TLM 2023: Speech analyzed from a smartphone app recording identified MHE and predicted OHE with results comparable to those from validated assessments.
In a cohort of older adults with cirrhosis, analysis of brief audio recordings made using a smartphone app identified minimal hepatic encephalopathy (MHE) and predicted future overt HE (OHE), according to findings from 1 of a pair of analyses presented at The Liver Meeting, November 10-14, 2023, in Boston, MA.
A second analysis of data from the same study (HE Audio Recording to Detect MHE or HEAR-MHE) found that whether audio recordings via app were made in the home or clinic setting did not affect the ability of the software program to predict MHE and prediction rates were comparable to those from traditional HE tests.
Study authors describe current tools used to diagnose MHE or to predict the risk of OHE as “burdensome, therefore rarely used in practice.” The lack of a simple approach to assessment can relegate patients to worsening disease and poorer outcomes. Patricia Pringle Bloom, MD, of Michigan Medicine, and colleagues, evaluated the use of recorded speech to identify MHE in previously unaffected patients and to predict future OHE.
Bloom and team enrolled 169 outpatients with cirrhosis from 2 geographically disparate centers to participate in the prospective HEAR-MHE study.
Patients were assessed using the Psychometric Hepatic Encephalopathy Score (PHES), a validated test for diagnosing minimal hepatic encephalopathy (MHE) and the animal naming test (ANT), also an established assessment, and were recorded while reading a paragraph aloud. Bloom and colleagues used the Winterlight Labs analysis platform, a smartphone app designed to characterize speech in dementia, to extract speech variables from the audio recordings, including acoustic, lexical, and syntactic features.
The team followed the cohort prospectively for 6 months to identify episodes of OHE. Participants had been categorized into 3 non-overlapping groups: (1) those with prior OHE and HE treatment, (2) those with MHE and no prior OHE (diagnosed by PHES ≤ -4), and (3) those without MHE or prior OHE. Researchers applied linear, logistic, and Cox regression analyses to predict PHES, MHE, and time to OHE, respectively.
The cohort had a median age of 63 years, 52% were men, and the median score on the Model for End-Stage Liver Disease (MELD) was 9 (scores can range from 6 to 40); more than one-third (39%) had alcohol-related cirrhosis and 33% had fatty liver-associated cirrhosis. The median length of the audio recordings was 35 seconds.
Speech correlates with PHES: 82 speech variables associated with PHES in the overall cohort (P<.05 with false discovery rate [FDR] adjustment). A model of 5 speech variables (2 speech tempo and 3 acoustic) was associated with PHES (r2=0.28, P<.001).
Speech associates with MHE: 71 speech variables significantly differed between patients with and without MHE (ie, groups 2 and 3; P<.05 with FDR adjustment) – 4 speech tempo and 67 acoustic variables. A model of 2 speech variables (1 speech tempo and 1 acoustic) was comparable to animal naming test to identify MHE (AUC 0.70 vs 0.66, P = .19).
Speech predicts future OHE: There were 13 OHE events in 10 patients with cirrhosis (6 with prior OHE, 2 with MHE, 2 without OHE or MHE), which occurred a median 33 days from enrollment. A model including speech rate and 2 acoustic variables predicted time to OHE (P = .001). Specifically, a speech rate <133 words/min was associated with a significantly shorter time to OHE than ≥133 words/min (HR =0.11, P = .001).
Office audio: The researchers reported that speech rate was significantly slower in patients with MHE (152 words/min) and history of OHE (155 words/min) when compared with rates in patients either without MHE or controls (169 and 172 words/min). Results using speech rate to predict MHE were comparable to those when the animal naming test was used (AUC 0.73 vs 0.66, P =.12).
Home audio: Patients performed 43 audio recordings with personal Smartphone apps. Analysis found that 23/225 (10%) speech variables were significantly correlated (<0.05) between home and office recordings, even though recordings were made on different days. Speech rate at home was highly correlated to that in the office (r =0.53, P < .01) and to PHES score in the office (r =0.71, P < .01). Speech rate findings were similar regardless of site (p =.44) or patient region of origin (P = .30), across the cohort or within patient subgroups.
“Recorded speech while reading a paragraph – a quick and easy test – was associated with PHES, was able to identify MHE, and predicted future OHE,” concluded Bloom et al. “Speech is simple to record, provides immediate point-of-care data, and represents a promising biomarker in HE.”