New data presented at the 2024 AD/PD Conference revealed the potential of smartphone speech analysis metrics as a biomarker for hypokinetic dysarthria in early Parkinson disease.
New data from the phase 2 PASADENA study (NCT03100149) presented at the 2024 International Conference on Alzheimer’s and Parkinson’s Disease (AD/PD), held March 5-9, in Lisbon, Portugal, showed that automated vowel space metrics derived from smartphone speech samples exhibited good-to-excellent test-retest reliability in the detection of dysarthria in patients with Parkinson disease (PD). The findings suggest that these metrics may serve as reliable and sensitive biomarkers of hypokinetic dysarthria and its progression in early PD.1
Conducted by lead author Damian Kwaśny, MSc, consultant of voice data analysis at Capgemini Engineering, and colleagues the study assessed the reliability and validity of automatedly generated vowel space metrics from remotely collected smartphone speech samples as biomarkers of articulatory impairment severity and progression in patients with early PD. Investigators included data from 316 treatment naive patients with PD randomized to placebo or prasinezumab (low dose [1500 mg] or high dose [3500 mg] for body weight <65 kg or 4500 mg for body weight ≥65 kg) in PASADENA. Researchers analyzed available baseline speech from 296 patients. Only placebo arm data, further censored at the start of levodopa treatment, were included in the analysis to model natural disease progression (n = 94).
The vowel space-based acoustic measures used in the study, vowel articulation index (VAI, men, r = -.14 [95% CI, -.28 to 0] P <.05; women, r = -.21 [95% CI, -0.39 to -0.01] P <.05), and vowel space area (VSA; men, r = -.25 [95% CI, -.38 to -.11] P <.001; women, r = -.26 [95% CI, -0.44 to -0.07] P< .01), were impacted by sex even after accounting for lower clinical impairment in women. The VSA and VAI measurements showed clinical validity in both men and women, and displayed significant progression over 1 year, independent of sex (P<.05; P<.01; respectively).
READ MORE:
In the study, investigators used the Roche PD Mobile Application v2, which targets the key motor signs in PD, for smartphone based-data collection among the participants. The application assessment suite includes a reading and spontaneous speech test scheduled for administration every second day. There were also 3 questions displayed to the participant 1-by-1, cycling through 14 questions sets. For the participants, each were instructed to read the question out loud and then answer it out loud in 20 seconds. In addition, researchers had the corner vowels’ formant frequencies estimated in voiced speech using a custom, unsupervised algorithm where the 3 questions and answers of the prompted reading and speech test were analyzed together to increase the quality of the findings.2
Although previous research has provided evidence of the clinical validity of these metrics in PD, authors noted that they were usually estimated manually, which prevented utilizing them in remote patient monitoring.3 Thus, researchers noted that existing automatic methods needed further assessment of clinical validity and sensitivity to progression in early PD. All told, the outlook of the current study included the need for further research on assessing the impact of language on features, evaluating the sensitivity of features to detect potential treatment effects such as with prasinezumab, and building upon multimodal speech impairment progression models.
Presented at the
