Sunny Tang, Feinstein Institutes for Medical Research

As a psychiatrist and scientist, I see the devastating toll taken by psychotic illnesses and the almost overwhelming gap between available and needed remedies. In my view, we can only achieve major leaps forward as a field by approaching problems with fresh perspectives and novel solutions. That is why I have dedicated my career to leveraging innovations in technology to better understand psychotic disorders and optimize treatment outcomes.

My presentations at the 2022 SIRS Annual Congress addressed this issue through three approaches.

While schizophrenia is related to significant functional impairment for some individuals, others are able to thrive socially and occupationally. In our talk on “Biopsychosocial Contributions to Functional Outcomes in Schizophrenia: A Data-Driven Machine Learning Approach,” we used machine learning methods to identify different patterns of functional outcomes in schizophrenia, and to relate these in reliable ways to biopsychosocial characteristics. We found that, in addition to a group of individuals who are resilient across the board, and a group who are more impaired, there is a cluster of people with schizophrenia who function well socially and with regards to independent living, but are impaired in occupational functioning. We would not have found this pattern without the machine learning approach, because we would not have known to look for it. We also found that the functioning pattern for each person was highly related to their internal sense of motivation and enjoyment, as well as the volume of some brain structures and cognitive ability. Importantly, race, sex, and socioeconomic status were not strong contributors to the functioning pattern of individuals with schizophrenia.

We know that social cognition – or the ability to process social information – is very important to functioning for people with schizophrenia. In our talk, “Speech and Language Disturbance in Schizophrenia are Related to Social Processing,” we showed that social cognitive is also related to how people with schizophrenia communicate. In particular, the ability of individuals to correctly identify emotions was closely related to how their speech was organized. This was found using traditional clinical ratings for speech, as well as with automated speech analysis – computerized methods for objectively quantifying speech characteristics.

One major roadblock in psychiatry is that treatment is often a guessing game. We have a variety of effective treatments available, but we do not yet know who will respond to standard care, and who will not. In our poster presentation on “Predicting Treatment Outcomes with Computational Speech Features in Hospitalized Patients with Schizophrenia,” we used an app developed by Winterlight Labs to record speech from hospitalized patients with schizophrenia just after they were admitted to the inpatient facility. The speech samples underwent automated processing. We found that speech features soon after admission significantly contributed to predicting how symptomatic people were when they were discharged (2 weeks later, on average).

We still have much to do before our research can tangibly benefit people with schizophrenia and their loved ones. However, I believe that taking advantage of advances in technology and machine learning has enormous potential for understanding what goes wrong in schizophrenia and for guiding personalized medical treatment to improve outcomes.