A computer analyzing speech has correctly identified five individuals who would later experience a psychotic episode against 29 who would not among a group of high-risk patients in a proof-of-principle study.

[disturbed young man]Share on Pinterest
Researchers say a speech-analyzing computer could be a useful tool for predicting psychosis among young adults.

The findings raise the prospect of a clinical tool to aid the diagnosis and prognosis of severe mental disorders such as schizophrenia.

The researchers, publishing in the journal NPJ Schizophrenia, also believe there may be potential for monitoring treatment.

They also say early identification through speech could help to delay, mitigate or even prevent the onset of serious mental illness.

Computerized analysis provided a more accurate classification than clinical ratings, found the researchers from Columbia University Medical Center, New York State Psychiatric Institute and the IBM TJ Watson Research Center.

The authors conclude:

“Computerized analysis of complex human behaviors such as speech may present an opportunity to move psychiatry beyond reliance on self-report and clinical observation toward more objective measures of health and illness in the individual patient.”

The capacity of psychiatry to diagnose and treat serious mental illness has been “hampered” by the absence of objective clinical tests, say the authors. Other fields of medicine enjoy routine use of such tests.

Computers are only now beginning to be explored in psychiatry, but they have long been in use with models combining demographic data and purchasing behavior to personalize advertising content, “and automated language assessment is employed to screen job candidates and score essays.”

In a variety of altered states, the unique window into the mind presented by speech is particularly relevant to psychosis. Speech gives “important clues about what people are thinking and feeling.”

A clinical psychiatrist may intuitively recognize signs of disorganized thoughts during a traditional interview, but a machine can precisely measure speech variables.

For the study, participants gave an open-ended, narrative interview during which they described subjective experiences. The interview transcripts were computer-analyzed for patterns of speech, including semantics (meaning) and syntax (structure).

Under analysis was each patient’s semantic coherence – how well they stayed on topic – and syntactic structure, such as phrase length and use of words that link phrases.

The participants were followed up for 2.5 years for onset of psychosis, assessed every 3 months. The authors found:

“Classification based on automated analysis outperformed that based on clinical ratings, indicating that automated speech analysis can increase predictive power beyond expert clinical opinion.”

The features of speech that were found to be predictive included breaks in the flow of meaning from one sentence to the next, and speech that was characterized by shorter phrases with less elaboration.

The speech classifier tool to mechanically sort symptom-related features achieved 100% accuracy, correctly differentiating between the five individuals who later experienced a psychotic episode and the 29 who did not.

“Although needing to be replicated in larger samples,” the researchers say the method could identify thought disorder – a key component of schizophrenia – in its earliest, most subtle form years before the onset of psychosis.