Construction Of Prediction Model For Rehabilitation Of Swallowing Disorders In Ischemic Stroke Patients

Background and Purpose:Stroke has five characteristics:high morbidity,high disability,high mortality,high recurrence and high economic burden.In 2019,the number of stroke cases in China was as high as 28.76 million,resulting in 2.18 million deaths.Swallowing disorders are a common complication after stroke,with an incidence of 46.3%in the acute phase of stroke,and approximately 15%of patients experience long-term swallowing disorders.The Chinese Handbook of Stroke Swallowing Disorders and Nutritional Management suggests significant heterogeneity in the swallowing rehabilitation trajectory of patients with post-stroke swallowing disorders.Currently,there is a lack of systematic methods and tools to predict the outcome of swallowing rehabilitation in patients.Early assessment of patient prognosis is essential for health care providers to develop better treatment care plans prospectively.It is critical to develop tools that can accurately predict the risk of prognostic difficulties in patients.Therefore,this study aims to construct a predictive model for the rehabilitation of swallowing disorders after ischemic stroke based on evidence-based medical evidence and real-world data.The study performs internal and external validation to ensure the model’s discrimination,calibration,and clinical utility value.Materials and Methods:This study conducted a comprehensive review of domestic and international studies related to post-stroke swallowing rehabilitation using evidence-based medical Meta-analysis.After literature screening and quality evaluation,data were extracted and quantitatively analyzed to obtain important factors influencing post-stroke swallowing rehabilitation.Based on the results of the Meta-analysis and consultation with clinical experts,a database of clinical information of patients with swallowing disorders in ischemic stroke was constructed.To construct prediction models,a retrospective study design was used,including patients who received nasogastric tube feeding due to swallowing disorder after ischemic stroke attending the inpatient neurology ward of a tertiary care hospital in Chengdu from May 2018 to April 2022 as the modeling set.They were divided into a difficult prognosis group and a good prognosis group based on different swallowing function recovery.Feature screening was performed using the minimal absolute contraction and selection operator Lasso regression,and prediction models were constructed by binary logistic regression analysis,plotted nomogram and developed web calculators.Internal validation of the model was performed using the bootstrap method.A prospective study design was used to externally validate the model,including inpatients who met the same inclusion exclusion criteria from May 2022 to December2022 as the external validation set.C-statistics,receiver operating characteristic(ROC)curves,Hosmer-Lemeshow test,calibration curve,and clinical decision curve analysis(DCA)were used to assess the discrimination,calibration,and clinical utility value of the model.Result:Initially,a total of 1023 papers were collected by searching Chinese and English databases.After literature screening and quality evaluation,only 20 papers were deemed suitable for inclusion.These papers reported on a total sample size of 3845 cases.The meta-analysis revealed several influential factors for swallowing rehabilitation in stroke patients.These included age,aspiration,degree of initial dysphagia,early intervention,cognitive impairment,nutritional disorders,National Institute of Health Stroke Scale(NIHSS),Modified Rankin Scale(MRS)and bilateral stroke.In this study,287 patients were included in the modeling dataset,of which 111(38.68%)had a poor prognosis for dysphagia.These patients had a mean length of stay of(18.74±11.57)days.Lasso-logistic regression analysis revealed that seven variables entered the mathematical model.The columnar graphical models indicated that several factors were predictors of prognostic difficulties in swallowing disorders,including age>70 years[OR(95%CI)=4.861(2.323,10.172),P<0.001],number of strokes>2[OR(95%CI)=5.645(1.634,19.504),P=0.006],previous hypertension[OR(95%CI)=11.647(5.227,25.950),P<0.001],electrolyte disturbances[OR(95%CI)=7.196(3.531,14.666),P<0.001],tracheal intubation[OR(95%CI)=2.879(1.146,7.234),P=0.024],NIHSS score at admission[OR(95%CI)=6.236(1.111,34.990),P=0.038],and Barthel index at admission[OR(95%CI)=0.059(0.007,0.469),P=0.007].The prediction model’s C-Statistic was 0.879(95%CI,0.839-0.918),and the Hosmer-Lemeshow test yieldedχ~2=3.620,P=0.306,indicating a well-fitted calibration plot.The Bootstrap test resulted in a C-Statistic of 0.878(95%CI,0.877-0.880).At a significance level of 0.05,the maximum Jorden index was 0.657,with a model sensitivity of 0.801,model specificity of 0.856,and correct prediction rate of 79.8%.For external validation,84 patients were included,yielding a C-Statistic of 0.789(95%CI,0.683-0.890)and a Hosmer-Lemeshow test result ofχ~2=4.650,P=0.200,indicating a well-fitted calibration plot.The decision curve analysis(DCA)suggested that the prediction model constructed using this study could achieve good net clinical benefit within a relatively large range of threshold probabilities(0.06-0.81 for the training set and0.11-0.75 for the external validation set).Conclusion:The prediction model developed in this study includes seven predictors that demonstrated good discrimination,calibration,and clinical utility values.This model provides a visual aid for accurately identifying individuals at high risk of developing prognostic difficulties in PISD,offering a scientific foundation for precise care,and improving the economic effectiveness of medical investments.