Katie first spoke with a demon when she was 14. He perched on the edge of her bed, and would persistently urge her to do bad things — like blow up her Arkansas high school.

She spoke to God, too.

Her parents, Pentecostal Christians, believed her visions made her special. So she received no therapy, and no medications, and no diagnosis as her schizoaffective disorder began to take root.

Katie, who is now 35, has been homeless and hospitalized several times. She tried "just about every drug there is," she said, before she found a medication — the antipsychotic risperidone — that works well for her. She's got a happy and stable life these days, living with her husband in Texas. But she knows it's tenuous.

"The thing is: A lot of times, a drug will work for you for several years, and then it'll just stop," said Katie, who asked that only her first name be used to protect her privacy. "At some point, I know I'll have to find another drug."

It's a common, and well justified, fear for people with psychiatric disorders. While scientists have made tremendous advances in decoding the genetics of physical illnesses, such as cancer, and developing precision therapies, treatments for mental health remain blunt tools.

They work by blasting entire mechanisms in the brain, without addressing the specific chemical pathways that have gone awry. As a result, the side effects can be quite substantial — and the efficacy is often low. Some people develop resistance to a drug years into treatment, and then have to go through a long period of trial and error to find another.

While demand for mental health drugs has surged, big pharmaceutical companies have largely backed away from investing in the field; the number of psychopharmacological drug research programs has shrunk 70 percent in the last decade, according to NeuroPerspective.

In part, that's because generic versions of popular drugs like Prozac dominate the market, leaving little incentive for companies to sink tens of millions into developing alternatives. But it's also because the biological causes of mental illness are so complex. There hasn't been much innovation in psychiatric medications in more than two decades.

To advance the field, researchers say they need to find biomarkers — tangible biological clues that can help diagnose mental illness, just the way high blood glucose levels can signal diabetes. The hope is that those biomarkers could help pinpoint what's gone wrong in the circuitry of a particular patient's brain and offer clues for drug development — and, perhaps one day, even precision psychiatric therapies. But that's far easier said than done.

Some biomarkers might not be physical: Research suggests that voice analysis could give clues as to a patient's mental illness, as certain sentence structures and cadences can objectively be linked to psychiatric disease.

There may be hints in the blood, too, suggests an intriguing small study published in Analytic Chemistry. It's based on research from the University of Maryland that found that testing for oxidative stress — that is, the imbalance of free radicals in the bloodstream — might help doctors diagnose schizophrenia more quickly.

As more genes are linked to various mental illnesses, the number of psychiatric biomarkers should increase, said Dr. Stephen Strakowski, chair of psychiatry at Dell Medical School at the University of Texas Austin.

Yet drug development veterans are careful to add a note of caution: Even when the mechanism of a brain disease is better understood, it'll still be a very long road to better treatments. The physical characteristics of brains disordered by Alzheimer's are well known, for instance, yet drug companies have failed for decades to come up with effective drugs, despite pouring hundreds of millions into research.

When it comes to mental health, one big help would be developing better animal models on which to test potential psychiatric drugs.

Researchers studying cancer and many other illnesses can use mice, primates, and other lab animals to understand how the disease behaves and whether a treatment looks promising.

But the human brain and mind are extraordinarily difficult to replicate. Mice simply don't have the same thought processes as humans, and therefore the misfirings in a rodent brain can't compare to a person's, said Robert Desimone, director of the McGovern Institute for Brain Research at MIT. So the brain mechanisms that lead to PTSD, or anxiety, or schizophrenia — while all biological — are harder to understand.

"It took literally decades of work in the cancer field before we got to a level of mechanistic understanding," Desimone said. That's vital to achieve in psychiatry, he added: "That's what the field has been missing all this time."

That's why much of the institute's focus these days is to engineer better animal models for mental illness. Using tools like CRISPR gene editing and optogenetics, scientists hope to create animals whose brain misfirings can more closely mimic what's going on in mental illness.

A new center for autism research at MIT, meanwhile — launched with a $20 million donation — will focus on unraveling the genetic and biological basis of that disease.

"When it comes to pharmaceutical development, we're not necessarily looking for a new treatment that totally cures a psychiatric disease," Desimone said. "But we'd like to make a substantial impact: If we could improve the intellectual disability in autism by 50 percent, for example, that'd be huge."

The global pharma company Novartis is also pumping resources into studying the genetics of psychiatric diseases, focused on the relatively small populations of people who have a clearly inherited predisposition to mental illness.

Big pharma used to test its therapies on broad swaths of people with mental illness — say, anyone with a diagnosis of bipolar disorder. Novartis plans a much more targeted approach, looking for subgroups of patients with similar genetic traits or biomarkers, according to Ricardo Dolmetsch, a former Stanford researcher who now heads up neuropsychiatric drug discovery at the company.

Novartis is also exploring the idea of conducting clinical trials that couple medication with a digital therapy tool — creating an app, for instance, that helps strengthen neural circuitry in tandem with a drug.

That's part of a broader push among doctors to test out video games, for instance, as "electronic medicine" to help treat mental illness such as depression. "The brain is very plastic, and with the combination of a drug with a sort of digital therapy as well, you could create a better clinical trial," Dolmetsch said.

Researchers are also testing out transcranial magnetic stimulation of nerves in the brain to quickly improve symptoms of depression. Doctors have long used electroconvulsive therapy, which deploys electric shocks to jar a patient's nervous system out of treatment-resistant depression. Now, scientists are looking at updated devices that might cause fewer side effects.

The lowest-hanging fruit, according to Dolmetsch, is the potential of so-called street drugs like ketamine, psilocybin, and ecstasy in treating mental illness. Ketamine, long used as an anesthetic, is showing remarkable promise in treating depression. Psilocybin, the active chemical in hallucinogenic mushrooms, is being studied as a way to ease anxiety and depression in terminally ill cancer patients. And MDMA, or ecstasy, is being studied as a psychotherapy tool to treat PTSD.

"Perhaps the right combination of a hallucinogen with a psychiatric or a behavioral therapy might help these drugs make an impact," Dolmetsch said.

Not everyone agrees that science has advanced far enough to start thinking about precision treatments for mental illnesses.

Jeff Jonas, for instance, has spent his career working on psychiatric drugs, such as the antidepressant Lexapro, the anxiety medication Xanax, and ADHD drug Vyvanse. He's now the CEO of a small biopharma company, Sage Therapeutics, working in the field of mental health.

Jonas said he thinks the treatment of mental illness is "not amenable to a target-based approach" — at least not yet.

Instead, he's focused on looking for compounds that might act in a novel way to modulate brain activity for most patients with a specific condition. His experimental postpartum depression drug, for instance, aims to tinker with a neurotransmitter that's known to soothe anxiety. It is showing early efficacy in helping to ward off serious symptoms like panic attacks, inconsolable sadness, and feelings of inadequacy.

Improvements can't come quickly enough for patients.

Psychiatric drug prescription rates have increased substantially in the past two decades, with the rate of antidepressant use alone quadrupling. One in six Americans now takes a psychiatric drug, such as an antidepressant or a sedative, according to a recent report.

Yet finding the right drug can be a painful process of trial and error. And waiting.

Most psychiatric medications take weeks or months to work, so patients and their doctors must wait to see if a drug is effective, wait to tinker with the dose, and wait to see if another medication works better.

It took Katie's doctors 15 years to properly diagnose her condition, and even longer to find the right drug. Each medication she tried had its own spate of side effects: Celexa, for instance, made her gain more than 100 pounds, and she developed diabetes.

On Risperdal, the diabetes has gone away — but she still has high blood pressure and high cholesterol, and must take additional drugs to keep them in check. "With Risperdal, the voices don't go away — but they get really quiet," Katie said. "They hardly bother me anymore."

She isn't holding her breath for a miracle drug that will cure her disease. But she is holding on.

"I find that bravery is highly relative: If you have no other option, it's easy to be brave," Katie said. "But my biggest takeaway: For all those who are suffering, there is hope."

This story was produced by STAT, a national publication covering health, medicine, and life science. Read more and sign up for their free morning newsletter at statnews.com. You can also follow STAT on Twitter and like them on Facebook.