br Patients read aloud the first paragraph of La
Patients read aloud the first paragraph of La chèvre de Monsieur Seguin by Alphone Daudet:
“Monsieur Seguin n’avait jamais eu de bonheur avec ses chèvres. Il les perdait toutes de la même fac¸ on. Un beau matin, elles cas-saient leur corde, s’en allaient dans la montagne, et là-haut le loup les mangeait. Ni les caresses de leur maître, ni la peur du
loup, rien ne les retenait. C’était, paraît-il, des chèvres indépen-dantes voulant à tout prix le grand air et la liberté”.
This text was chosen for its length and balanced phonetic make-up; it AWD 131-138 is widely used in clinical assessment in France .
Patients randomly chose one out of several photographs, all showing a sea-front with boats.
They were asked to describe the picture so that a listener some distance away could copy it. The description task lasted at least 1 minute, to give the listening panel sufficient material.
22.214.171.124. Listening panel. The completed corpus of recordings was analyzed by a listening panel comprising 6 speech therapists, work-ing in the ENT department and thus considered to be experts.
2.3. DESC task analysis
Individual recordings were rendered anonymous and presented to the panel, who listened to them via headphones in a calm setting.
Panelists were informed that they would hear recordings of sub-jects describing a photograph related to the sea. After listening, they were to make two assessments:
• impairment of intelligibility and of speech signal, on a 0–10 scale (the lower the score, the greater the perceived disorder) by unit steps (discrete scoring). The two terms were defined in the instructions as follows: “Intelligibility is the comprehensibility of the message carried by the signal; signal impairment severity is the degree of overall vocal signal impairment.”;
• the degree of difficulty of understanding the information given by the speaker, on a visual analogue scale (VAS), placing a bar along a line going from “I understood nothing” to “I understood every-thing”, via “I understood some words but not the message” and “I understood the message as a whole but not all of the words”. From this, a continuous quantitative score was derived.
Thus, on the DESC task, there were 3 global scores per patient: signal impairment (“sig”), intelligibility (“intell”), and degree of dif-ficulty of understanding (“intelldeg”).
2.4. READ task analysis
The reading task was analyzed 6 months after the picture description task. The perceptual assessment procedure was similar except for the absence of “Inteldeg” assessment on VAS.
For the READ task, there were thus 2 scores per panel per patient; signal impairment and intelligibility.
Description of distributions, correlations between variables and differences in score shed light on the metric quality of the various parameters.
Population characteristics were described. Perceptual assess-ment results were panel-dependent, and inter-observer reliability was assessed on interclass correlation coefficients (ICC), cor-responding to the proportion of variance associated with inter-patient variability free of measurement error: i.e., the pro-portion of variance accounted for by a given factor. In the present case, ICC assessed the proportion of variance in perceptual score due to inter-observer variability compared to overall variation in perceptual score: ICC = (inter-observer variation)/(inter-observer variation + random error). Thus, the higher the ICC, the lower the random error and the more homogeneous the inter-observer scores. Normal values for interpretation purposes have been
Please cite Lethal locus article in press as: Balaguer M, et al. Assessment of impairment of intelligibility and of speech sig-nal after oral cavity and oropharynx cancer. European Annals of Otorhinolaryngology, Head and Neck diseases (2019), https://doi.org/10.1016/j.anorl.2019.05.012
M. Balaguer et al. / European Annals of Otorhinolaryngology, Head and Neck diseases xxx (2019) xxx–xxx 3
Interclass correlation coefficients per test.
ICC 95% confidence interval
reported, indicating moderate reliability for ICC values between 0.5 and 0.75, and good reliability for values > 0.75 .
Panel scores were reported as mean per panel. Normal distribution was checked on Shapiro-Wilk test, and residual homoscedasticity on Breusch-Pagan test. The various test results were reported on box-plots. Comparisons and correlations used non-parametric tests as appropriate: Wilcoxon and Spearman cor-relation tests, respectively. Bonferroni correction was applied on multiple tests, to take first-order risk inflation into account.