Every few weeks a patient walks in holding a GeneSight or Genomind report, or a clinician tells me they ordered one to find out which medication will work best. The reports rarely say outright "this tells you which antidepressant will work," but that is the message everyone walks away with. The color-coded columns, the "personalized medication guidance" language, the whole way these tests are sold and talked about points to one conclusion: swab the cheek and the genes will tell you what to prescribe and what to avoid.
That is the promise. Here is what is actually true.
The bottom line up front: psychiatric PGx has two well-supported uses and one major limitation.
That is the whole post. The rest is just showing my work.
Short answer: No. This is the part that is not supported.
Several large, careful trials asked exactly this question. Does using the gene test to pick the drug get patients better than a clinician choosing normally?
What the guidelines say. The American Psychiatric Association's review concluded the evidence does not support currently available combinatorial PGx tools for selecting depression treatment. The 2022 VA/DoD guideline stopped just short of that and found insufficient evidence to recommend either for or against testing to guide antidepressant selection. Neither one endorses routine testing as a way to identify the antidepressant most likely to work. The VA/DoD group also noted that in the studies it reviewed, testing was actionable in only about 15 to 20 percent of patients, meaning their prescribed medication had a predicted gene-drug interaction, and that much of the evidence was funded by the companies selling the tests.
What I do. I do not order it to pick a drug. And I do not let a report talk me out of a medication that is working, or into one that is not.
Short answer: Yes. This part is true and useful.
Some people break certain drugs down fast, some slow. That is genetic, it is well established, and there are real published dosing guidelines built on it. A fast metabolizer may burn through a standard dose and never reach a therapeutic level. A slow metabolizer may pile it up and get side effects at a normal dose. That is the part of the report I actually read.
The catch: knowing how someone processes a drug is not the same as knowing whether it will treat their depression. And honestly, in most patients, careful dosing (start low, go slow, watch how they respond) gets you most of this same information without a swab. I am not claiming the clinical outcome equals a test. I am saying good clinical reasoning already covers a lot of this ground.
What I do. When someone has failed a bunch of trials with tolerability that makes no sense, side effects at a tiny dose or nothing at a big one, metabolizer status can explain it and change how I dose. That is a good reason to order it.
Short answer: Yes for a few specific drugs, and this one can genuinely save a life. But in everyday psych practice, it rarely comes up.
There are a handful of gene-drug combinations where a variant predicts a serious, sometimes fatal reaction:
These are not predictions that someone might feel nauseated or tired. They identify an elevated risk of potentially life-threatening reactions, and they can meaningfully change whether the medication should be used at all.
But here is the practical catch: in routine outpatient psychiatry, many of us rarely reach for carbamazepine or oxcarbazepine. That means one of the strongest, best-validated psychiatric uses of PGx testing is genuinely important when it applies, but it applies to only a small slice of our everyday prescribing. Even the part of PGx with the hardest science behind it does not come into play most days.
And notice how narrow it is regardless. It is a targeted safety check for a couple of specific drugs, not the broad "here is how you will tolerate any psych med" promise the panels imply. Everyday side effects like nausea or sexual side effects? The panels do not reliably predict those.
What I do. In the rare case I am heading toward carbamazepine in an at-risk patient, I screen. Outside of that, this particular strength of the test almost never changes my day.
Order it for the two things it actually does:
Do not order it for the reason it is usually sold. It will not tell you which antidepressant will work for the person in front of you. It gives you a couple of useful data points and flags a few specific dangers, real things, but it does not make the call.
Because here is what all of it comes down to: there is currently no way to get around clinical judgment. The test does not replace it, shortcut it, or outperform it. At best it feeds one more input into the same reasoning you are already doing: the history, the prior trials, the response in front of you, the dosing, the follow-up. A color-coded report can inform that judgment. It cannot be that judgment. Anyone selling it as a way to skip the hard part of prescribing is selling something the evidence does not support.
Order it for what it is proven to do, and keep making the treatment call yourself.