Abstract Body: Background and Purpose: Dysphagia is a common complication of stroke that can lead to increased aspiration pneumonia and malnutrition, prolonged hospital stay, and can necessitate the placement of a gastrostomy tube. Early identification of these patients may lead to significant reductions in hospital stays. In this single center study, we demonstrate a scoring system to predict the likelihood of a gastrostomy tube.

Methods: Pooled clinical data of consecutive adult stroke patients from a single center from April - December 2023 were retrospectively collected. Patient baseline characteristics, demographics, imaging findings, and clinical assessments were assessed in regards to gastrostomy tube placement. Using a Lasso procedure, features with an independent association with gastrostomy tubes were found. An XGBoost model was used to predict placement of a gastrostomy tube to derive a final scoring model based on feature importance using Shapley values.

Results: Of the 635 patients screened, 74 (10.4%) underwent gastrostomy tube placement. Demographic and clinical gastrostomy placement predictors were identified and included Rosenbek scores 4-8, pharyngeal residue rating, oral cavity residue rating, Anterior loss on dysphagia assessment, hemorrhagic infarction type 1 (HT-1), National Institutes of Health Stroke Scale (NIHSS) > 10 after 24 hours, intraventricular or subarachnoid hemorrhage (IVH/SAH), cardioembolic etiology, NIHSS > 10 on presentation, and prior stroke. The comprehensive model unbiased estimated out-of-sample performance, using a 10-fold cross-validated area under the curve (AUC), was 0.81 [0.77-0.85]. The simplified scoring model performed similarly to the comprehensive model, although slightly higher than expected given a lack of an independent out-of-sample testing set due to data availability, with an AUC of 0.88 [0.80, 0.92]. This outperformed the previously published PEG score (0.86 [0.84-0.95]) and PRESS score (0.81 [0.75- 0.88]). However, a completely out-of-sample evaluation of the final scoring model remains necessary.

Conclusions: This scoring system provides clinically relevant predictive value to identify acute ischemic stroke patients who will ultimately require gastrostomy tubes as opposed to those who experience transient dysphagia and premature gastrostomy tube placement can be avoided. Future studies will be required to determine if this scoring system can result in reductions in hospital stays and other clinical outcomes.