Cecilia (not the patient’s real name) was 15 the first time she tried to kill herself. She sliced into her left wrist with a razor she had hidden away. The initial sting silenced her emotions, but as she went deeper her arm tensed. Her head dizzied with pain. Too much. She screamed out and threw the razor at the wall.

Cecilia’s built-in self-destruct safeguard—pain—had worked. Her parents brought her to the hospital, where she remained for two months.

Six years later, when I met her in the emergency room, she told me she was thinking about driving her car into a tree. Or perhaps jumping off a building—something between seven and nine stories; “High enough to do the trick, but not so high I’d splatter,” she told me.

Six years, nine medication trials and months of psychotherapy later, Cecilia still suffered from what her doctors labeled depression and borderline personality disorder.

Curled in a ball on a gurney in the hallway, she told me her story, hands pressed to her tear-filled eyes. She had fallen into the wrong crowd and had recently been raped. I noticed the 20 or so parallel scars that ran from her left elbow to the base of her wrist, notches for each time her emotions overpowered her reason.

I wondered what her reality felt like, where her only reprieve involved mutilating herself. What on earth was going on in her brain?

I had seen patients with depression. This was different. The texture was different. And I couldn’t say quite why.

So I began to look critically at the language I was using to describe her. To call Cecilia’s mood “sad” was an understatement. “Emotion dysregulation” is a more clinical term meant to capture Cecilia’s fluctuations in mood and how she’s often at the mercy of those mood swings. But this still failed to capture what was going on: What facet of emotion? How dysregulated? Is it actually dysregulated emotion or is it another back-end process, perhaps a more cognitive function like attention or memory?

Cleaving the ambiguous term “emotion” into its component processes provides greater precision and, therefore, ability to act. Drilling into “emotion” with language such as “negative and positive valence”—which captures different poles of emotional stimuli—or other biologically driven language gets closer to what Cecilia’s brain is doing: computing information.

Over the last decade research in Catherine Harmer’s Psychopharmacology and Emotion Research Laboratory at the University of Oxford has shown that increasing the brain’s serotonin levels affects how the brain processes information like pictures of fearful faces (negative valence) and happy faces (positive valence). Harmer’s studies have shown that the brain’s response to negative valence is decreased whereas the brain’s response to positive valence is increased. Because depression is thought to be related to a negative emotional bias, modifying bias from negative to positive is thought help improve mood.

Harmer’s work has shown that selective serotonin reuptake inhibitors (SSRIs) like Celexa (citalopram) decrease “negative bias” with a single dose, suggesting that neurotransmitter levels directly affect this facet of emotion. Harmer’s lab has also shown how SSRIs affect valence in different ways by altering how information is processed in specific brain networks. These observations led Harmer to propose a model for how antidepressants work: They don’t directly affect mood by making you “less depressed” but rather change the way your brain processes emotional information and, by changing this negative processing bias, you learn to be less depressed over the space of weeks because the world doesn’t seem as bad.

I worked with Harmer’s lab for six months and enjoyed their computational framework for mental illness. Essentially, the framework suggested a clinical algorithm: If you want to change the brain’s emotional bias, use an SSRI. But could I have sorted out whether Cecilia needed a change in emotion as opposed to, say, attention?

And because other antidepressants affect  other different neurotransmitters—serotonin, norepinephrine, dopamine—how could I sort out which of Cecilia’s neurotransmitters was at fault and which antidepressant to give?

This is a general problem that psychiatry is beginning to address. Because our broad drug categories like “antidepressants” and “antipsychotics” act on a variety of physiological processes, this adds another layer of imprecision to our treatment plans. Far better to call an “antidepressant” a “selective serotonin reuptake inhibitor” or an “antipsychotic” a “dopamine blocker,” because it describes a more actionable target.

As I typed up Cecilia’s treatment plan, I wanted to learn what exactly was wrong in her brain so I could find a medicine to target the problem.

Yet because there is equal imprecision in diagnosis as there is in treatment, I was in a bind. I didn’t have tools to think clearly and precisely about Cecilia, so I couldn’t find a clear solution (if one existed).

Even without invoking hard neuroscience, as I have before, I think there are two things we can do now to work toward biomarker-level precision in mental health care.

First, we could use biologically driven language in our formulations of mental illness. The language we use sculpts how we consider problems and, therefore, the types of solutions we develop for those problems. Words such as “thought” and “mood” rather sloppily ignore underlying brain processes and leave us linguistically unarmed as we approach the organ of mind.

Parsing emotion into valence—or, say, attention into orienting, alerting and executive processes—allows us to enter the realm of intervention, where facets of brain function have been and can be behaviorally isolated, measured and tracked with treatment.

And this is a second step we should take to biomark  psychiatry. We should actually measure brain function by quantifying behavior. Questions such as “How’s your mood?” should only be screening tools in the same way that cardiologists ask, “Do you have chest pain?” Such questions are not a diagnostic terminus but rather a guide toward more precise, quantitative tests.

There is an abundance of biologically informed measures of emotional responsivity, of attention, of working memory, etcetera. Many of these behavioral measures have been shown to activate discrete brain networks that respond to specific therapies.

Of course, no single behavioral measure will fully capture “emotion” or “attention”—and it will certainly not fully reflect some symptom-based diagnostic category. This is not proof that the test is invalid. It rather suggests that multiple measures are needed to evaluate a patient and that we have much progress to make in formulating mental illness.

To be honest, I am unsure precisely how to go about this but I’m interested in learning more precise ways to consider my patients and their illnesses. So if you have ideas, please e-mail! Maybe next time I have a Cecilia in clinic, I can combine behavior and biology in a more productive way.