Date of Award

Summer 2021

Document Type

Open Access Dissertation

Department

Communication Sciences and Disorders

First Advisor

Julius Fridriksson

Abstract

It remains largely unclear what factors determine who responds to aphasia therapy and to what degree. The current study sought to ameliorate this issue by addressing three aims: 1) To identify baseline predictors that dissociate between therapy responders and nonresponders, 2) to identify predictors of degree of treated recovery in therapy responders, and 3) to examine the generalizability of predictors identified under Aims 1 and 2 in randomly selected subsamples of study participants.

Method: Stroke survivors (N = 102; 43 females; age = 60.5y +/- 11.0y) with chronic aphasia (>12m post-stroke) were recruited as part of a multisite trial. Participants underwent a comprehensive baseline assessment, including a detailed case-history interview, multiple cognitive-linguistic assessments, and a neuroimaging workup prior to undergoing 30 hours of language therapy. Baseline data was used to predict binary and continuous therapy response after therapy completion (T1), one-month (T2), six-months (T3), and one-year (T4) after therapy. Primary statistical analyses utilized LASSO binary and multivariate regression approaches for variable selection, classification, and prediction.

Results: Classification of participants as responders and nonresponders was most accurately predicted based on functional Magnetic Resonance Imaging (T1 accuracy = 77.0%), biographical (T2 accuracy = 76.3%) and neuropsychological predictors (T3 accuracy = 80.3%; T4 accuracy = 85.7%). Highest prediction accuracy for treated recovery as a continuous factor in therapy responders was achieved by models incorporating neuropsychological predictors (T1 R2 = .645), saturated (‘combined’) prediction models (T2 R2 = .664; T3 R2 = .629), and diffusion tensor imaging data (T4 R2 = .830). Aphasia severity, age, lesion to and functional/structural integrity of the temporoparietal junction emerged as consistent predictors across models and timepoints. Generalizability of predictors of binary therapy response varied between 84.3% and 100.0%, whereas generalizability of predictors of continuous therapy response varied between R2 = .629 and R2 = .925.

Conclusions: While these results largely fall in line with prior research findings, the current study incorporated the largest sample size to date and a comprehensive dataset of baseline data. To this end, the novelty of the current study primarily lies in the scientific rigor of the reported findings as compared to prior literature.

Share

COinS