AI as an Early Warning System for Mental Health

Richard Zwicky: Hi everyone, it's Richard Zwicky with High on Healthy, the show where we know there's no one path to healthy, so we explore them all. As I'm a big proponent for the wise use of AI and I believe that AI can help predict mental health crises before they happen. But it's going to take a lot of data and it's going to take a lot of learning for us to get things And that's quite honestly why I built the Sparks Health platform is part of that process. And it's a small part, but it's a step towards it. And we've had a number of discussions on the show about AI and how it applies into mental health. And most guests we've interviewed have focused on AI as a use when somebody's already struggling as an adjunct to therapy, chat bots, in between session support, and the like. But when we take a step back and think about it, imagine if medicine worked the way mental health often does today. We'd discover heart disease after the heart attack. We'd diagnose diabetes after organ failure. We'd have death from cancer before we discovered that the cancer was there. But mental health, that's often how we're behaving today. We're waiting for someone to reach burnout until they collapse until anxiety is disabling or addiction hits and then We don't look for the signals actively beforehand. We don't question each other and we don't think about the question how do we treat mental illness. We think what can we do about the existing crisis? And today's guest believes artificial intelligence could fundamentally change not just mental health care but the timing of care itself. Dr. Sar Van Tangara Jan Rajan sorry is thinking at the population level how entire health systems fail to identify people before they reach crisis and I'd frame it around the premise mental health isn't failing because we don't know how to help people it's failing because we don't recognize the signs before it's too late welcome aboard the show Saravan Nan

Dr Saravanan Thangarajan: Thank you Thanks for setting the stage and making it clear what the context is about. Yeah. So that's one of very important topics which is currently less addressed and as with the advent of AI technology the amount of mental health pressures is more than the more aid or what is being kind of facilitating around. So that's a very essential and more important topic of discussion which and thank for bringing it today.

Richard Zwicky: It is. And just for context for the listener, you're with Harvard University and you're working on these questions day along with what else?

Dr Saravanan Thangarajan: I'm basically a clinician. So I started the journey at talking onetoone with a patient trying to do a treatment for them. Trying to plan treatment for them and due course of time I kind of understood okay fine this is not how the system works or this is not how the care is going to work then I had to step back and see okay fine the person is not just talking about just individual care there's something which needs to be done at a system level who gets taxes who needs that care who's within the system who's out of the system and what happens to that so that those kind of conversation help me to think beyond okay fine let's think beyond what an individual care a person cares about to think how thousands of people are lost in between the process of being identified being addressed and being treated and being in the system basically so that kind of thought process made me to think beyond the clinical setup to get into a space where a research could really help get me those answers to and that's how I kind of landed up at Harvard where I did my research which is focused on how my primary focus on vulnerable communities especially women and children disabled communities. Trying to focus so all the answer we know that everyone is going through a lot of pressures in life but there is something more to I would call this an invisible community or invisible population because we all assume a mother has to take care of a child and mother has to take care of the home. It becomes then duty of a mother to take care. But what are the things she under goes during the early phase of life when she's a new mother. What are the ch leave behind The psychological impact are more to it which is never addressed which is never brought to the light or even understood. These kind of conversations never happened as an open talk or discussion. As a clinician or being I was also part of the government during my earlier phase where I was u heading a couple of mental health projects, maternal child health projects. So all the addressed basically rock idea was like are you okay? Did you get your blood tests done? Did you get your routine? Did you get your vaccination done? Did you get your scans done? That's the conversation. And when you have those data as yes, you feel your mother and child is safe. That's how the vulnerable communities are at risk.

00:05:00

Dr Saravanan Thangarajan: But that's just the mere survival of someone a mere a person exist live or die. You're not talking about quality of life here. We not talking about the being. We want to talk address about something which is more human. We just talk about numbers whether the person lives or die. So that's something which really struck me hard are we just getting those numbers? Are we considering human as just data like any numerical numbers or data is that all we care about being clinic because I've been at the other end of the spectrum being clinician treating them and then coming to administrative role where I just see them as numbers. So I couldn't really kind of what to say to beat myself or kind of understand is this the right thing to do. So that's kind of made me to deep dive into it and I kind of worked out with my project with my mentor there at HR world and we did a research project which is focused on how climate is impacting these families and later I kind of did a couple of other consulting projects for the California health human services and then a project in Vietnam. All of these are focused on climate or climate change indicators. Climate is I would say an environmental issue. It's more of a public health issue which has deep impacts on mental health. It's not just we talk about what's happening around us. It also impacts your inner self inside. How safe and secure you feel. So it is in mental health context too.

Richard Zwicky: It does and what you're describing definitely has value but thing that strikes me through it is most health care systems around the world are designed to diagnose illnesses before the symptoms become too severe. You do the tests, you look for the markers and you prevent things. Mental health care is always dealt with as reactive. Is there a system which is looking at part or all of it from a preventive perspective? And if so, how

Dr Saravanan Thangarajan: That's really an interesting question. So I see the value in technology where it kind of helps a person to reach appropriate helps health support or human support earlier but it can help identify the change in the risk because everyone does not undergo the same kind of risk same kind of situation and kind of reduce the administrative burden the support system which is available for them. So one help or one kind of assistant which a person need may not be the same for the other. So it is not universal system which is going to help and also the health system also does not have the same policy recommendation for all and everywhere. So talking about the mental health technology becomes dangerous when it creates the appearance of care without the accountability of care. So we just talk about okay this person been addressed or not but what is the final output? How has the person gone through the process? Was there a real change? So a system may sound empathetic but it does not automatically understand the context, the culture, the family dynamic or the clinical urgency. So all these technology help you to diagnose. Yes or no? Yeah. But that's helpful but it helps you to shorten the time. The technology help you to maybe screen more population. That's all. I believe it cannot replace the real human healthare intervention which is needed which is very I would say personalized care which cannot be like a technology cannot do that. So my standard is simple. Does the technology kind of strengthen human judgment access or it just gives you a continuity of care? I feel I kind of keep juggling between the other two end of the spectrum or does it just quietly transfer the responsibility of just giving those numbers and keeping the trained people or the institution away from where they are supposed to take up the responsibility and keep moving forward. So I think artificial empathy is not the same as accountability of care. I believe

Richard Zwicky: No, it's not. But it's a tool that a lot of doctors don't develop, but is valuable for training or for dealing with mental health because we create the bridge across which a patient can cross to speak about and explain what they're suffering from. That otherwise they're fact facing shame or even unawareness of how it's affecting them everything around them. And across almost every interview in the show whether we've been discussing burnout, addiction, loneliness, suicide, any trauma, the same patterns keep emerging. People struggle silently for months and years before anyone notices.

00:10:00

Richard Zwicky: Sometimes before they can even step back and

Dr Saravanan Thangarajan: All right.

Richard Zwicky: Recognize what's affecting them. Is our biggest problem? No. Not really treatment but detection. And is that where you're looking to focus

Dr Saravanan Thangarajan: So that's something that's core work which I'm kind of because as I mentioned there was not much of people talking about the impact of climate or the impact of technology happening on mental health. We just saw about how the system is going to make it speed faster the work. But my research kind of made a ground or kind of formed a base that we could develop more content or kind of have more research questions on it. At the same time I'm kind of working on technology which kind of works on early detection such as your wife residence because most of the apps tell you to only when the patient has been diagnosed with some specific situation that's when these kind of so-called wellness apps kind of help you to intervene. So I'm kind of being thinking in a more proactive way and trying to work on technology which just simply need a person to just have a conversation with them. It doesn't have to be anything related to your mental health or the way you go. It could be even a very normal healthy person. And one more thing which I wanted to talk is many of us do not really reveal ourselves in the first phase. Be it a person, we always try to present the best part or the best version of us. So if even for an app they might initially try to hide the emotion. So that is a very natural human tendency we don't want to burst out or be ourselves in front of something a new person or a app anything it could be like with These are the very critical phases the critical time which we kind of miss I would say that the early diagnosis or early warning signs I would say so I'm kind of developing a system where it kind of helps at the very early age just a simple conversation you don't have to talk about yourself you don't talk about your problems you don't talk about anything that disturb you just a normal hi I mean I got up this morning I had this breakfast and I'm getting back to work so I'm trying to capture the voices the neural tones I'm just trying capture those pictures, the silence, the pauses, the phrases of words. So my kind of the wellness app which I'm developing kind of going to capture that and help you to identify what the person mood because a person can undergo an array of mood over a period of day over a day right from morning they could be happy midday they could be stressed evening they could be relaxed. So this is kind of going to capture all your emotion over a period of day and try to analyze whether the person is really stressed because of work stress or it could be kind of a temporary or it's something which is kind of building up within them which could lead up to some kind of a mental disturbance. So this is just a normal conversation which is happening and the app is going to run in the background do its own algorithm and trying to understand what's happening with the person. It is just a temporary fit or something which is more deeper. So definitely what I'm trying to develop is not something which is a replacement of a therapy, replacement of a physician or replacement of human intervention. This kind of healthy therapist a physician or a psychologist to address in a better way usually what we tend to do in the current day as clinical therapist when we see a person that's disturbed we usually evaluate them for two three weeks and then most of them has similar symptoms which can overlap depression or PTSD any kind of viral thoughts could overlap to lot of different diagnosis so we should put them on medication or drug and that becomes lifelong. So there's no stopping or no getting back from it. Maybe it could take years. Some of them even die having those medications. So we all categorized at a very fast because we don't have much time to really diagnose and categorize each individual in that space. So we give them medication within two weeks of time. So what I'm thinking is an app which is kind of going to work along with the physician along with the doctor.

Richard Zwicky: But I'm big proponent of apps and

Dr Saravanan Thangarajan: So it helps you to categorize the patient or the individual through a span of two to three week. So you can modify your treatment plan which could help the physician in different way.

Richard Zwicky: Tools and different things and aspects and it was interesting Elon Musk was on TV yesterday the day before and he was talking about his Optimus robot and the conversations focused around surgery and his argument was that most surgery is routine and that it's standard moves practices if the patient is laid out properly and robots could do that more efficiently than doctors and don't require constant training whereas doctors require they age out and the new one has to train up to that level whereas the robots are continuous.

00:15:00

Richard Zwicky: That's absolutely true in some contexts because there's a defined number of factors that have to be dealt with in health care. If we just look at the AC study and then add gender, age, two other factors, where your ethnic background and where you live are going to affect how any of the aces affect you. But those 14 combinations can be combined. 265 million different ways for an individual. That's a lot of factors. When you factor in all of somebody elses somebody's life experiences, it's also the factorization becomes enormous. And humans are remarkably good at hiding distress and high performance especially. AI can be used because it's a mathematical function to help detect those patterns incredibly rapidly. But where the challenge is the human therapist is used to making the leaps and the connections between what would be disperate data points and recognizing how they could fit together in different ways and testing them on the fly. AI still has to be taught a lot of that. How are we going to do that? And how are we going to tie in some of the biomarkers that AI also relieves in or believes relies on? Not just, I'm not doing well, but also the factors like somebody's blood pressure, cholesterol, glucose, other things that are also markers of mental health disease or injury more importantly.

Dr Saravanan Thangarajan: I mean I do share a part of the ambition which alon has been talking about and using AI robotics to make health high quality healthcare which is more widely available for and accessible for all but the cap capability alone is not care I believe. So healthcare also requires trust the context the continuity as I already mentioned accountability. So these are some things which cannot be replaced. So my view is that AI should extend the reach of these trained professionals help recognize the risk earlier. So all these technology help you to identify the risk earlier and kind of reduce the avoidable burden such as which could not become a substitute for a human judgment. I would always leave as a human judgment can never be substituted even transfer the institutional responsibility in the first place. So the real test is whether the technology can perform a task. Whether not it is understood the technologies are doing really well but the real question is are they improve the safety check is it equating an equitable access for all people who are currently being missed out. So that's the real question about these technology or these robotics are yet to answer and I'm sure these are something which has not been touched on. We just talk about the efficiency how it's going there's a lot of talk about replacing the physician but that's never going to healthare profession that's not the point which I strongly believe a clinician or healthcare profession can never be treated it just kind of shortens the time for the early diagnosis

Richard Zwicky: And prediction is powerful but prediction without intervention is almost cruel at times. If AI can identify someone at high risk of depression or suicide, what's the ethical responsibility and how would you account for it?

Dr Saravanan Thangarajan: So as I already mentioned the accountability is something which also relates to ethics and account ethics and responsibility because the AI can help you to kind of identify it but moral ethics of how you handle the situation to because these are large big number of data which could be exposed to anyone and that could be manipulated in different ways how we already have these kind of our data being used up by lot of big M\&P and corporate to manipulate right now I'm taking loans or buying a new shoe anything so that's how the algorithm the being used but that definitely cannot work in terms of mental health which is I would say a confidential space or confidentiality has to be maintained and that's where the ethic comes to play it's a very crucial topic or a crucial situation where this needs to be so I feel these corporate trade or the AI can never replace the human intervention which has the ethical or moral decision to make when treating an individual where cannot it is irreplaceable that's what I feel and whatever technology whatever algorithm you bring that is not going to be something which is going to be incorporated or a part of the technology

00:20:00

Dr Saravanan Thangarajan: Thank you.

Richard Zwicky: But that becomes a difficult part because in my system today if somebody voices information that's indicative of a risk of self harm the system brings up a message questioning them on that but suggesting they dial into one of, let's say, 988 or something. And it could be tied in directly to asking if they want to speak to a therapist. But if you're in a therapist's office and you voice markers that they deem to put you at risk of self harm, they're have a duty to act. When we're extending AI into therapeutic practice, drawing that line is on one hand, it's an established practice. On the other, we've never given that type of consent to a system before, but we still need to protect the user. Where are the boundaries? Where do you see do you see the boundaries actually negatively impacting people's willingness to participate or do you see it as a necessary step for consistency?

Dr Saravanan Thangarajan: Self harm is something which is most challenging areas in terms of healthcare because it is often reflect the combination of emotional distress, your trauma, mental illness and the social circumstance. So it's not a single thing which could be directly addressed by so-called your AI or technology has the potential to help identify those patterns to some extent and we kind of indicate who's at a higher risk or who could wait or kind of c or I would say triaging system but it should never be viewed as a replacement of a clinical assessment on human connection. So ITI is an early warning and support tool but it may help recognize the cha changes and challenges. However the language is the nuance the behavior or I would say the speaking patterns that kind of warrant a clinical attention may not be definitely done by an AI. So however any signal generated by AI should lead to a timely human evaluation intervention and it's not an automated decision like deciding for a human. So this is what a process like as I mentioned it fast in the process but definitely none of these can be replaced by AI

Richard Zwicky: No, it is a challenging issue which a lot of people do struggle with. Suppose your vision succeeds and AI detracts stress earlier and clinicians are able to intervene sooner and outcomes improve. Where does personal liability shift at that point? Do we expect the people to manage more and differently? Right.

Dr Saravanan Thangarajan: I would keep this in different levels as an individual and as a over system level. So, AI can detect the high suicide risk and system the can assign a 95% score or nobody. Then there's also a possibility where we have so much of data coming and pulling in but who going to view review all of these alerts do we have session manpower do you have session and who's going to do the followup and what happens when the patient has lost due to some kind of a mishap during this process. So who's going to take up this liability and whose is it you going to blame the AI or the clinician because AI is going to give you pull in thousands of data about several client or patient you have but a physician have limited time of scope limited time even they have their own burnout they have their mental stress so if something must happen if everything goes we all looking for a good positive response from AI and improve our mental health but it's not going to happen everywhere and every time there is always a possibility of missed between the data which we receive from the AI and from the clinical intervention it's not blaming the eye is giving the mito wrong or the clinician is not working it is the gap who's going to balance that gap so we don't have the sufficient manpower to handle all of this so I feel AI should not only detect the risk but it should also kind of trigger I

00:25:00

Dr Saravanan Thangarajan: And as you spoke about probably taking the responsibility and if an AI system could identify someone who may be at risk for self harm there should be a clear defined clinical pathway for it that's kind of I mean thinking in a more forward futuristic way it should also have some kind of a clinical pathway which should have who views the alert and who should be assigned and what is their case load that is the next big algorithm which has to be addressed And how quickly are they being If they are not addressed then what is the next clinical pathway or next channel because AI should not just stop at just trying to find if that person is being addressed or not the next what is the channel what has been done to that specific patient and what documentation is updated documents available has there been any improvement. So that is something which is kind of iterative next big phase which has to be because so far what we have done is just identify it and get that data what happens the clinical followup is usually done by so I think these are all thinking down the lane which might help or reduce the burden but I mean there's not been any kind of work which has been happening this way so the technology without governance can create a dangerous illusion of safety I believe so I These technology if they are not like guide or channel life or have a clinical pathway it could be definitely dangerous than what we think it could be beneficial.

Richard Zwicky: And that whole issue is basically around it's incredibly good at prediction but fundamentally incapable of wisdom. It may know what that I'm struggling with but not understanding why. Is prediction enough to save lives?

Dr Saravanan Thangarajan: Enough to save lives. Yeah, that's a good question. But I mean I would treat that as more of a hypothetical because tradition is something which you feel could go both the ways. A tradition is not something this is going to happen like person is going to have a negative impact. It is just trying to understand which direction a person might move. So when that is kind of a hypothetical and I don't think AI can kind of really be a perfect guide and that is going to help you to do the necessary step. So when a human intervention is there I think that I would call it personal instinct or a clinical judgment is something that kind of help you to really understand whether the person might definitely go through that or need that specific or kind of detecting that earlier can and also talking about how ethically or more moral responsibility of that individual has to be taken care by the physician or the treating physician so that the so-called initial definition can be considered or taken or maybe translated into action. So deduction without an accountability response accountability is definitely not worth and it's not worth the time or the process of what we're doing because without knowing and what are the next step having those high-end sophisticated technology or is not going to really work because until the implementation phase or implementation signs work along with the technology these are just going to be mere tech based platform which does the work faster but doesn't have an final output.

Richard Zwicky: Saran, we're out of time for today, but people want to learn more about the work you've been doing and more in general with regards to everything you do. How can they reach out? How can they learn more?

Dr Saravanan Thangarajan: So as I mentioned I'm kind of a person who more active on more of academic research. And you could reach I'm more active on LinkedIn that's the space I work on and I'm also updated with Harvard and other I've also been updated with some of the organization who are working on dementia research and with welcome press which is working on mental health and climate impact so I'm working in different research and academic spaces and also do I'm doing a lot of work with the government and different parts of the world so you could reach me out on LinkedIn so that the best page I'm on And it's just my name LinkedIn saran\_ tangent. So that's how you can reach me out.

Richard Zwicky: All right. Thanks so much for joining us today on High on Healthy S.

Dr Saravanan Thangarajan: Thank you so much. Thanks for the time and it was really wonderful to chatting with you and as I mentioned already it was a very essential topic which definitely needs more conversations about it so that people are aware and also the technologies well adapted to suit our population and our community. Thank you so much.

Richard Zwicky: Thank you so much and thanks to everybody for listening. I'm Richard Wiki with High on Healthy and we'll be back again with you soon.