in group activities. We also, we found that patients who assigned to the e-wellness group, this was an intention-to-treat approach, so folks assigned to this group, whether or not they used the treatment, and that this population of folks assigned, people who were obese lost weight. And the average weight loss was 6.2 pounds. So this is a clinically significant level of weight loss, meaning it's a level of weight loss that may not seem like a lot, but in fact, numerous studies that show that even with small increments of weight loss, that there can be important changes in metabolics, things like hypertension, diabetes, lipids, and those are, of course, directly related to health over the long term. The in-person and the usual care saw no change in weight. In terms of significance, the third, more than a third of the e-wellness group lost more than 5% of their weight, again, believed to be a clinically significant number, and that was certainly less common than the other two groups. This is a graph of weight change. You see zero, three, and six months. BMI score on the left, and again, you can see a starting BMI on average greater than 35. So this is a population that was significantly obese, 30 or above being the definition of obese, and you see basically some bump up in usual care, but not much, and then returning to baseline, and then in-person group, no change in weight substantially over time, but you can see the weight, the web peer group starting to diverge from zero to three months, and then maintaining their weight loss on average. So in conclusion, what do we find? And this, at the top, this is one of the articles that we published from this work. There's a number of other articles either published or in process, and I'm happy to provide any of those to people. People can get in touch with me after this. And so, on my weight management, peer coaches can be tailored for people with serious mental illness as feasible, well-received results, reduces lower results in lower weight, and is a patient-centered approach with modest or little clinician burden, and relatively inexpensive to broadly disseminate. While there was a very large amount of work to produce the program, the marginal cost of providing it by computer to one additional person is, of course, almost nothing. So this is the sort of, there are other success stories out there of clinical informatics. We're certainly making too little use of these, I think, as clinicians and healthcare settings with patients for various reasons that we can discuss. But something people can think about, and I think there is a, people can start, there's some of these that folks can start recommending to their patients that may be useful. So, now let's move on to risk prediction and predictive analytics. This is another popular area of clinical informatics. So, what do we mean by this? Well, predictive analytics are predictions about the future or unknown events. And the idea is that you're taking data and algorithms and figure out what's gonna happen next. Sort of like a pre-crime kind of approach to beginners. And the idea is, this is a risk prediction approach. I'm gonna show here about suicide and health outcomes and the near future. The, this is experience from the Department of Veterans Affairs. So, the VA has had an interest, of course, in preventing suicide as everyone does in their other healthcare systems that have actually done this same thing. I've been thinking of group health and other systems that have had approaches like this and published on it. The idea here is to predict death by suicide. It's a challenging problem as clinicians are very aware, it's very hard to do. And this is, I'm gonna talk a little about some algorithms developed by researchers from the VA and the National Institute of Mental Health. This is somewhat of a big data approach. It took clinical administrative data from electronic data systems and systems such as the VA now are virtually fully electronic, of course. And so, pretty much all the clinical administrative aspects of care are captured somewhere. To the extent that they're in the record, there are, of course, many things that are involved in the clinical interaction that are not. But they made use of everything that was available. The approach was a, in a sense, a very large logistic regression. So, a large standard sort of statistical models that are often used. It is certainly possible to use like machine learning or other approaches. I think there are, as we can have as a discussion, there's pros and cons of either approach. And there are advantages and disadvantages of each. But the data's elements that were used to predict suicide were demographics, diagnoses, utilization, and medications prescribed. This is a slide that shows how accurate the model is. So, suicide is a very rare event. But, of course, a critically important one. And so, predicting very rare events is very, can be very challenging. And the, if you, changing the sort of the level of risk. So, the model, can I make a prediction as to what the level of risk is of an individual for suicide? Of course, the, and then you can see what happens to people at various levels of risk. Of course, if you take the people at the very highest level of risk in your model, their risk is gonna be greater, but at the same time, you're gonna miss a lot of people because of the nature of these algorithms. And, on the other hand, if you take a larger pool of different cutoff of risk, you will capture a lot more people at risk, but then very few of them will actually truly be at risk for suicide or go on to suicide. Suicide is a dynamic event. Of course, very time-bound, involves often sort of sudden labile things, hopelessness, events, changes in people's lives that go beyond, it's not a static phenomenon, though there is some static risk, which is what's captured here. And there's also, it's important to realize that even with this sort of modeling, it's highly changeable, meaning that if you do run this model one week and the next week, you will get different populations of people based upon how data is changing. So this is an area under the curve, which is a measure of how accurate the model is. And so, the area under the curve here is 0.76, which shows some reasonable accuracy. If you look at the highest 0.1% risk, which is the red arrow on the left, you'll see that they have a 30 times higher elevated suicide risk compared over time. So it's clearly, so that 30 times elevated is of course clinically, that's a clinically meaningful number, though it's from a very low baseline. And one approach that the VA has been taking is to identify these 0.1% of people and then have clinicians call them to see if they can figure out what's going on and offer services and help reach out. And this is consistent with other, will be described as low-intensity suicide interventions, which is in other settings have also, in other settings these interventions have proven to be helpful, things like mailing postcards to patients and that kind of thing. It's sometimes interesting that relatively low-intensity efforts to offer services, engagements and help to people can be helpful, particularly at areas of high instances in their life of high need. So this is, I'm gonna talk about another risk prediction modeling approach. This is again, some data from the Department of Veterans Affairs, though it's not, they're again also, there are other systems that are of course doing this. This is a very important area. Systems are very interested in high utilizers. So who are these patients who they are, who they're responsible for and are providing care for? Who are the ones who are not doing well? Who are coming into the emergency room on a regular basis, used in the hospital a lot, their diabetes is out of control, that kind of thing. Who are they, can we predict folks who we identify from a predictive way? Who's doing poorly so that services can be targeted at them? We can identify them and help them try and intervene to prevent these poor outcomes. And so this was an effort to model, use this somewhat big data approach. And I say somewhat big data because there just doesn't necessarily meet the idea. Sort of the definition of big data is a little different, but when people think of big data, they think of big data and that's what this is. And the idea here is to predict hospitalization or death as negative outcomes. Again, this is a very large variable, many variable logistic regression. You can see many of the variables on the right. It's an effort, their efforts to trim these models and identify which variables are helpful and which are not. But you can see that it's a very wide range of variables. So it's things like, are they getting particular scans? Are they getting particular lab tests? It's not necessarily even the results of lab tests, but are they getting certain lab tests? What are the results of certain lab tests? Are they prescribed certain medications? So these are things that are sort of indicators of health status and indicators of what clinicians are concerned about in doing that when put all together can help to identify these folks. So in this case, how accurate is this model? This is the C-statistic for these models, which is again pretty good. And what you're looking for is a predictive accuracy. And this is predictive accuracy. This is identifying people who in terms of their risk for death or hospitalization over one year in this example. And everyone, it's what's useful thing to think of is that everyone in the healthcare system has a number. So everyone's number is a percentile from one to 99. So there are a lot of people out there, of course, with 99s, which means that the chances of them being dying or winding up back in the hospital are like 99 plus percent, so very, very high. People don't necessarily know what their number is, but the system knows what their number is and is adapting services. To focus on these folks, to try and make a difference. And of course, there are gonna be some people who are 99 who have end stage cancer and hospice or something like that. But then there are also gonna be folks who just have a number of numerous acute medical problems or serious medical problems in particular areas that are placing them at risk. And so what you see here is that the veterans in the highest percentile of risk in this study had a 58% probability of admission, 23% probability of death, and 64% probability of either event. So it's reasonable in terms of the, what was expected and observed, which is what you see in the graph with red and blue are relatively similar. So the model is doing a pretty good prediction, a pretty good job of prediction from a statistical perspective. And this is the sort of thing where teens, clinical teens can be constructed or focused by knowing the risk of the folks who they're treating and providing the services. So, and what are predictive analytics being used for today? It's for, you know, it's been used for case management, care management, selecting patients at high risk, managing a panel by health risk. And also for research, it's possible to use these approaches to design and carry out and develop research studies. So you can, if you want to do a study where you want to see if some particular intervention or approach is helping people that are particularly in need and are at high risk of a poor outcome, as you then, if you can focus your study on those folks, since they're the ones who stand most to benefit, and then see whether it works. And so it's, so I think I would say both use for research and augmenting practice. And this is also, this is a dynamic area of development. So there are various other approaches. Again, people have also used machine learning approaches or sometimes referred to as artificial intelligence approaches to this kind of effort. And, you know, the methods that are used, I think depend in part on the data, what data you have available, whether you care to know about the process of what's going on for the risk, what is determining the risk, or whether you just want to accept whatever the computer, what interrelationships among the data that the computer can find for you as being generalizable and useful. So where is this heading? You know, so on the right, you see an example from the facial recognition, which is of course an area that is both very interesting and also concerning to people from a privacy perspective. So basically computers now through facial recognition software, as I think people know, do a very good job of identifying who someone is and basically matching their face to previous images of them. So if everyone has an iPhone, I think you've drilled around in your photos thing, you can see that your iPhone is able to group all the photos of you, different people, family members, with a reasonable degree of accuracy by identifying who they are. And this, of course, has been stories in the news about the Chinese government using this with cameras on the street to monitor people and certainly could be used by security in the United States too in places where there are frequent monitoring to identify who people are and follow them. It's, and you can argue the pros and cons of this, but it's happening. So at the bottom, you see an example. This is just a picture that's used for machine learning, which is another talk, but it's an example of how data, the complexities in data can be understood by machine learning approaches, which are, machine learning approaches are particularly well-suited to things like facial recognition, image processing, image recognition, and audio processing, those kind of things. You know, Siri, for example, or a Google Translate or things like that or machine learning algorithms. And it's a computer approach that is very effective in that sort of use case. So what are we headed for tomorrow in terms of mental health assessment and treatment? Increasingly, we will be using really big data. And by really big data, I mean things like data streams generated by sensors from phones or all the other sort of internet of things or electronic devices that are increasingly around us that are constantly producing data. So these produce data that is a very high volume, high velocity, high variety, and low veracity, meaning that the data is enormous flow of data and the veracity, meaning the validity or the meaning of all the data elements is highly inconsistent. And so these are things that big data algorithms have to deal with. Patient-generated data. Again, these can be things like clinical status. Patients are increasingly using patient-reported outcomes or other ways entering data into tablets, that kind of thing. Speech recognition is, of course, going on all the time with Siri and other Alexa and your other various machine learning friends. There's a visual processing activity and purchasing behaviors. Activity and purchasing, as everyone knows, that buys anything online. Facebook has, and Google, have managed to pull together these data using cookies and other technologies so that when you express a search and interest in a particular type of shoe in Google, you'll then be bombarded with offers to sell that to you for a period of time in whatever various apps and phone places you are. So then also diagnosis. There's an increasing interest in trying to use all data from electronic medical records systems and so forth to see if we can improve diagnosis. There's interesting work in each one of these areas. Predicting behavior is, of course, something that people would like to do. Human behavior is challenging to predict. But again, I've showed you some things in terms of suicidal behavior and health behavior. There's, I think, certainly opportunities to improve the prediction of behavior, the future prediction of behavior, using these machine learning and neural network approaches. So this is a famous quote, which it's, big data is like teenage sex. Everyone talks about it. Nobody really knows how to do it. Everyone thinks everyone else is doing it, so everyone claims they're doing it. And there is a lot of work going on. It seems like a very sexy topic, big data and machine learning and so forth. But I would maintain a certain degree of skepticism about what is actually going on and which instances this is actually going to be useful and produce meaningful things and where it may, in fact, be a dead end. In terms of folks who are interested or maybe thinking about clinical informatics or think this interesting area and so forth, the American Medical Informatics Association is a large professional organization in this area, which is interesting. They have journals. They have a lot of trainings. They develop clinical informaticists. Office of the National Coordinator. This is the federal government office that oversees data sharing and EMRs, electronic medical records, which is, of course, very interested in this area. There's a whole variety of conferences in clinical informatics, fellowships, board certifications, other certifications of various sorts, and then job opportunities for folks who want to develop their work in this area. So I'd like to thank for your attention and joining today, and for those of you who are live with us, we'd be happy to take any, I'd be happy to take any questions.

clinical informatics

mental illness

treatment

eWellness program

weight management

peer coaching

predictive analytics

risk prediction