Use of Beiwe Smartphone App to Identify and Track Speech Decline in Amyotrophic Lateral Sclerosis (ALS)

Kathryn P. Connaghan, Jordan R. Green, Sabrina Paganoni, James Chan, Harli Weber, Ella Collins, Brian Richburg, Marziye Eshghi, J.p. Onnela, James D. BerryInterspeech 2019

TL;DR

This paper explores the use of a smartphone app called Beiwe to remotely monitor speech decline in individuals with Amyotrophic Lateral Sclerosis (ALS). The study demonstrates that Beiwe can effectively identify differences in speech function between ALS patients with and without bulbar symptoms and track the rate of decline, suggesting its potential as a valuable tool for early diagnosis and disease progression monitoring.

What Problem Does This Paper Address?

Amyotrophic Lateral Sclerosis (ALS) is a devastating neurodegenerative disease that affects motor neurons, leading to progressive muscle weakness and loss of function. A significant challenge in managing ALS is the early identification and tracking of disease progression, particularly the involvement of bulbar systems, which control speech and swallowing. Bulbar symptoms can significantly impact quality of life and are associated with reduced survival. Traditional methods for assessing speech decline rely on subjective measures like clinician or patient reports. While instrumental speech measures have shown promise in detecting bulbar symptom onset earlier than subjective methods, they often require specialized equipment and in-person visits, posing barriers for ALS patients who experience mobility issues. This paper addresses the need for a more accessible, frequent, and objective method to monitor speech decline in ALS, leveraging the ubiquity of smartphones.

Methods

The researchers utilized the Beiwe smartphone-based digital phenotyping platform, which includes native applications for Android and iOS devices. This platform allows for both passive data collection (e.g., sensor data) and active data collection, including voice recordings and self-reported surveys. For this study, participants used the Beiwe app weekly to record themselves reading a standardized passage and complete the Revised ALS Functional Rating Scale-Revised (ALSFRS-R), specifically focusing on bulbar function questions.

Key methodological aspects include:

  • Speech Data Acquisition: Participants recorded themselves reading a short passage designed to aid in pause detection. Speech samples were recorded at a 44.1kHz sampling rate with no compression.
  • Speech Feature Extraction: Offline processing using Audacity was performed to reduce noise and remove non-speech sounds. Speech Pause Analysis (SPA) software, a semi-automated MATLAB procedure, was then used to extract speech and pause variables. The primary variables analyzed were speaking rate, articulation rate, and percentage of pause time.
  • ALSFRS-R Data: Participants self-reported their bulbar function using specific questions from the ALSFRS-R.
  • Participant Stratification: Participants were divided into two groups: "Bulbar" (those with speaking rate at or below 150 words per minute for at least the first two sessions) and "No-Bulbar" (those with speaking rates above 150 WPM). This stratification was based on previous research indicating that speaking rate can be an early indicator of bulbar symptoms.
  • Statistical Analysis: Two types of analysis were conducted:
  • Baseline Group Differences: Mann-Whitney-Wilcoxon Test and Welch two-sample t-tests were used to compare group differences in speech function and ALSFRS-R scores at baseline.
  • Longitudinal Analysis: Linear mixed-effects models were fitted to examine changes in speech measures over time, including fixed effects for week, bulbar group, and an interaction term between week and bulbar group to assess differential rates of decline.

Datasets

The study utilized data collected from 12 participants diagnosed with ALS according to the El Escorial Criteria. These participants were part of a larger study on smartphone-based digital phenotyping of ALS progression. No new benchmark datasets were introduced; rather, the study leveraged data collected through the Beiwe platform from these ALS patients.

Key Insights and Ideas

  • Smartphone-based monitoring is feasible and effective: The study successfully demonstrated that a smartphone app can be used to collect high-quality speech data and self-reported functional status from ALS patients remotely and frequently.
  • Beiwe can differentiate between bulbar and non-bulbar ALS: The platform was able to identify significant differences in speech function at baseline between participants with and without bulbar symptoms, suggesting its potential as a screening tool.
  • Speech decline is accelerated in bulbar ALS: The longitudinal analysis revealed that participants with bulbar symptoms exhibited a significantly faster rate of decline in several speech measures, including speaking rate and articulation rate, compared to those without bulbar symptoms.
  • Objective speech measures are crucial: The findings reinforce the idea that objective acoustic measures can detect subtle changes in speech that may not be immediately apparent through subjective assessments.
  • Percentage pause time as a sensitive indicator: While not significantly different at baseline, the percentage of pause time showed a significant week*group interaction, indicating a differential increase over time in the bulbar group, highlighting its potential as a sensitive marker of progression.

Key Results

  • Baseline Group Differences: Significant differences were observed between the Bulbar and No-Bulbar groups for ALSFRS-R speech question (p=0.006), ALSFRS-R bulbar total score (p=0.05), speaking rate (p=0.001), and articulation rate (p=0.002). Percentage pause time did not differ significantly at baseline (p=0.56).
  • Longitudinal Changes:
    • When data were pooled, only ALSFRS-R bulbar total score and articulation rate showed significant changes over time.
  • However, the week*group interaction revealed that the Bulbar group demonstrated significantly faster rates of decline compared to the No-Bulbar group for ALSFRS-R speech question (p<0.001), ALSFRS-R bulbar total score (p<0.001), speaking rate (p<0.001), and articulation rate (p<0.001).
  • The percentage pause time also showed a statistically significant week*group interaction (p<0.0001), indicating a faster increase in the Bulbar group.

Limitations and Future Work

Limitations:

  • Small Sample Size: The study involved only 12 participants, which is a small sample size. The authors acknowledge that findings should be interpreted with caution due to this limitation.
  • Limited Data Points for Long-Term Analysis: For some participants, data beyond 38 weeks were limited, affecting the robustness of very long-term trend analysis.

Future Work:

  • Larger Sample Investigation: The authors emphasize the need for larger sample investigations to confirm these findings and to explore a comprehensive set of speech variables.
  • Inclusion of Other Acoustic Measures: Future research aims to incorporate acoustic analysis of cough recordings to further inform bulbar function assessment.
  • Integration of Passive Data: The researchers plan to integrate passive data collection capabilities within the Beiwe platform (e.g., GPS, phone/text logs) to develop a more comprehensive understanding of communicative participation changes in individuals with ALS.
  • Diagnostic Screening Tool Development: The findings suggest potential for developing Beiwe into a diagnostic screening tool for early detection of bulbar symptoms.

Summary updated

BibTeX
@inproceedings{connaghan2019use,
  author = {Kathryn P. Connaghan and Jordan R. Green and Sabrina Paganoni and James Chan and Harli Weber and Ella Collins and Brian Richburg and Marziye Eshghi and J.P. Onnela and James D. Berry},
  title = {{Use of Beiwe Smartphone App to Identify and Track Speech Decline in Amyotrophic Lateral Sclerosis (ALS)}},
  booktitle = {Interspeech},
  year = {2019},
}
Figure