Key points are not available for this paper at this time.
Background: Guillain-Barré syndrome (GBS) constitutes an immune-mediated inflammatory polyradiculoneuropathy. Stroke may coexist with GBS during the same clinical episode, but the associated clinical predictors remain insufficiently characterized. The present investigation therefore sought to construct a machine learning-based model for identifying concomitant stroke in patients with GBS. Methods: This retrospective cohort study included 260 patients with GBS who received care at the Second Affiliated Hospital of Army Medical University from January 1, 2015, to December 31, 2024. All candidate predictors were collected at admission. Feature selection was conducted using LASSO regression, and seven machine learning algorithms were developed and compared. An independent external validation cohort of 60 patients was obtained from the First Affiliated Hospital during the same period. Patients were subsequently grouped according to model-estimated probabilities, and short-term functional outcomes were compared between groups. Results: Nine clinical predictors were selected to construct seven machine learning models. The neural network architecture exhibited the best performance for identifying concomitant stroke. Internal validation yielded an AUROC of 0.838 (95% CI: 0.739-0.923) for the optimal model. Sensitivity analysis excluding patients with documented prior stroke showed comparable performance. For all outcome measures, time displayed a substantial primary impact (p 0.05). Conclusion: The ANN model showed good performance for admission-time identification of concomitant stroke in patients with GBS and distinguished clinically different functional profiles among model-defined groups. Notably, while the longitudinal interaction between temporal factors and assigned risk strata did not achieve statistical significance, the stratification methodology successfully discerned clinically distinct outcome profiles in GBS.
Zhou et al. (Thu,) studied this question.