Connected Speech Alterations and Progression in Patients With Primary Progressive Aphasia Variants.

Abstract

Background and objectives: Diagnosing the different variants of primary progressive aphasia (PPA) is challenging, but more accurate characterization can improve patient management and treatment outcomes. This study aimed to identify the following: (1) which speech features, alone or combined with language assessment and gray matter volumes (GMVs), best distinguish PPA variants and (2) how connected speech evolves in PPA.

Methods: This prospective study was conducted at IRCCS San Raffaele Hospital in Milan, Italy, between 2010 and 2021. We included patients with PPA who underwent neuropsychological assessments, including standard evaluation of language and the "Picnic Scene" speech test, and, when available, brain structural MRI. Clinical and language assessments were also performed at follow-up in a subgroup. Sequential feature selection models identified speech parameters that best differentiated groups, incorporating age, sex, education, standard language tests, and GMVs. In each PPA group, linear mixed-effect models analyzed speech changes over time.

Results: We included 95 patients with PPA (mean age 69 ± 9 years, 55 women [58%]; 40 with nonfluent variant PPA [nfvPPA], 35 with semantic variant PPA [svPPA], 20 with logopenic variant PPA [lvPPA]), of whom 82 underwent brain MRI and 34 had a follow-up visit after 10.2 months. Each model distinguished svPPA from the other PPA groups with high accuracy (R² range 0.93-1.00; p < 0.001). No differences in accuracy were observed among models for this distinction. In differentiating nfvPPA and lvPPA groups, the models incorporating speech parameters (R² = 0.92; p < 0.001), GMVs (R² = 0.95; p < 0.001), and their combination (speech + GMVs; R² = 0.97; p < 0.001) outperformed those using only standard language scores (R² = 0.75; p = 0.01). Over time, patients with nfvPPA showed more phonological errors, the svPPA group exhibited more semantic and morphosyntactic errors along with difficulties in naming and syntax production, and patients with lvPPA exhibited reduced number of words per second and fewer words per sentence.

Discussion: All models were equally effective in distinguishing the svPPA group from the other 2 PPA subtypes. However, compared with using standard measures alone, incorporating speech measures from the "Picnic Scene" speech test, GMVs, or their combination into the models significantly improved accuracy in differentiating nfvPPA and lvPPA groups. The PPA variants showed distinct speech trajectories. These variables can aid in understanding disease progression, predicting patient outcomes, and planning speech therapy interventions in clinical practice.