Psyrin and the Future of Vocal Biomarkers in Psychiatry
"I Can Hear It In Your Voice" was a good title, but it has worse SEO
Since 2013, when I was carpooling with Dr. Leonardo Lopez to my general psychiatry residency training program at the Zucker Hillside Hospital in a too-long drive from Brooklyn to Outer Queens, I have been fixated on the possibility of ahypothetical psychiatric diagnosis. I wanted to understand what was wrong with the people I was caring for without the conceptual anchor of a—frankly—made-up diagnosis in a book that is an approximately $25m/year source of revenue for its publisher. The DSM-5 is a product of a committee, or, rather, a series of committees, and reads like it. To suss out what is wrong with a patient, we rely on the pattern recognition skills of physicians, which are not very different from the rest of medicine. However, I longed for a better way. Leo had been a neurologist prior to his training as a psychiatrist. He also lived in Brooklyn and liked to tell it to me straight: “The way you get good at psychiatry is by taking your interests outside the field and bringing them in, mixing them up, and that will become something new. For example, I’m reading this book about cognitive neuroscience. Dr. Pennebaker uses natural language processing, which hinges on pronouns to predict things about the writer. I bet I can do the same with schizophrenia.”
He wasn’t wrong. I couldn’t shake that the robots would be better diagnosticians than I could ever be someday. I haven’t stopped working on the problem since. It started with publications on the use of AI to monitor the experience of esketamine. In even lesser-known work, I was the PI on a study with the remarkable team at Mind Med to build detection systems for anxiety. The robots were here!
I went on to work with iRxReminder and Videra Health on our now commercial Tardive Dyskinesia Risk Assessment tool. Health professionals can go to tdscreen.ai — scroll to the bottom to submit a request for access to TDScreen. One publication on that work is out already, and more are coming.
Those tools solved the classic problem of “hot dog/not a hot dog” a la Silicon Valley. It’s easier to teach a Machine Learning Algorithm to differentiate between two things than to go through the much more challenging process of figuring out which of many things is the “match.” I had the privilege of meeting the team at Psyrin a few years back and was shocked by the insights this scrappy and, frankly, very young team was bringing to their work. It was mature thinking, serious science, and a relentless focus on a scalable diagnostic solution.
I have since become an advisor to the company, and in today’s column, I present an interview with Raheem Chaudhry (RC), a UK-trained physician and co-founder.
RC: I grew up in Cheltenham, which is in the Cotswolds, a beautiful scenic area. It's great if you're retired—a beautiful landscape. Horse racing is popular. Cheltenham is also renowned for its festivals. So, in addition to the Cheltenham festival, which is one of the biggest betting events in the UK, it also has a literature festival, science festival, and jazz festival. This is great for some, but for me, I ended up moving to London to follow my dreams of becoming a doctor in the big city.
I'm 26. Most people go to medical school straight out of sixth form, or, I guess you’d call it high school in the US— you would go straight through a six-year or five-year program from 18 years old.
OM: So you are a physician?
RC: I am a physician by background. I don't practice clinically now; I'm spending a hundred percent of my time on the Psyrin. To do that, I have to withdraw from the licensing board after moving to the U.S.
OM: I'm always interested in that moment when we're young; there are so many things we have yet to see. That is a moment when new experiences can be formative. Especially the things we enounter earlier in our clinical training are very impactful. I'm wondering if, while doing that training, something changed your mind about what you would do with your life. I'm curious what was it?
RC: How did I decide, “I'm going to abandon being a practicing physician for now, at least?"
OM: …and go on to do something not that.
RC: I didn't plan on going to medical school to start a startup. I really didn't. It was a serendipitous thing insofar as my university flatmate saw Julie, my co-founder, having this idea about using voice to get at diagnosis. We saw them present at a competition. She brought me along for some due diligence because my research area in medical school was non-imaging biomarkers in schizophrenia. So she wanted to see whether Julie's idea was any good!
OM: “Is this real?” was the question…
RC: Exactly. That was the question she was asking. So I went along, saw this presentation, met Julie and Ed, and, long story short, I ended up joining the startup, but my flatmate didn't.
OM: It wasn't a clinical experience that did it?
RC: The clinical experience is the reason why I went into the startup.
OM: I’ll describe my moment in a way that makes my question more clear. I remember on the second day of my first rotation of medical school— inpatient psychiatry— I met someone with catatonia for the first time, and it just struck me. I had seen suffering. I had been a research subject for J. John Mann! He gave me a lumbar puncture when I was 20. I had spent time in a schizophrenia research unit. I don't have schizophrenia, but that's where they had a place to put me for the bipolar study I was in. Still, when I saw catatonia? Holy cow, this person is sick. This person is suffering from a devastating illness—did you have a moment like that?
RC: I did. I think one patient that sticks with me. He was a guy my age. He was going to university like me. He was writing on the walls. He had been an A student. On the unit, when we met, he was tearing up pages from a book.
He started university. I think he was doing engineering. His life has just changed on a dime; he ended up being sectioned, and we still weren't able to control his psychosis. We had a team call with his family and saw how it affected them all. That was my first experience with the impact of psychosis. Because he was someone I felt was close to home, It struck a chord. That was my first patient experience in inpatient psychiatry, and I ended up staying in inpatient psychiatry for a whole year.
OM: And you've been pursuing that problem he has been facing ever since, it sounds like?
RC: Yeah, I think that's right. In my extended family, there's a long history of mental illness.
OM: My entire family is in the NIH family study!
RC: Oh, wow.
OM: As a kid. I had bipolar disorder. And so my mom did what any parent does. She calls the persons she trusts in her world.
RC: Right.
OM: She called Denny Zeitlin, a jazz piano player who happens to be a psychiatrist at UCSF, and asked, “What do I do for my kid?”
He said, “You get them into a study. Because they're not going to drop a study patient.”
I was enrolled in a study with Ramin Parsi, who's now the chair at Stony Brook, as the PI. Maria Oquendo was my study doctor. J. John Mann gave me a lumbar puncture. After that study, I got my entire family to enroll in the NIMH family study on genetics in bipolar disorders.
Thus, everyone in my family has a SCID1 rating. I ended up with this tremendous respect for research from Jump, and it sounds like you do, too. I'm curious: you were doing this research, and so you saw a problem in this person that was motivating, you spent a year on that unit, and nothing dissuaded you in that time, but eventually, there's going to be a point where you decide academics isn't how you do it, and a startup is. Take me from thinking you will be a physician or academic physician to this moment where you leave that aside.
RC: Thank you. That links to why we pivoted to the U.S.: trying to get through innovation or change is a prolonged and bureaucratic process in the NHS. It's a massive organization—it's bigger than anything else. If you do want to, there are ways you can make small changes.
You can put posters in this waiting room or that waiting room, instead of the left room instead of the right room. These little, small changes can be made locally, but to do something big, to make a massive change? That takes years and years. It's a safe system.
Changes go through many layers. There needs to be more space for innovation— things die. Healthcare Innovation was not supported until very recently. And when they finally launched the clinical entrepreneurs’ program, I think they found it challenging even then. People ended up thinking about other markets as well. It's frustrating because there are a huge number of bright people in the UK, especially doctors, but across academia, who want to launch these novel approaches. They want them to make an impact as soon as possible.
Our culture of academia is great for research and it's great for learning. However, to do, I think you need to think about startups. You need to think about your markets in terms of which patient population you can practically get this to. In which populations can you prove a tool’s usefulness? From there, it can be used as a launchpad to go into NHS or other large institutions with a slower process.
OM: If I hear you right, you saw the importance, on that inpatient unit, of identifying suffering to tackle it quickly—before it caused devastating changes in the lives of people who could have been…you. You also saw the impossibility of changing that in the NHS.
RC: It's tough to think about how on earth you go from going as a physician, from patient to patient, person by person; you've got a waiting list as long as ten football fields, and make any difference. Going from one person at a time and trying to put out fires as they come? It feels hopeless. You want to stop it at its root cause and help as many people as possible, all at the same time. I was drawn from a firefighting approach to thinking about things from above.
OM: You want to take a bird's eye view.
RC: You want to think about how can I help as many people as possible, as early as possible so that we're ameliorating the suffering before it even begins.
OM: Can you tell the story of meeting your co-founders for the first time?
RC: It feels like such a lifetime ago now. We met at this bar on the Thames is called the Oyster Shed. It's got a beautiful view. It's a pub bar element. It's got a glass wall and a beautiful view. There's a round table when we're Julia and Ed were sitting by that window, looking out over the Thames. We order some food to come to the table, and in that time, we start having our first conversation. We're not even talking about the company or Julie’s research. We're talking about mental health philosophy. We're talking about the issues with the DSM. We're talking about what we want to see: a long-term big vision for mental health. Before we even get to the nitty gritty of the detail. This was before they even brought the Oysters!
It was a conversation about treatment matching, not “just” diagnosis. We were all interested in precision in psychiatry, not just the one-size-fits-all treatment algorithms approach I was using as a physician. We dove heavily into the science of cognitive markers—imaging was all the rage at the time. Julie's work was on speech.
It's great if you have; you can get everyone an fMRI tomorrow as a screening tool. That would be one thing, but you can't do that, so understanding the clinical utility, the reality on the ground that resources are limited? This is why speech was important.
Even over the first meal on the ancient Thames, we drilled all the way down to how this works in reality. I knew then that this was for me. This is what I want to do. The details mattered!
OM: What does Psyrin’s technology do?
RC: The vision is a tool that can help clinicians predict, detect, and monitor mental health conditions.
OM: Explain it to my mom.
RC: We're trying to do is we're trying to catch those earliest signs of disorders. We started with psychosis, which includes disorders like schizophrenia and bipolar disorder. We've expanded our technology’s capabilities across multiple conditions, picking up the earliest signs of those disorders in speech.
OM: And so people talk into their phone?
RC: Patients log into the app for five minutes and talk to the AI in response to prompts. Next, your psychiatrist will get a readout of your percentage likelihood of different disorders. That's version one. Right now, that looks like the percentage likelihood of schizophrenia and percentage likelihood of bipolar disorder, but also differentials for clinical high risk. That might be autism, or it might be OCD.
OM: Is all of that explanatory power essentially hiding in people's everyday speech?
RC: Yes, it is. If you go to your psychiatrist, part of what they do will be assessing your speech, both what you're saying and how you're saying it. It's part of that process already baked in. We're making it more accurate by increasing the resolution. We're trying to get the earliest hints of disorders and automate them. Instead of a 90-minute structured interview, for example, you can have a five-minute interaction with an AI that can give similar data to your physician. It's all there in your speech, from the basic stuff that we learned in medical school, flattened affect or depression or thought disorder for schizophrenia, all the way down to things that are imperceptible to human beings, and only a computer can pick up on it.
OM: In everyday life, when you meet somebody, you hear them talk, and sometimes you get a sense that something's wrong. Is that sense that we get our brain doing pattern recognizing math?
RC: Yes, very much so.
OM: Doctors listen to a broader range of speech, which we listen to from our patients. We have some structure to learn that pattern recognition skill, but that only goes so far. For example, I heard about measuring blood glucose and hypertension from a speech last week!
RC: there's so much in there that we have been picking up on;
OM: We've been listening to the words, not the music, as they were.
RC: That's a good way of putting it. I've also heard people discuss “system one and two” thinking. Physicians have this sort of understanding - they don't know why they know, but they know when they hear something is off. You can speak to someone on the phone and tell if they're stressed or worried; you can tell that something's not right if they're ill. You can tell a lot from someone's speech, and you only need to pick up a phone to get that information.
Thanks for joining us for Part One of my interview with Raheem from Psyrin.
Structured Clinical Interview for DSM-IV, at the time, was the gold standard of diagnosis in psychiatric research.
Won't this tech be given to GPs allowing us to bypass BH altogether?
If a GP passively uses this tech during a consult can they enter a diagnosis based on the AI assessment straightaway? Does the patient have any autonomy or Informed consent in this scenario?