Spontaneous Speech Alterations and Evolution in Primary Progressive Aphasia Variants
Elisa Canu1, Laura Lumaca1, Veronica Castelnovo1, Silvia Basaia1, Sofia Santicioli1, Elena Gatti1, Alessandra Lamanuzzi1, Edoardo Spinelli3, Giordano Cecchetti4, Francesca Caso2, Giuseppe Magnani2, Paola Caroppo6, Sara Prioni7, Cristina Villa6, Stefano Cappa8, Massimo Filippi5, Federica Agosta3
1Neuroimaging Research Unit, Division of Neuroscience, 2Neurology Unit, IRCCS San Raffaele Scientific Institute, 3Neuroimaging Research Unit, Division of Neuroscience; and Neurology Unit, 4Neuroimaging Research Unit, Division of Neuroscience; Neurology Unit; and Neurophysiology Service, 5Neuroimaging Research Unit, Division of Neuroscience; Neurology Unit; Neurorehabilitation Unit; and Neurophysiology Service, IRCCS San Raffaele Scientific Institute; and Vita-Salute San Raffaele University, 6Unit of Neurology 5 - Neuropathology, Fondazione IRCCS Istituto Neurologico Carlo Besta, 7Fondazione IRCCS Istituto Neurologico Carlo Besta, 8Department of Humanities and Life Sciences, University Institute for Advanced Studies IUSS Pavia
Objective:
To identify: (i) which features of speech (alone or in combination with standard language tests and/or brain gray matter [GM] volumes) most effectively distinguish primary progressive aphasia (PPA) variants; (ii) how spontaneous speech evolved over time, and (iii) the best combination of features predicting speech evolution in each PPA variant.
Background:
An accurate speech characterization in PPA could positively impact on diagnosis and patient management.
Design/Methods:
95 PPA patients (40 nfvPPA, 35 svPPA, 20 lvPPA) underwent the “Picnic Scene” test and structural MRI. A subgroup of 34 patients underwent a similar follow-up. Stepwise regression models were used to identify the speech parameters that best distinguished the groups, also incorporating GM volumes, standard language tests, age, sex and education. In each PPA group, linear mixed effect models were performed for defining speech changes over time, and the prediction analysis were conducted using variables from the best stepwise models.
Results:
The best models to differentiating PPA variants included: left temporal and middle frontal volumes, and syntax production features when comparing nfvPPA vs svPPA (R2=0.89); lexical contents, syntax complexity, left temporal and insular brain volumes in nfvPPA vs lvPPA (R2=0.81); left temporal volumes and speech production rates in svPPA vs lvPPA (R2=0.86). Over time, nfvPPA patients showed more phonological errors, which were predicted by syntax production features at baseline. SvPPA and lvPPA showed reduced naming and reduced number of words in sentences, respectively, which were predicted by baseline left temporal volumes.
Conclusions:
The stepwise regression models that best distinguish the three PPA variants included both speech features and GM volumes. By combining all these variables, increased accuracy was found mainly when nfvPPA and lvPPA were compared. Over time, the PPA variants showed different speech trajectories, which were predicted at baseline by speech features and integrity of specific brain volumes.