This story is from The Pulse, a weekly health and science podcast.
Julie Bates-Maves was only a couple years out of school when she was working as a counselor at a methadone clinic. She had 89 patients, a pretty crazy caseload that happens to be par for the course in her field. Years later, though, there’s one patient that she still thinks about every single day.
“I remember his face vividly,” she said. “I remember his typical style of dress, and he had a particular coat that kind of looked like a letterman’s jacket, and it was this big, bulky leather coat, it was red and white and black.”
She calls him Michael to protect his privacy, and she remembers that she liked him.
“He was just such a lovely person, and thinking back, he was very sarcastic in his humor and so am I, so we just kind of enjoyed being with each other,” she said. “He was very vulnerable and honest about what he was experiencing.”
When he first came to Bates-Maves, Michael was in pretty terrible shape. He had been working construction when he fell from the third story of a job site, breaking his back and neck. He got hooked on his narcotic pain meds, and later moved on to crack. All the while, the doctors never could get rid of the pain.
“It was something that was going to be chronic and persistent and largely managed forever,” Bates-Maves said. “So he came in with depression on top of his addictions, and it was all one big glob of a problem.”
In the early days, it was rough going. Michael talked about hopelessness, and he talked about killing himself.
“He had a lot of the so-called traditional risk factors that we would be taught to assess for,” Bates-Maves said.
But Michael kept coming to every 6:30 a.m. appointment, every Monday. They helped, and his life outside the clinic was turning around in big ways.
“He was engaged and looking forward to many things: marriage, just becoming a father,” Bates-Mave said. “He was on the verge of having a workers’ compensation claim paid out that would have exceeded a million dollars.”
Over nine months of treatment, it seemed to her that Michael’s life was getting better, and that Michael was getting better. It had been months since he’d mentioned hurting himself when a Monday morning came around and Michael didn’t show.
“I just thought he overslept because it was six in the morning,” Bates-Maves said. “So I called and left him a voicemail saying, ‘Hey, you know, I’m sorry [we] missed each other, call me back when you can.’”
Michael never called back.
“About 9 a.m., when I had a break and I checked my voicemail, I had a call from his fiancée saying, ‘Michael’s no longer with us, I’m sorry.’ And that was it,” Bates-Maves said.
Later, she learned that around midnight, only six hours before their appointment, Michael had killed himself.
One thing Bates-Maves repeated often during an interview is the idea that a client is not a friend. They can’t be. It would cloud clinical judgment. But that doesn’t mean you don’t care for them.
“I think in some ways the relationships are a step up from friendship,” she said. “What I mean is, if I think about my friends, there’s a handful of them where I know, like, their deep dark secrets, and I know the reality of their insides. And then many others, I don’t, I know what they show me.”
It’s a kind of intimacy that’s hard to understand if you haven’t worked in the field. Bates-Maves had that with Michael, and it made his death devastating. Wrapped around her grief, always, were the questions she asked herself: “What went wrong? What did I miss? Oh my God, is this my fault?”
Bates-Maves put herself on trial in her head, producing evidence for and against. Combing through every word she said, everything Michael said. It was a constant examination and cross-examination.
“It’s just this really awful teeter-totter between tremendous guilt and then, kind of counter-thoughts to that of ‘It’s not your fault,’” she said. “But you know, we search for explanations when things get really hard, and the easiest explanation at the time was I missed something, you know? So every now and again, I would land on that, and that would hit me like a truck.”
Bates-Maves failed to predict Michael’s suicide. That much was indisputable.
The question that haunted her was: Should she have been able to?
Suicidal or not? A coin toss
Was there something in her counselor handbook, in everything she learned in school and in clinical rotations, that she could have used but didn’t?
Psychologist Jessica Ribeiro of Florida State University has explored suicide risk factors. She and colleagues analyzed hundreds of longitudinal studies that represented 50 years of research. These are the types of studies that follow people for years, even decades. The researchers could see who ended up attempting a suicide or having suicidal thoughts and who didn’t, and whether having risk factors made suicidality more likely.
It turns out, they didn’t.
“Our ability to predict suicide was about on par with a coin flip,” said Ribeiro.
Using commonly held suicide risk factors to assess people — things like substance abuse, hopelessness, depression — was as good as flipping a coin, pure chance. Suicide science is pretty new, but still decades of study, of trying to figure out what makes someone suicidal, and we’re no better at predicting that than we are at predicting heads or tails.
“And even predictors that we thought would be particularly strong, that are often cited as really strong predictors, or, for instance, something like having a prior suicide attempt — that’s often cited as by far the strongest predictor of suicide death — even something like that wasn’t much better than chance prediction,” Ribeiro said.
The most surprising risk factor that didn’t actually pan out was “clinician prediction” — even a trained professional like Bates-Maves, saying a person was suicidal or not. It was still just a coin toss.
“So suicide, much like everything else, like biology for instance, is likely to be a cause of many things coming together all at once in a way that sort of defies us understanding,” Ribeiro said.
The basic mystery of suicide remains what leads to it. Ribeiro and her team’s work is only a slice of a growing body of research that basically says, “We don’t know.”
Part of the reason is that suicide is incredibly difficult to study. You’re always in a position where you’re trying to puzzle together what happened, retrace somebody’s steps after the fact.
You can’t experimentally induce hopelessness in a group of people and see if that causes suicide. You can’t randomly assign depression and watch and wait.
“In psychology, our strongest interventions are based on targeting those necessary causes of a phenomenon, and disrupting those causes,” Ribeiro said.
As an example, exposure therapy for people with phobias or traumatic stress disorders works because psychologists know the fear and stress responses are learned, and then reinforced by avoidance. They know those responses can be unlearned by exposure to the distressing things or ideas in a safe setting.
Suicide is different.
“Because we don’t know what the causes of suicide are, our interventions have largely been kind of flying a little blind,” Ribeiro said.
So knowing just how little is known about suicide, how its causes may be too complex for humans to grasp, the research team’s members looked to machines, and to machine learning.
They had access to the electronic health records of a few thousand people from Tennessee who had attempted suicide at one point. They used that data to train their algorithms to look for suicidality.
It was like giving a bloodhound a whiff of a fugitive’s clothing?
Basically, they looked at every characteristic those suicidal people had: everything they had ever been diagnosed with (medical history or mental illness), their medications, demographics and socioeconomics, a number of factors beyond traditional suicide-related issues.
Then, the researchers broadened things out. They asked the algorithms to look at the records of some 2 million Tennesseans and guess who among them had attempted suicide.
And the computer was much, much better at it than a coin toss. It took all those factors, looked for and found patterns, and predicted correctly — suicide or not — up to 90% of the time.
Though the researchers know the machine was accurate, they don’t really know why.
“Suicide is kind of like a painting. So, asking questions like, ‘What in those particular algorithms comes out as particularly important?’ is kind of like asking what brushstroke is the most important in [a] painting,” Ribeiro said. “So what these algorithms can do is sort of capture more of that painting. How a lot of these factors come together to create something that emerges to become suicide.”
Ribeiro’s team is working on developing a system that uses health records to assign a kind of score, just like blood pressure or blood sugar, that doctors can track. The thing about their machine is it can’t predict when a suicide will happen — tomorrow or next year — but it appears to accurately predict risk.
Just like a cholesterol number can say a person needs to be on medication, the suicide risk number could say someone needs a mental health intervention of some kind.
Experiments only done in virtual reality
To get a better understanding of it all, the research team is designing the kind of experiments you can only do in virtual reality.
“The virtual reality paradigms can be particularly evocative for folks, and when I describe them, it’s sort of like, ‘Oh God, that seems, dangerous and whatnot,’” Ribeiro said.
There’s a gun scenario, and a tall structure jump scenario. It looks like a video game, but it feels real.
“It’s incredibly vivid. I still have graduate students in my lab who will not jump, even though they know it’s safe and they’re in the lab room on solid ground and have a ton of exposure with it,” Ribeiro said.
It’s not real, but it can feel real. They subject participants to a factor they’re trying to test, say, social rejection (a program excludes participants from a game they think they’re playing with humans), or anxiety (participants write speeches on short notice, and then are critiqued harshly), and then see if the participants feel compelled to act out a suicide in the virtual reality setting.
None of this will decode suicide all at once, but the idea is you have to start somewhere. The team hopes that by by subjecting participants to stressors that can be recreated in the lab, they’ll see whether something pulls the rate of virtual suicide one way or another, even if ever so slightly.
They’re chipping away, experimentally, at the causes of suicide, basic science that hasn’t really been done all that much, and it’s slow-going.
“I think we’ve been trying to build things because of how urgent this problem is. We’ve been trying to build something that we hope will be helpful,” Ribeiro said. “But I think what the data is showing is that we really need to take a step back in the interim and really understand how this works and then build from there.”
Until that happens, counselors will still face patients in crisis, and they’ll have to do the best they can with what may be an incomplete set of tools.
Counselors like Bates-Maves, who lost a patient early in her career, said that after many years on the job a kind of intuition starts playing a bigger role in their assessments.
“What I’ve learned from that experience … certainly from Michael’s death, being a big part of that, is that pain is not often what it seems, and people don’t have to appear sad to be hopeless and people don’t have to be depressed on the outside to be suicidal,” Bates-Maves said.
She looks for the painting of suicide, not for each brushstroke.
“It’s more like, ‘Hey, you know, here’s what I’m noticing, and I can tell that you’re really hurting, and I can’t exactly explain to you what I’m picking up on, but I know there’s something there and I’m hoping we can talk about it,’” she said.
With Michael, Bates-Maves eventually realized part of the reason she didn’t see the risk is because Michael didn’t want her to.
“Something that I’ve really settled on is that when he came out of his depression he regained a significant amount of energy that he lacked for a long time. Sometimes, you will find that to be true, that people have the energy to act now, where when they’re in a depressive state, they don’t always have the energy to make happen what they want to happen,” she said. “And so, that day we met, the week before, he could have been actively planning his death and just never alluded to it or didn’t want me to know.”
When Michael died, Bates-Maves didn’t know a single other counselor who had ever lost anyone, and it’s not something that was really talked about in school. It’s part of what made the weeks and months after he died so torturous.
“We don’t want to sit with the fact that anybody could die this way any day,” she said. “It’s so pervasive, and yet we pretend it isn’t because it’s so scary.”
Bates-Maves is a professor now, training future counselors. She tells each student about Michael. That way, if any of them loses a patient, and almost certainly some of them will, they will at least know one other counselor who’s been through it as well.
They’ll know her.
If you or someone you know is in crisis, call the National Suicide Prevention Lifeline at 1-800-273-8255.