Construction of a Prediction Model for Naming Recovery in Subacute Poststroke Aphasia Based on Multivariate Analysis: Evaluation of Accuracy and Reliability.

Authors

Zhifang Sun, Linlin Ye, Yi Shan, Lei Cao, Bixiao Cui, Jie Wang, Dahua Zhang, Yinan Cheng, Tiantian Zhang, Yuting Zhao, Weiqun Song, Jie Lu

Published in

Journal of speech, language, and hearing research : JSLHR. Pages 1-20. Jul 08, 2026. Epub Jul 08, 2026.

Abstract

Anomia, a common dysfunction in poststroke aphasia (PSA), impacts daily communication, and its prognosis remains challenging. This study aimed to develop a model to predict naming rehabilitation in patients with subacute PSA.
Data of PSA were collected retrospectively. Logistic regression (LR) analyses by a nested fivefold cross-validation were performed to identify predictors and construct a predictive model. The efficacy of the model was evaluated using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). Additionally, the net reclassification index (NRI) and integrated discrimination improvement (IDI) were calculated to assess the discrimination of the model. Shapley additive explanation (SHAP) values were utilized to determine the prediction role of each feature in the model. To reduce the variance of the best model, data resampling was performed using fivefold cross-validation and three distinct random seeds.
A total of 199 patients and 21 clinical variables were analyzed. Lesion site, Broca's area damage, education level, Boston Diagnostic Aphasia Examination grade, and transcranial direct current stimulation (tDCS) therapy were significant in ≥ 50% of inner folds across three different random seed conditions in univariate LR analysis and were incorporated into the predictive model. The model achieved the mean area under the curve in the fivefold cross-validation, surpassing the single-predictor models (Seed 42: 0.792 ± 0.099; Seed 123: 0.828 ± 0.045; Seed 456: 0.816 ± 0.065). The ROC curves, calibration curves, and DCA demonstrated high accuracy, consistency, and clinical utility in the prediction of rehabilitation of PSA. Furthermore, the NRI and IDI indicated that the model offered better discrimination than single-predictor models. The model showed high balanced accuracy (Seed 42: 0.892 ± 0.031; Seed 123: 0.874 ± 0.061; Seed 456: 0.883 ± 0.074). The SHAP analysis identified the importance of these features. Notably, tDCS, as the sole modifiable core factor, holds a prominent position in the predictive model, highlighting its potential therapeutic value in accelerating language functional recovery through the facilitation of neuroplasticity. However, its therapeutic efficacy is modulated by stimulation target status (Broca's area integrity), suggesting that alternative therapeutic targets may be explored for patients with Broca's area damage.
The predictive model demonstrated a strong performance and may assist clinicians in stratifying patients with aphasia for early intervention, thereby potentially improving rehabilitation outcomes.
https://doi.org/10.23641/asha.32774427.

PMID:
42418163
Bibliographic data and abstract were imported from PubMed on 08 Jul 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.