ABSTRACT
Background
While persons with aphasia (PwA) are often dichotomised as fluent or nonfluent, agreement that fluency is not an all-or-nothing construct has led to the use of continuous variables as a way to quantify fluency, such as multi-dimensional rating scales, speech rate, and utterance length. Though these measures are often used in research, they provide little information about the underlying fluency deficit.
Aim
The aim of the study was to identify how well commonly used continuous measures of fluency capture variability in spontaneous speech variables at lexical, grammatical, and speech production levels.
Methods & Procedures
Speech samples of 254 English-speaking PwA from the AphasiaBank database were analyzed to examine the distributions of four continuous measures of fluency: the WAB-R fluency scale, utterance length, retracing, and speech rate. Linear regression was used to identify spontaneous speech predictors contributing to each fluency outcome measure.
Outcomes & Results
All the outcome measures reflected the influence of multiple underlying dimensions, although the predictors varied. The WAB-R fluency scale, speech rate, and retracing were influenced by measures of grammatical competence, lexical retrieval, and speech production, whereas utterance length was influenced only by measures of grammatical competence and lexical retrieval. The strongest predictor of WAB-R fluency was aphasia severity, whereas the strongest predictor for all other fluency proxy measures was grammatical complexity.
Conclusions
Continuous measures allow a variety of ways to objectively quantify speech fluency; however, they reflect superficial manifestations of fluency that may be affected by multiple underlying deficits. Furthermore, the deficits underlying different measures vary, which may reduce the reliability of fluency diagnoses. Capturing these differences at the individual level is critical to accurate diagnosis and appropriately targeted therapy.
Disclosure statement
No potential conflict of interest was reported by the authors.
Supplementary Material
Supplemental data for this article can be accessed here.
Notes
1. Package gvlma in R was used to assess linearity, skewness, kurtosis, and homoscedasticity of residuals. To diagnose problems with the data, we used the commands vif to assess for variance inflation; skewness to assess normality; ncvTest to assess homoscedasticity; Durbin–Watson test to assess the independence of errors; OutlierTest, outliers, and hist to identify outliers; and plotted Cook’s Distance to identify influential cases.