Melanie Cole, MS (Host): There's no handbook for your child's health, but we do have a podcast featuring world-class clinical and research physicians covering everything from your child's allergies to zinc levels. Welcome to Kids Health Cast by Weill Cornell Medicine. I'm Melanie Cole. And this is part two of our two-part series highlighting the signals we miss in youth suicide prevention.

In this episode, we cover how better measurement reaches the young people we are currently missing and how social determinants of health can really help us to assess risk. A quick note before we start, this episode does contain material of a sensitive nature, including discussions on suicide and teen mental health that some listeners may find disturbing. We know this can be a difficult subject, so please take care of yourself while listening.

Joining me today is Dr. Yunyu Xiao. She's an Assistant Professor of Population Health Sciences and Psychiatry at Weill Cornell Medicine. Dr. Xiao, thank you so much for joining us today. We have done a part one, and now this is part two, and you've said that we're measuring the wrong signal. You told us about the paper that you wrote. And we're measuring the wrong signal when it comes to the mental health of our children. Tell us what you mean by the wrong signal.

Yunyu Xiao, PhD: Yeah. Thank you, Melanie, for having me again. So, the wrong signal means that almost everything we built in suicide preventions listens to one signal, which is one person telling someone that they have suicidal ideation or thoughts. So, the rise of the last two decades is happening in people, however, who never tell anybody. In the recent work that we're doing, we use the National Registry of Violent Deaths to split the suicide rates of deaths of people into two groups. Those with documented disclosures of intent beforehand and those without.

Essentially, the entire rise in the US suicide mortality sits in the group, however, with no documented disclosure. And then, the gap between these two groups have been widened substantially. And then, the deepest rise in an undisclosed group are in exactly the people we most need to watch. Young people aged to 10 to 19 years old, and the fastest are all among adolescent girls and Black adolescents. This is very important because most people we know that died by suicide at their first attempt. So if someone never disclosed or, like, having a prior attempt documented, every system is keen to really flagging these two signals but never seeing them.

So, what I mean about measuring the wrong signals means that we optimize for the people who speak up and then who we can document. But the epidemic moved to the people who don't. So, this is why in our research, we look at those quieter or less frequently studied signals, and then we wanted to uncover those signals like behavior patterns, like addictive use, the text of these death records, and the conditions like the neighborhood of people living, as well as their social networks.

Melanie Cole, MS: This is a really fascinating part of this whole topic, which as I said at the beginning, can be a little bit difficult to discuss and hard for people to hear about, but hear about they should because it's so important. Now, we're learning more and more, Doctor, about machine learning and AI and all of these things coming into play and in your research. Can a machine read a death investigation note? I'd love you to speak a little bit about what is even a death investigation note? And what are you finding when you look at these records?

Yunyu Xiao, PhD: So, every suicide death in the national registries carries a coroner, we call it medical examiners as well as law enforcement examiners' narratives. So, free text right now becomes a very rich resource for us because it describes the circumstances, the stressors, what was said, and what was found. And ten thousands in the years that was produced by this free text. And with the colleagues in biomedical informatics as well, we use natural language processes to read them at the skills. So, what at the surface would say that circumstance is at the chalkbox, like a very granular, like general, in terms of like their causes of death. And however, these intentions have never gotten to be documented in details.

So in the decades of the datas, we find that disclosures before death went to the intimate partner, like 40% of that, and then 35% went to the families, and 5% only went to these healthcare workers. So, the safety net, we find, is not where the real signals goes. And that means that with the tools of reading those thousands of these records of deaths, they do not predict individuals who would actually have the suicidal thought attempt. But this really tells the policymakers, and also actually tells ourselves, like as the family members, that prevention should go to the social determinants of health because that is where it connects people to 24/7 instead of just staying the suicide prevention at the hospital.

Melanie Cole, MS: In your Nature Mental Health paper, grouped those social determinants of health and linked them to county suicide rates. So, Doctor, how does ZIP code shape a young person's risk? Because I think this is also just such an important point.

Yunyu Xiao, PhD: ZIP code is a fascinating digit for us to look at because we ask simple questions at a national scale, which is the first times of documenting it. Say that if you let the data just speak by themselves, what kinds of places in Americas is made of, and how does each kind of this clusters of neighborhood or social determinants of health would associate it with suicide deaths? So, we took around 284 social determinants across 3,000 counties over the decades, and let the machine learnings group, each of these counties, nationally and naturally. So, three profiles or clusters, we said, emerged. And we named them as their characters. We call them remote, cope, and diverse.

So, remote means that they are living in the places which is having a lot of rural area, older, marginalized environment, older housings, empty houses. And about 60%, in fact, of the US counties belongs to this group. And the second one we call it COPE, C-O-P-E. That means that they have complex family dynamics and structure. They have absolute poverties and have these like healthcare consumptions and extreme heat weather. And the third one we call it DIVERSE, which is acronym for living in a densely populated cities that are having immigrants and inequalities is very high, very expensive housing. So, those are the three types of the counties. And each of these county profiles have their own suicide signal. We find that the remote counties had the highest overall suicide rates, especially among men. And then, the COPE have elevated rates over the decades of white residents' suicide rates. And the final ones, which is what we are interested in, that for children and adolescents, they have the highest rates when they are living in the diverse counties in terms of suicide, as well as also if you are a woman as well. Particularly, we also find that Black and Hispanic adolescents living in those diverse counties also showing the highest suicide rates.

So, the points of these profiles over just poverty scores, like we usually see that in index scores, is that the index tells us where the county is doing badly, right? But the profile is kind of like, say, MBTI personality test. It tells us how it is doing badly and which is what the policymakers can act on immediately.

Melanie Cole, MS: Well then, if prevention is partly housing, income, and access, as you were discussing about access to healthcare, and that's the story rather than just a clinical one that we would look at that way, what does that change when we are looking at the ways to identify and assess that risk?

Yunyu Xiao, PhD: Melanie, you ask a really interesting question. And in fact, before I came to machine learning and health informatics and AI to this field of research, I was a political science major undergraduate. So, what I always advocate is in terms of the policy changes, and of course, it relates to the budgets, right? So if the risk lives partly in the conditions where people live, then some of the most powerful suicide prevention doesn't have the word of suicide in its name. In fact, it goes to social determinants of health policies. Say that in the remote counties, prevention would look like investing for the broadband and telehealth, housing, transportations, repair, and firearms storage partnership. Because services are far and means are so close to them, and that's why it would increase the risk of suicide. And say in the counties, in the diverse counties, that it looks like language access in terms of, like for example, we have research showing that if you're communicating with people who are more proficient in the language of your native language, you're more likely to seek out help. And also, culturally competitive cares, and also like unemployment support, housing affordabilities. Because those excess risk sits with the women and the communities of colors and would squeeze the hardest, the like social conditions that they are struggling with. So, the government should take the responsibilities in terms of taking care of these social factors. And the clinical system still matters enormously because the shift is from asking which patients only to also asking which places and which conditions and which neighborhoods that they are living in. So, upstreams and downstreams are not in a barrel. They're actually the two halves of the one strategy that we should really advocate.

Melanie Cole, MS: When we look at national averages, and I think I know the answer, because you've mentioned a few times some of the people that are at higher risk in which communities, but who are we missing?

Yunyu Xiao, PhD: Yeah, exactly. The average is not comforting fictions, like almost no one lives at average. And even in the year when the national like suicide rates fell, Asians, young adults rose about 12%. And over the long period, suicide amongst Black adolescents, especially Black boys aged 10 to 19 years old, rose about 160% from 2005 to 2021, and more than doubled, which is the increase among their white peers. As well as like people living in rural areas, those suicide rates grow really twice as fast as those living in the urban areas.

So in our research, we also find that young Black individuals who die by suicide, most had not touched the healthcare systems in their final years, and often for the reasons that they never labeled their care as mental health. So, they were seen, but not seen. And then, the fastest like rising individuals in 10 groups are adolescent girls as well. So, they are the persons who may not be able to disclose and not give you a safety net.

Melanie Cole, MS: So, I'd like you to take us from bench to bedside here. How does one study travel from a dataset, which is what you're discussing, to a national strategy? And I'd like you to really emphasize for us whether suicide risk prediction models are really ready for real health systems. When we're talking about everything you're giving us information on today, how does this travel to our national collective?

Yunyu Xiao, PhD: I'm happy to give you some concrete examples that, first of all, we talk a little bit about this ABCD data set. It can be an example that the findings that these addictive patterns, not hours, would carry the risk. So practically, the translation could be that in a pediatric visit and the school programs, that we could shift from just asking how many hours does the kids have to four questions about the loss of control, the repetitive, like, check-ins, and then, also the distress without using it.

And then, to check it, like in a checking point, repetitively throughout their developmental stages, starting like age 11 till the 14 years old, because the risk is the trajectories, not just one time point. And similarly, for the county examples, that's why I'm interested in studying individuals and then the counties. That is warning signs saying that after we're profiling the three types of the county clusters of social determinants, and each have the distinct suicide signals, right? So, that means that the federal and the state prevention dollars should match the profiles rather than spread evenly. They're not the average.

Not every county get the same type of the treatment for the health economy and healthcare. So, the remote counties should just really be invested more about telehealth and safety stories, while diverse counties should be more investigating in terms of investing their, like, professional services for health and language access and housing stability, support about employment, et cetera. And those data does not just describe the map, but is really in terms of, like, where the money goes in terms of the budget.

Melanie Cole, MS: So, the budget, where does the money go? How is it being used? Dr. Xiao, I mean, you've just given us so much to think about in this two-part series that we're doing on risk and assessment and not missing the youth that are really at that highest risk. If you had to give us your best advice, the single most important thing that you think parents and policymakers should take note of in your research and that we all need to, again, I'm going to say as a collective, understand so that we do not miss signals, because that's what this is really all about.

Yunyu Xiao, PhD: The strategies of youth suicide prevention is not very simple. And as we discussed throughout all the research agenda of my whole journeys, I think it's across at least the three areas, like for the parents, and then for the healthcare systems in terms of research, and also for the policymakers. And I would say that starting from our health systems, because I work in hospitals as well, and I would say that our models right now in terms of research of suicide prevention is ready, but our systems are not. And this is literally the title of our newest, like, JAMA Psychiatry viewpoint.

So, the honest bench-to-bedside status is that we have models that perform respectfully where they were built and weakened when they travel to another systems. And validating them is very expensive, and then the precisions is the rarest outcomes, like, say, suicide deaths remains very low. So, the equality problems, as we also discussed, is the sharpest, like, problem, like safety net, rural and community systems, those serving the highest-risk patient hospitals are the least resourced to validate models by themselves. So without a shared, like, infrastructures and also the research investigated, the people at greatest risk are always being missed. So, we want to propose that we want to build, like, a grasp system to grasp, like, each of these missing element that we want to build and advocate that we should definitely build a shared, like, and then this infrastructure where systems benchmark models on the common data standard without centralizing the patient records. And then, the question I ask for every health systems is not does the model detect risk, but what prevention can this model actually trigger and with the staff and service that you have?

And of course, for the policymakers then, prevention that waits for people to raise their hands and now misses where the epidemics is growing. So, really funding and specifically say using the three type of social determinants of health profiles to guide your budget is very important because then that makes this upstreams preventions more targeted.

And then finally, for the parents and the families, like, the children at risk are often looks quiet, not dramatic, and you should always ask directly, and then keep the device out of the bedrooms at night when they sleep, lock the means, and know that when the adolescents, like, do tell someone, they tell someone like you. And you are actually the front line, and you are just the person to start the conversations. And every element of their social networks and their social environment are actually important, and they teach each other together. And then, we can really create, like, a system to protect our kids.

So finally, I would say that we saw the national numbers moved in the right directions, but the next move really belongs to the person that's the average that the surveys hides, and we know, and we have the tools to find them. So, I always make the analogy that we've seen the tip of the iceberg, but now we actually have the X-rays to see what is below. And now, we should take the action because if it's not now, then when? And if it's not you, then who should take the responsibility to save the kid's life? And one person is too much.

Melanie Cole, MS: Beautifully, beautifully said. And such important information. And I'd like to point out that the Suicide & Crisis Lifeline is 988. So parents, if you have any worries, red flags are going off, you think that there's something going on with your children, your instincts are very often correct. You can call that number. You can contact mental health professionals. Your pediatrician is a great place to start. So, I hope that you will share this show with your friends and family on social media, because that's the way that we are all learning from the experts at Weill Cornell Medicine together. And Dr. Xiao, thank you so much for joining us today.

This has been part two of our two-part series highlighting the signals we miss in youth suicide prevention. Please check out part one if you missed it. And Weill Cornell Medicine continues to see our patients in person, as well as through video visits, and you can be confident of the safety of your appointments at Weill Cornell Medicine.

That concludes today's episode of Kids Health Cast. We'd like to invite our audience to download, subscribe, rate, and review Kids Health Cast on Apple Podcast, Spotify, iHeart, and Pandora. And for more health tips, go to weillcornell.org and search podcasts, and don't forget to check out Back to Health. Thanks so much for joining us today.