Big Data and AI May Hasten Precision Medicine in Psychiatry

Smoller presented data from a January 2022 study in JAMA Open Network showing that clinicians might improve their ability to identify patients at high risk of suicide by using data from a brief patient self-report scale and electronic health records (EHRs). In an analysis of suicide risk among patients visiting a hospital emergency department (ED) for psychiatric symptoms, Smoller and colleagues found that clinicians’ assessment alone was slightly better than chance at predicting suicide attempts; it was slightly higher using models based on data from EHRs and was best using the brief patient scale combined with an EHR-based algorithm.

“A composite approach that combines patient self-report with data from an EHR-based risk score may improve the prediction of suicide attempts by patients after visiting an ED with levels of accuracy beyond what is possible with clinician assessment or EHR-based risk scores alone,” Smoller and colleagues wrote.

Smoller also described an app for calculating suicide risk that has been devised for use with EHRs and is now being used in a randomized clinical trial by the NIMH-funded Center for Suicide Prevention and Research, which he co-directs with Matthew Nock, Ph.D., at Harvard.

Smoller presented data from an April 2023 report in npj Digital Medicine describing a study in which he and his team used EHR data and AI to predict response to selective serotonin reuptake inhibitors (SSRIs), serotonin and norepinephrine reuptake inhibitors (SNRIs), mirtazapine, and norepinephrine four to 12 weeks after initiation. Several AI models successfully estimated the probability of treatment response between patients and between antidepressant classes for the same patient.

“That is a step toward implementing clinical decision support that could address one of the major problems we have: What is the right treatment for this person up front?” Smoller said.

“We have used large-scale genomic data to observe at a genetic level the boundaries among these disorders,” he continued. “We were able to identify basic underlying factors that seem to capture the genetic basis for all of these conditions.” A 2023 report in JAMA Psychiatry described a study in which Smoller and colleagues used “transcriptome-wide structural equation modeling” (T-SEM) to identify 466 genes associated with risk for 13 disorders. (T-SEM is a statistical method for studying the effects of tissue-specific gene expression across genetically overlapping traits.) Results also revealed five drug-gene pairs that may have transdiagnostic effect and 35 drug-gene pairs that may target risk specifically for bipolar disorder and schizophrenia.

Smoller presented work that used genomic and EHR data to calculate a polygenic risk score for depression to show that increased risk scores were associated with an increased white blood cell count. He also described use of AI to successfully calculate risk for bipolar disorder. Both of these efforts drew on data from PsycheMERGE (electronic Medical Record and GEnomics), a partnership of researchers across the world who are applying genomics and health system data to advance precision psychiatry.

“We are now in an era where we can bring precision medicine to psychiatry using the vast amounts of data now available and new technologies in the realm of AI and genomics to improve how we diagnose and reduce morbidity and mortality,” Smoller said. “We are still in the early stages, and this promise is yet to be realized in clinical care. But I am quite optimistic that pursuing these approaches and leveraging these tools can bring us closer to answering some of the clinical challenges for which we have had too few answers for too long a time.” ■