Development and Validation of a GULP-Based Predictive Model for Dehydration in Elderly Patients with Post-Stroke Dysphagia.

The background for establishing and verifying a dehydration prediction model for elderly patients with post-stroke dysphagia (PSD) based on General Utility for Latent Process (GULP) is as follows: For elderly patients with PSD, GULP technology is utilized to build a dehydration prediction model. This aims to improve the accuracy of dehydration risk assessment and provide clinical intervention, thereby offering a scientific basis and enhancing patient prognosis. This research highlights the innovative application of GULP technology in constructing complex medical prediction models and addresses the special health needs of elderly stroke patients. Based on GULP criteria, the study aims to establish and validate a dehydration prediction model for elderly patients with dysphagia following a stroke. Two hundred patients with post-stroke dysphagia treated at Beijing Rehabilitation Hospital Affiliated with Capital Medical University, from January 2020 to December 2023, were selected retrospectively. The patients were randomly matched at a ratio of 1:4 to establish a verification group (n = 40) and a modelling group (n = 160). Based on the occurrence of dehydration, the modelling group patients were divided into two groups: the dehydration group (n = 55) and the non-dehydration group (n = 105). Univariate and multivariate logistic regression analyses were used to identify the influencing factors of dehydration in elderly patients with dysphagia after a stroke, and to establish a predictive model based on GULP. The predictive value of the model was evaluated using receiver operating characteristic (ROC) curve analysis. The results of univariate and multivariate logistic regression analyses showed significant differences in age, lesion location, muscle strength grade, homocysteine (Hcy), and swallowing function score ( < 0.05). When these influencing factors were included in the model, the slope of the calibration curve in both the training set and the validation set was close to 1, indicating that the predicted dehydration risk was consistent with the actual risk. ROC analysis results revealed that in the training set, the model predicted dehydration in elderly post-stroke patients with dysphagia with an area under the curve (AUC) of 0.934, a standard error of 0.034, and a 95% confidence interval (CI) of 0.916 to 0.981. The optimal cutoff value was 0.78, yielding a sensitivity of 88.84% and a specificity of 90.00%. In the validation set, the AUC was 0.867 with a standard error of 0.025 and a 95% CI of 0.694 to 0.934. The optimal cutoff value here was 0.66, with a sensitivity of 80.16% and a specificity of 85.94%. This study successfully established and validated a GULP-based dehydration prediction model for elderly patients with dysphagia following a stroke, demonstrating high application value.