Cognitive Impairment in Schizophrenia: How Machine Learning Is Helping

With increased attention on cognitive impairment in schizophrenia, it is growing more important to address and understand the domains involved. Robert Chen, MD, PhD, shared highlights of his research leveraging machine learning at the 63rd Annual Meeting of the American College of Neuropsychopharmacology in Phoenix, Arizona.¹

Chen, a PGY-2 at the University of Washington's Psychiatry Residency Research Program, has a keen interest in biomarker and therapeutic development for psychiatric diseases, especially those with a strong genetic component such as schizophrenia, he told Psychiatric Times in an exclusive interview. To better understand the domains of cognition impacted by schizophrenia, he conducted a case-control study that included 600 patients with schizophrenia and 750 healthy comparison individuals. The study included 15 cognitive assessments (eg, Degraded Stimulus and Identical Pairs Continuous Performance Tests, Letter Number Span, Penn Computerized Neurocognitive Battery, the California Verbal Learning Test II).

As a result of the machine learning models, 2 cognitive measures appeared to be most indicative of schizophrenia when comparing results between the treatment group and the healthy participant group. These measures involved verbal learning and emotional identification, Chen explained.

The results proved to be very interesting, Chen said. “Interesting because verbal learning is essential in human communication, including daily tasks such as shopping, writing checks… and Emotional reading is important for social cognition and being to interact with others.”

Moreover, having 2 cognition domains to assess as compared with 15 makes it more real world applicable, he added. Administering 15 tests could take days if not longer, and this finding can help simplify things and get information to clinicians faster.

The team also collected electrical markers in the brain using EEG signals as part of the study, and Chen said incorporating that data into machine learning would be next step in the research. Ultimately, he hopes that looking at biosubtypes using EEG signals, cognition markers, and maybe even genetics we will be able to not only predict symptomology and treatment issues.

The sharing of data, like that collected from this study and so many other studies presented at the ACNP meeting, help to move the field forward, he said. Although it is his first time attending the annual meeting, Chen said it is helpful to hear what has been working, what hasn’t, and to learn from other researchers. He also appreciates the unique way ACNP brings together researchers, clinicians, and industry.

Chen said it makes him hopeful that psychiatry can move “from a shotgun approach toward precision therapy, which is being done in other fields of medicine, like cancer and cardiology.

Looking for more updates from the ACNP Annual Meeting? Subscribe to our e-newsletter.

Reference

1. Chen R. Machine Learning Reveals a Sparse Set of Cognitive Domains Impaired in Schizophrenia. Presented at the 63rd Annual Meeting of the American College of Neuropsychopharmacology. Phoenix, Arizona; December 8-12, 2024.