One size fits few. Precision psychiatry: where we are, how far we've come, and where we go next.

A 30-minute read, with eight interactive tools · Reid Robison, MD

In oncology, a biopsy can tell a doctor which mutation is driving a tumor and which drug will target it. In psychiatry, we still mostly choose a medicine, wait six to eight weeks, and see. Precision psychiatry is the effort to change that: to match the right treatment to the right person at the right time, using biology and behavior rather than symptoms alone. This post covers how far that effort has come, where it has stalled, and what two new roadmaps say should happen next.

One heartbeat, a slide rule, and a BlackBerry

One of my most memorable and influential mentors was Homer Warner, widely considered one of the founders of medical informatics. I did a postdoctoral fellowship in biomedical informatics at the University of Utah, in the program he founded¹². Early in his career, he once stayed up all night analyzing a single heartbeat with a slide rule. That night led to a paper on analyzing cardiac waveforms with computers, in the 1950s. LDS Hospital in Salt Lake City (today part of Intermountain Health), where he worked, took a smart gamble and bought him a room-sized computer in 1960 for hundreds of thousands of dollars. A new field was born. By the early 1970s, Dr. Warner and his colleagues had taught computers to analyze an EKG and interpret lab results, and before I was born, doctors at that hospital were entering prescriptions into computer terminals.

I remember sitting with him one day in the medical school library, right next to a larger-than-life statue of him. In the statue he is holding a handheld device. I smiled, because in real life, sitting beside it, he was holding one too: a BlackBerry with a 240×240-pixel color screen and a whopping 16 MB of storage. He was comparing it with that first giant computer, and how the little thing in his hand was now hundreds of times faster. My mind was blown. Then he told me: the most sophisticated healthcare measurement device ever created is now the one in my hand.

Cardiology took that idea and ran with it. Psychiatry is only now starting to.

Why psychiatry needs precision

“Is relating the initial cellular effects of a drug to behavioral benefit in heterogeneous patient groups of syndromes defined by experts and not by biology really such a difficult task?”Stephen M. Stahl, MD, PhD, with tongue in cheek

Since DSM-III in 1980, psychiatric diagnoses have been defined by symptom checklists. That brought reliability: two clinicians can usually agree on who has major depression. But it never promised biological validity. Two people can share a depression diagnosis with almost no symptoms in common, and the same biological problem (disrupted reward circuitry, for example) can show up across depression, addiction and psychosis¹. In 2010 the US National Institute of Mental Health launched the Research Domain Criteria to study mental illness by dimensions like threat, reward and cognition that cut across diagnoses¹³. Each diagnosis we name may turn out to be many diseases, perhaps hundreds.

The psychiatrist Daniel Barron makes the problem concrete with a comparison I’ve borrowed for talks ever since¹⁴. Picture a patient with chest pain, a vague symptom that could be heartburn, a panic attack or a heart attack. A doctor asks about it, of course, but doesn’t stop at conversation. A calibrated monitor counts the heart rate. An ECG records the heart’s electrical activity millisecond by millisecond. A blood test looks for the protein a damaged heart releases. Within minutes there is a stack of objective, recorded measurements. Now picture a psychiatric evaluation. I listen for rapid or disorganized speech, watch facial expressions, notice recurring ideas, and ask finely honed questions. It’s a trove of clinical data, and the only instrument gathering and analyzing it is my own brain. Most of it is never recorded at all.

The cost of not knowing who will respond to what is easiest to see in depression treatment:

How far we've come

Precision psychiatry isn't only a future idea. A few tools are already in practice, and the research has moved through some instructive hype cycles. Brain-based measures have sorted psychosis into biotypes that cut across diagnoses¹⁶, and depression and anxiety into six circuit biotypes¹⁷. An EEG signature predicted antidepressant response¹⁸. Automated analysis of speech predicted which high-risk young people would develop psychosis¹⁹. Gene-guided prescribing improved response and remission in one large trial²⁰, and inflammation has emerged as a possible way to pick out who might benefit from anti-inflammatory treatment²¹.

What's ready, what's close, what's research

Here is my read of where each major approach stands. The categories follow the biomarker types used by the FDA and NIH: diagnostic, predictive, prognostic, monitoring, safety, and others¹¹.

The cheapest precision tool is a questionnaire

Before any scan or gene test, there is measurement-based care: a validated scale, like the PHQ-9 for depression or the GAD-7 for anxiety, completed at every visit and used to decide whether to stay the course or change it. In a randomized trial with blinded raters, depression care guided this way led to remission in about 74% of patients, compared with 29% in standard care⁶. Yet surveys keep finding that fewer than one in five behavioral health clinicians use it routinely^22,7. That gap is a familiar one: it has been estimated that it takes 17 years for just 14% of research findings to reach everyday patient care²³.

I’ve tried to practice what I preach here. At our ketamine and esketamine clinics in Utah, patients completed standard scales at each treatment, which turned routine care into evidence. Led by my graduate student at the time, our team analyzed 171 people with treatment-resistant depression who received esketamine nasal spray between 2019 and 2021. Over a median of 11 treatments, average depression scores (PHQ-9) fell from 16.7 to 12.0 and anxiety scores (GAD-7) from 12.0 to 8.7, and scores for suicidal thinking dropped too. Dissociation was common (73%) but nearly always resolved within two hours, and there was one serious adverse event²⁴. We then used those real-world numbers to compare value. From the health system’s perspective, IV ketamine was the more cost-effective choice; from the patient’s side, effectiveness was similar, and insurance coverage and assistance programs made esketamine the cheaper option out of pocket²⁵. A retrospective study can’t replace a randomized trial, but none of it would exist if we hadn’t measured.

The test you're most likely to get: pharmacogenomics, decoded

If you've had any "precision" test in psychiatry, it was probably a pharmacogenomic (PGx) panel: a cheek swab, sent to one of several commercial labs, that comes back as a colorful report sorting medicines into green, yellow and red. Insurance increasingly covers it, and it can be useful. It is also one of the most misunderstood tests in medicine. Here's what it does, what it doesn't, and how to read one.

How a PGx report gets made

What the genes actually do

Most of what a PGx test measures is speed. Enzymes in your liver, especially CYP2D6 and CYP2C19, break down many antidepressants and antipsychotics. Your genes decide how many working copies of each enzyme you have. That makes you a poor, intermediate, normal, rapid or ultrarapid metabolizer of drugs that rely on that enzyme. Poor metabolizers clear the drug slowly, so the same dose builds up to higher blood levels and more side effects. Ultrarapid metabolizers clear it fast, so the drug may never reach a level that works⁴. These types are common: roughly 2% of people of European ancestry and about 12% of people of East Asian ancestry are CYP2C19 poor metabolizers, and frequencies of every type vary widely around the world²⁹.

Two things in that tool matter a lot in real life. First, a dose change often fixes a gene–drug mismatch; switching medicines isn't the only option. Second, other medicines can override your genes. Fluoxetine, paroxetine and bupropion strongly block CYP2D6, so a genetically normal metabolizer taking one of them can behave like a poor metabolizer. This is called phenoconversion, and a gene test can't see it³⁰.

Decode a sample report

Here's how the same kind of report reads for three different people. These are illustrative reports, not any company's format. Tap any medicine to see what its color really means.

So why did Zoloft land in red, and Pristiq in green?

Sertraline (Zoloft) is cleared partly by CYP2C19. For a CYP2C19 poor metabolizer, the guideline from the Clinical Pharmacogenetics Implementation Consortium (CPIC) suggests starting at a lower dose and titrating more slowly, or choosing a drug not mainly cleared by that enzyme⁴. Some vendors label that yellow; others label it red. Desvenlafaxine (Pristiq) is cleared mostly by a different process (conjugation), with little help from the enzymes these tests measure³¹. So there's almost nothing for a gene test to flag, and it shows up green for nearly everyone. Green means “no known gene interaction,” not “more likely to work for you.” CPIC makes no genetic recommendations for desvenlafaxine at all⁴.

And prescribing sertraline without a test wasn't an error. No guideline requires PGx testing before starting an antidepressant; CPIC guidelines explain how to use results once you have them⁴. Most people are normal or intermediate metabolizers and do fine on standard dosing, and sertraline remains one of the most widely used and best-tolerated antidepressants. Even for a poor metabolizer, the usual fix is a lower dose, which is exactly what careful prescribers do when side effects show up, test or no test.

What's on the panel, and how much each gene can tell you

Not all genes on a PGx report are equal. Enzyme genes that change drug levels have strong evidence and formal guidelines. Genes said to predict whether a drug will work (often serotonin-receptor or transporter genes, or MTHFR) do not. CPIC reviewed SLC6A4 and HTR2A for antidepressants and found the evidence too mixed to recommend any action⁴. In 2018 the FDA warned that the link between DNA variants and antidepressant effectiveness has never been established⁸.

Who benefits most

• People who've had side effects at ordinary doses or no benefit at full doses, where a metabolizer type can explain what happened.
• Before starting a drug with strong gene–drug guidance: carbamazepine (HLA screening), tricyclic antidepressants, citalopram/escitalopram, paroxetine, vortioxetine, and several antipsychotics with genotype dosing in their FDA labels^32,33.
• People taking many medicines, where gene results plus drug interactions together shape the safest plan.

Used this way, PGx is a genuine precision tool, and the trials back a modest benefit^27,20,34. Used as a crystal ball for which drug will work, it disappoints. Two practical tips: different labs can give different colors for the same genes²⁸, so look at the underlying metabolizer types, not just the colors; and keep your report, because your genes won't change.

The hard lessons

Both roadmaps put overfitting front and center. A biomarker has to be validated analytically (it measures what it claims), internally (it works in the development sample) and externally (it works in new people, at new sites). Prediction models need locked analysis plans, external validation and prospective testing in both biomarker-positive and biomarker-negative people^1,11.

What the roadmaps say comes next

In 2025, an international group from academia, industry and the European Medicines Agency published a Precision Psychiatry Roadmap in Molecular Psychiatry. It argues for moving from a fixed, symptom-based classification to "a continuously flexible and iteratively evolving" biology-informed framework, built in three stages¹:

1. Align globally (years 1–3): agree on principles, harmonize data and measures, and build shared platforms.
2. Build consensus on predictive validity (years 3–7): validate biomarkers and run intervention trials in biologically defined groups.
3. Put it into practice (7+ years): fold validated biology into classification and clinical care.

The model it holds up is Alzheimer's disease, where amyloid and tau biomarkers now detect pathology years before symptoms and made targeted antibody treatments possible. It also insists that biology isn't destiny: the "exposome" of trauma, stress, poverty, pollution, inflammation and the microbiome interacts with genes, and about 75% of mental disorders begin before age 25¹.

In 2026, the American College of Neuropsychopharmacology followed with a practical roadmap for precision clinical trials¹¹. Its key points:

• Define the context of use. Every biomarker needs a precise statement of what decision it informs: enrichment, prediction, monitoring, or a surrogate endpoint.
• Expect multimodal biomarkers. No single gene, scan or blood test is likely to be enough; combinations of genetic, molecular, neural, digital and environmental measures probably will be.
• Use the regulatory paths that exist. Biomarkers can be used inside a drug program without formal FDA qualification. The first psychiatric biomarkers to enter the FDA's Biomarker Qualification Program, an EEG face-processing measure and an eye-tracking measure of social attention from the Autism Biomarkers Consortium for Clinical Trials, show it can be done.
• Collaborate before competing. Large consortia like AMP Schizophrenia (about 2,000 high-risk young people at 40+ sites) are building the shared, standardized data the field needs.
• Be honest about genetics. Polygenic risk scores remain only modestly predictive for individuals.

The visit we should be building now

None of this needs to wait a decade. Some of the pieces exist today, and the rest are being built and tested right now. The job for the field is to pull them into everyday care on purpose, rather than waiting for them to arrive.

What you can ask for today

• Pharmacogenomic testing, with the right expectations. It's most useful after side effects or failed trials, or before drugs with strong gene–drug guidance. It helps with choosing and dosing; it can't tell you which medicine will work best. See the decoder above^4,34.
• HLA screening before carbamazepine or oxcarbazepine if you have Asian ancestry. This one is standard of care³.
• Measurement-based care. Ask your clinician to track your symptoms with a validated scale at every visit and to show you the trend. It's the simplest precision tool there is, the evidence for it is strong, and many of the scales are on this site⁶.
• Advanced options after two or more failed medicines: esketamine, ECT, or accelerated, brain-mapped TMS where available⁵.
• Clinical trials. More and more trials stratify participants by biology. Joining one is a way to get access to new treatments and to help build the field.

Where I land

Early in my career I worked on the genetics of autism in Utah families, and later I co-founded a genomics company, Tute Genomics, betting that sequencing would quickly transform how we choose treatments. On September 13, 2016, we launched a Kickstarter campaign offering people whole-genome or exome sequencing with a report on their actionable variants and disease risks. It caught on fast: $74,560 toward a $100,000 goal from 114 backers. Two days after launch, a letter arrived from the FDA about marketing genome sequencing directly to consumers. In plain terms: shut it down. It didn't land well with me, but we had no real choice. We suspended the campaign, refocused our technology on genetics labs doing hospital-based testing, and kept working to move the field forward³⁶. What I wrote to our backers then is still how I feel: “I still strongly believe that everyone should have a right to, and own their DNA, but they should also respect the FDA and their credo to protect the consumer. I think both aims can be achieved.”

It has transformed a lot, but in psychiatry the road has been longer than any of us hoped. The brain's complexity, small studies, and our symptom-based categories have all slowed things down. What has changed is that the field now agrees on how to get there: shared data, rigorous validation, and trials designed around biology from the start. As a trial investigator, I see that last part as the real turning point. For now, the most precise thing most of us can do is measure carefully, choose thoughtfully, and stop guessing sooner.

I’ve also learned some humility about timelines. In a 2021 keynote to investors, I described a coming era of curative psychiatry. Three years later, the FDA declined to approve MDMA-assisted therapy for PTSD³⁷. Progress is real, but it rarely arrives on the schedule we announce. Dr. Warner’s slide rule took decades to become a BlackBerry, and the BlackBerry took only a few more years to become the phone in your pocket, which can already sense sleep, movement and voice. The measurements are coming. Our job is to validate them, protect people’s privacy, and use them well.

Data moves science. Story moves culture. Precision psychiatry will need both.

Educational only. This isn't medical advice and isn't a substitute for care.