Construction Of Fall Risk Prediction Model For Stroke Patients

ObjectivesTo investigate the current situation of stroke patients’ fall event,to explore the risk factors of stroke patients’ fall event,and to construct logistic regression and decision tree risk prediction models in order to facilitate the screening of stroke patients at high risk of fall at an early stage.MethodsIn this study,388 inpatients with stroke from December 2021 to October 2022 were collected from three grade-A tertiary hospitals in Gansu Province.Baseline general data and risk factors were assessed and collected using the stroke patient General Information Questionnaire,Modified Falls Efficacy Scale,Barthel Index rating scale,mini-mental state examination,Performance Oriented Mobility Assessment,functional ambulation classification,10-Meter Walk Test and biochemical indicators.Fall occurrence was investigated within 6 months after baseline using the fall Occurrence questionnaireThe risk factors for falls in stroke patients were screened by univariate analysis,and those with P<0.05 were included in the multivariate analysis to establish a logistic regression model;the decision tree model was constructed by using the decision tree(CHAID)algorithm for factors with P<0.05,and the accuracy,sensitivity,specificity,Youden,s index and area under ROC curve(AUC)were applied respectively to evaluate the two models.Ten-fold cross-validation was used to internally validate the models,and Z-test was used to compare the AUCs of the logistic model and the decision tree model to determine whether the differences were statistically significant.Results(1)A total of 388 stroke patients with a mean age of(64.27±11.30)years were included in this study.There were more male patients(73.7%)than female patients(26.3%).(2)The incidence of falls in stroke patients was 28.1%.Among them,2 falls occurred most frequently within 6 months,accounting for 42.2%.The incidence of fallrelated injuries in stroke patients was as high as 58.7%,and the most common injury after a fall was hematoma(18.3%),but only 36.7% of stroke patients took preventive measures after a fall.The most common cause of falls in stroke patients in this study was loss of balance,accounting for 44.9%;60.6% of stroke patients had a fall indoors.(3)Logistic regression analysis identified seven independent predictors of falls in stroke patients including: age(OR=1.035,95%CI: 1.000～1.072),gender(OR=2.382,95%CI: 1.199～4.733),history of stroke(OR=1.975,95%CI: 1.033～3.775),NIHSS scores≥5(OR=3.607,95%CI: 1.282～10.144),muscle strength(OR=4.201,95%CI:2.060～8.566),Holden walking function grade≥4(Grade 4: OR=0.061,95%CI:0.013～0.287;Grade 5: OR=0.082,95%CI: 0.013～0.511),the level of triglycerides(OR=0.698,95%CI: 0.493～0.987);The decision tree CHAID algorithm was used to construct a fall risk prediction model for stroke patients.The maximum number of layers in the output model was 3,with a total of 20 nodes,12 terminal nodes and 12 decision paths.A total of 7 explanatory variables were screened,including functional ambulation classification,muscle strength,age,the level of triglycerides,BI scores,history of stroke and 10-meter walking function test.(4)The decision tree prediction model had a higher AUC(0.911),accuracy(86.1%),and specificity(0.876)than the logistic regression prediction model AUC(0.883),accuracy(82.7%),and specificity(0.835).However,the sensitivity(0.853)and the Youden,s index(0.688)of the logistic regression model were higher than those of the decision tree prediction model(0.809)and the Youden,s index(0.685).Comparing the AUC of the logistic regression prediction model with that of the decision tree prediction model(Z=1.712,P=0.028)indicated that the AUC of the decision tree model was greater than that of the logistic regression model,and the difference was statistically significant.Conclusions(1)Stroke patients have a high incidence of falls and serious injuries after falls.Targeted monitoring of stroke patients should be strengthened and interventions should be developed early based on risk factors.(2)Age,gender,history of stroke,NIHSS score≥5,muscle strength,functional ambulation classification ≥ 4,the level of triglycerides,BI scores,and 10-meter walking function test are independently associated with falls in stroke patients.In addition,these factors are entry points for clinical development of relative preventive measures.(3)Both the logistic regression prediction model and the decision tree prediction model constructed in this study showed good prediction accuracy.However,the decision tree model is better than the logistic regression model and has more clinical utility.