Listener effort quantifies clinically meaningful progression of dysarthria in people living with amyotrophic lateral sclerosis

Abstract

Amyotrophic lateral sclerosis (ALS) is a neurodegenerative motor neuron disease that causes progressive muscle weakness. Progressive bulbar dysfunction causes dysarthria and thus social isolation, reducing quality of life. The Everything ALS Speech Study obtained longitudinal clinical information and speech recordings from 292 participants. In a subset of 120 participants, we measured speaking rate (SR) and listener effort (LE), a measure of dysarthria severity rated by speech pathologists from recordings. LE intra- and inter-rater reliability was very high (ICC 0.88 to 0.92). LE correlated with other measures of dysarthria at baseline. LE changed over time in participants with ALS (slope 0.77 pts/month; p<0.001) but not controls (slope 0.005 pts/month; p=0.807). The slope of LE progression was similar in all participants with ALS who had bulbar dysfunction at baseline, regardless of ALS site of onset. LE could be a remotely collected clinically meaningful clinical outcome assessment for ALS clinical trials.

Figure 1:. Cohort Diagram.

Flow of participants from enrollment in the overall study, through quality assurance, inclusion in the overall speech study, inclusion in the Listener Effort Substudy and creation of cohorts within the LE Substudy. Criteria for inclusion in sub-studies and cohorts are noted. Cohorts described in specific Tables and Figures throughout the study are noted.

Figure 2 |. LE Inter-Rater Reliability.

Pairwise inter-rater reliability was excellent for two of the three rater pairs: (a) LE Rater 1 and 2 ICC was 0.92; (b) LE Rater 1 and 3 ICC was 0.91. (c) Inter-rater reliability was very good for the third pair: LE Rater 2 and 3 ICC was 0.88.

Figure 3 |. LE vs ALSFRS-RSE and acoustic features.

(a) Correlation matrix was computed over the population of controls and PALS. Correlations are expressed in absolute values. Speaking Rate (words/min; SR) and Articulation Rate (syllables/sec; AR) are remarkably highly correlated (Pearson R = 0.95), indicating that they measure the same aspect of speech. There is a good correlation between Listener Effort (LE) and AR (Pearson R=−0.75), SR (−0.74), ALSFRS-R Question 1 (−0.70) and ALSFRS-RSE bulbar subdomain (−0.65). This indicates that while LE measures similar concepts to these measures, it also contributes non-overlapping information. As expected, because ALSFRS-RSE covers many more domains than just speech, it shows low to moderate correlation with AR, SR and LE. (b) In participants with lower self-reported speech function on the speech question (Q1) of the ALSFRS-RSE, LE increases (lower ALSFRS-RSE scores, and higher LE, denote lower speech function). Because Q1 of the ALSFRS-RSE only has five categorical answers, each category contains a wide spread of LE scores. These differences are significant comparing the categorical answers 2–3 and 3–4, but not 0–1 and 1–2. This may be because of the low numbers of participants in the lowest categories of ALSFRS-RSE Q1.

Figure 4 |. Progression of LE in PALS and controls.

We analyzed ALS progression using linear mixed models (LMM) in different cohorts. In panels (a-c) we plotted LE data since onset from all participants, and in panels (d-f) we plotted LE data since enrollment from participants with onset of ALS within 3 years of study initiation. In panels (a) and (d) we compare PALS and controls. The slope of decline of LE is higher for PALS than controls in both all participants and those with onset <3yrs prior to enrollment. In panels (b) and (e) we compare PALS with bulbar and non-bulbar onset. PALS with bulbar onset show a faster slope of decline on LE than those with non-bulbar onset, consistent with the concept that PALS with bulbar onset have faster progression. In panels (c) and (f) we compare PALS with bulbar and non-bulbar onset, excluding PALS with LE scores in the normal range (0–10) at the time of enrollment. This partition focuses the analysis on PALS with current bulbar symptoms. In this analysis, LE slopes for PALS with bulbar and non-bulbar onset show no statistical differences. This suggests that once participants have developed bulbar symptoms, LE tends to progress at a similar rate, whether the disease began in the bulbar region or not.

Figure 5 |. LE Prediction Model.

(a) A Lasso regression model was trained and evaluated using nested cross-validation. The predicted LE output is plotted on the y-axis against the LE as assessed by SLPs on the x-axis. Each outer fold is shown in a different color. The model showed a robust performance: the RMSE averaged across the five outer folds was 8.56 ± 0.60, and the average R2 was 0.83 ± 0.07. (b) The weight of each feature averaged across the outer folds (error bars are standard deviations). Most of the predictive power of the LEPM comes from Speaking Rate and Whisper Confidence