Few things in life work out the first time you try them, and antidepressants are no exception. According to the National Institute of Mental Health, it’s not uncommon for patients to try at least two such medications before finding an effective one. But artificial intelligence is on its way to solving that problem.

Researchers at George Mason University in Fairfax, Va., have revamped MeAgainMeds.com, their free tool that uses AI to recommend antidepressants to patients based on their demographics and medical history. Farrokh Alemi, PhD, a professor of health informatics at GMU’s College of Public Health, spearheaded the effort.

“Me Again Meds, it’s a play on the fact that many people who take antidepressants feel that they are not themselves,” Alemi tells Fortune. “We want to help them with a selection of an antidepressant that has fewer side effects for them and is more effective for them.”

The pursuit is personal. After losing a loved one to suicide, Alemi has in recent years dedicated the bulk of his research to AI in depression management.

Alemi and his colleagues have published several studies in conjunction with the development of Me Again Meds. In research published in 2021 in the journal eClinicalMedicine, they used the OptumLabs health insurance database to analyze nearly 3.7 million U.S. patients who had been diagnosed with major depression and were taking antidepressants. From 2001–2018, patients collectively recorded more than 10.2 million treatment episodes, or courses of medication.

Researchers assessed patients taking 15 of the most commonly prescribed antidepressants—including citalopram (Celexa), escitalopram (Lexapro), fluoxetine (Prozac), and sertraline (Zoloft)—and found vast differences in how the medications benefitted distinct groups of people. For instance, 25% of teenage boys treated with fluoxetine experienced symptom remission, while 59% of women ages 65–79 saw symptom remission on the same medication.

No medication was best for everyone and within the age/sex subgroups, the best antidepressant was on average over 20 times more effective than the worst. Alemi’s team showed that if clinicians had prescribed the medications with the highest remission rates, 1.5 times more patients, or 1.6 million more treatment episodes, would have had symptom remission.

“People are going through three or four trials before they get the right medication. Many don’t even get the right medication,” Alemi says. “African Americans are not given the right medication, Hispanics are given the wrong medication; all kinds of minority differences are ignored. All kinds of medical history information is ignored.”

An engineer by trade, Alemi expects systems to function well. Despite clinicians’ best intentions, the U.S. health care system’s practice of prescribing antidepressants “reminds me of 18th-century medicine,” he says. “Why are we not getting the right medication the first time around?”

Can AI keep up with demand for antidepressants?

In the same 2021 study, Alemi’s team went beyond age and biological sex to match patients to the most effective antidepressants. They incorporated study participants’ medical histories to generate nearly 17,000 patient subgroups. Not expecting doctors and patients to sift through so many options, researchers turned to AI, delivering the first iteration of Me Again Meds.

For example, if you’re a man 41–64 years old with alcohol dependence, Me Again Meds determines sertraline may be most likely to relieve your symptoms based on an analysis of more than 700 patients with a similar medical history. If you’re a woman 20–40 years old with obesity and polycystic ovary syndrome, Me Again Meds recommends bupropion (Wellbutrin), with the caveat that the medication may be ineffective because so few patients in the database match your criteria. The website doesn’t ask for identifying information but provides a report ID you can share with your doctor.

Though patient feedback has been overwhelmingly positive, Alemi says clinicians’ reactions have been mixed. In focus groups and interviews, for example, providers said the analytical model failed to match the nuance of real-world antidepressant prescription and wasn’t representative of the patients they treat. Despite the database’s volume, some clinicians also took issue that it wasn’t a randomized sample of patients with depression.

Me Again Meds has been revised several times in the last three years. Most recently, in a study published in March in the Journal of Mental Health Policy and Economics, Alemi’s team analyzed roughly 2,500 of the site’s subgroups of patients who had received psychotherapy. Still, Me Again Meds remains a survey-based AI that outputs varying multiple-choice questions based on respondents’ previous answers. It’s also brief, taking just minutes to complete. A more advanced chatbot is coming soon.

“Our eventual goal is to create a standalone AI system that diagnoses patients and suggests treatments for the patient in behavioral health,” Alemi says. “That intake process is a long conversation, and I don’t see any long conversations right now in the published literature.”

Last year, GMU launched a prototype chatbot site that’s still active; the conversation kicks off with the bot asking the patient if they’ve experienced major depression. More than 13% of U.S. adults use antidepressants—including 18% of women and 8% of men—according to data the Centers for Disease Control and Prevention collected from 2015–18. The COVID-19 pandemic exacerbated their use.

“We are conscious that the demand for the service would be very large,” Alemi says.

Patient safety is a top concern in bringing an artificial clinician to scale, Alemi says. For example, if a patient is displaying risk factors for suicide, the chatbot would need to terminate the conversation and connect the patient to a live person who is trained to help. Even at a smaller scale, having people monitor chats in real time will help keep the chatbot running smoothly. What’s more, Alemi and his colleagues are working to reduce AI hallucination, or the generation of false or misleading information. They’re also developing a referral system to connect patients without a primary care provider to a prescribing clinician.

“This is a very complicated product; it’s not something that you click on the switch and it works,” Alemi says. “It has many significant parts, and we are working component by component to put it in place.”

Alemi expects the chatbot’s human monitoring feature will be live by the end of the year. He’s also tackling the disparities he sees in the prescription of antidepressants to patients of color. Alemi’s team recently received a grant from the National Institutes of Health (NIH) to research how Black patients with depression respond to medication, using Me Again Meds and the NIH’s All of Us database.

What do antidepressants treat?

Contrary to what its name suggests, antidepressant medication is prescribed to treat more than clinical depression. The Food and Drug Administration has approved certain antidepressants to treat these disorders:

• Bipolar disorder

• Bulimia

• Generalized anxiety disorder

• Obsessive-compulsive disorder

• Other depressive disorders

• Panic disorder

• Posttraumatic stress disorder

• Social anxiety disorder

In addition, clinicians may prescribe antidepressants for off-label use to treat conditions such as migraine, chronic pain, and insomnia.

While Me Again Meds may ask you about a variety of mood, depression, and anxiety disorders, it was designed to help people diagnosed with major depression.

Integrating AI into your next doctor’s appointment

Alemi hopes Me Again Meds proves a powerful resource for patients and providers but notes it doesn’t constitute medical advice. The website is meant to inform discussion between you and your doctor, and only a licensed clinician can prescribe medication Me Again Meds may recommend.

If you’re already taking an antidepressant, don’t stop doing so unless instructed by your doctor; antidepressant discontinuation syndrome may occur without a doctor’s guidance.