Speech pattern analysis may help diagnose, predict ‘more subtle’ hepatic encephalopathy
Click Here to Manage Email Alerts
Key takeaways:
- A speech rate of less than 137 words per minute was tied to higher risk for overt hepatic encephalopathy.
- Speech analysis surpassed animal naming test in spotting minimal hepatic encephalopathy.
A diagnostic tool based on speech variables helped identify minimal hepatic encephalopathy and may predict time to development of overt hepatic encephalopathy in patients with cirrhosis, according to a study published in Hepatology.
“Right now, more subtle forms of hepatic encephalopathy, or HE, are missed,” Patricia P. Bloom, MD, assistant professor of hepatology at Michigan Medicine, told Healio. “Our patients can have early stages of HE, which we call minimal HE (MHE), which is not being diagnosed by providers. This is because the current standard-of-care testing for minimal HE is cumbersome, painful and hard to perform in short clinic visits.
“Hepatologists also right now have very limited tools with which to monitor subtle signs of HE when patients are at home.”
In the prospective, HE Audio Recording to Detect MHE (HEAR-MHE) study, Bloom and colleagues investigated whether automated speech analysis correlated with validated HE tests and MHE diagnosis, and could predict future overt HE.
They enrolled 200 adult outpatients with cirrhosis (median age, 63 years; 50.5% men) and 50 controls with noncirrhotic liver disease from the University of Michigan and Baylor University Medical Center between October 2021 and July 2023.
The researchers recorded patients performing several speech tasks, including paragraph reading, picture description and animal naming, and conducted validated HE diagnostic testing using the psychometric HE score (PHES) at baseline. Patients were monitored for 6 months to identify episodes of overt HE and undergo additional speech measurements.
Bloom and colleagues used an automated speech analysis platform to extract 752 speech variables, which included acoustic, lexical and semantic properties, and developed models for each speech task. They then created the multivariable HEAR-MHE model.
According to results, 89 speech variables from the paragraph reading task significantly correlated with PHES, as did 126 variables for the picture description task and seven for the animal naming test (P < .05 for all).
The HEAR-MHE model predicted MHE with a similar area under the curve as the animal naming test (0.76 vs. 0.69; P = .11), with improved accuracy when age and MELD-Na were added (AUC = 0.82; 95% CI, 0.74-0.9). Sensitivity analysis that excluded 20 patients on opiates or benzodiazepines confirmed these findings (AUC = 0.87; 95% CI, 0.8-0.95).
The researchers reported 15 overt HE events among 12 patients during the 6-month period, which occurred at a median 35 days following study enrollment. The HEAR-MHE model predicted time to occurrence with a concordance of 0.74 (P = .06) and outperformed other models.
“Automated analysis of speech is a novel way to diagnose minimal HE and predict future, more severe episodes of HE,” Bloom said. “Certain aspects of speech correlate with gold standard measures of HE, and a model of five variables identified minimal HE and predicted future overt HE.”
When analyzing speech rate alone, the researchers also found that patients with a speech rate of less than 137 words per minute had a seven-times greater risk for overt HE during the follow-up period compared with those with a faster speech rate.
“This tool is not currently ready for patient care, but an application for easy home use is being developed now,” Bloom told Healio. “In the next year, we will test this application with the potential for patient use thereafter.”
She continued: “This discovery will potentially allow for patients to be easily and quickly diagnosed with minimal HE in clinic, and also may allow for improved outpatient monitoring and treatment of HE. The hope it that someday this tool may improve clinical outcomes and quality of life for patients with cirrhosis and HE.”
Collapse
Read more about
Click Here to Manage Email Alerts