| BackgroundStroke is the leading cause of adult disability and death in China.Early neurological deterioration(END)is an important factor for poor outcomes of stroke patients.Once END occurs,the patient’s neurological function recovery is poor and the mortality rate is higher in the short term.The probability of disability in patients with END is 35 times that of normal patients,and the probability of death is 5.8times of normal patients.Therefore,early identification of patients in high-risk of END and prevention are particularly important.However,there is still a lack of accurate and effective prediction tools for clinical nurses to predict the occurrence of END,and it is difficult to carry out effective early intervention measures.Therefore,there is an urgent need for an effective END predictive scoring tool to help nurses assess the patient’s condition and formulate corresponding care plans.Objective1.Explore the predictive power of IScore score on END.Risk factors and IScore score were combined to construct END prediction model,and internal validation of the model was conducted.2.To explore the independent risk factors of END in patients with acute ischemic stroke(AIS)and construct a predictive model for END.External validation of the predictive model in a prospective cohort was performed.Methods1.We retrospective collected data of AIS patients hospitalized in the Department of Neurology,Shenzhen Second People’s Hospital from January 2014 to December2018,IScore scores and END-related risk factors were collected on the study subjects,and the effectiveness of IScore scores for predicting END was discussed.We construct a predictive model of END by combined IScore score with the independent risk of END factors.and internal validation was conducted.2.We retrospective collected data of patients with AIS who were admitted to a tertiary hospital in Shenzhen,China between January 2014 and December 2018.Baseline clinical,laboratory,and neuroimaging variables were selected to construct predictive models through multivariate logistic regression.According to whether patients had END,they were divided into END group and non-END group.The risk factors of the two groups were compared through Univariate analysis,and independent risk factors were screened by multivariate analysis.We use R software to build a nomogram and draw a receiver operating characteristic(ROC)curve of the model.3.We prospectively collect AIS patients who were hospitalized in the Department of Neurology,Shenzhen Second People’s Hospital from March 2019 to December 2020as the validation group of the END prediction model.According to whether the patients had END,they were divided into case group and control group,and the risk factors of the two groups were compared.The ROC curve was representing the discrimination of the model.Hosmer-Lemeshow(H-L)test and the calibration curve graph was calculated for the consistency of the representative model for the calibration ability of the model.Results1.A retrospective analysis of the clinical data of 452 patients with AIS,Univariate analysis showed that the IScore score,atrial fibrillation history,admission blood glucose,white blood cells,MCA stenosis,and carotid artery stenosis of the END group and the non-END group were statistically significant(all P value less than 0.05).The logistic regression analysis showed that IScore score[OR=1.027(95%CI=1.012,1.043),P<0.001),white blood cell[OR=1.207(95%CI:1.089,1.336,P<0.001],MCA stenosis[OR=5.604(95%CI:1.906,16.475,P<0.002],carotid artery stenosis≥50%[OR=2.583(95%CI:1.142,5.84)P=0.023]are independent risks of END Factors.Combine IScore score and END independent risk factors to construct a model:Logit(END)=-5.586+0.027*IScore score+0.188*white blood cells+1.723*MCA stenosis+0.949*carotid artery stenosis≥50%,model AUC is0.790(95%CI:0.729,0.852),sensitivity is 0.635,specificity is 0.834,accuracy is 0.798,and the best threshold is-1.257.Bootstrap repeated sampling method(number of samplings=500)AUC is 0.782(95%CI:0.717,0.840).2.A total of 391 subjects were included in the Derivation cohort,of which 64patients had END.Regression analysis showed that the baseline NIHSS score(OR=1.267,P<0.001),MCA stenosis(OR=3.588,P=0.036),and carotid artery stenosis≥50%(OR=3.122,P=0.020)were independently risk factor for END in AIS patients.Through the method of model comparison,the final END model is:logit(END)=-3.406+0.251*baseline NIHSS score+1.677*MCA stenosis+1.159*carotid artery stenosis≥50%,AUC is 0.844(95%CI:0.788,0.901),the threshold is-1.570;the specificity is 84.40%,the sensitivity is 75.00%,the positive predictive value is 48.48%,and the negative predictive value is 94.52%.3.We prospectively enroll 155 patients as the Validation cohort,including 129 in the control group and 26 in the case group.There was no significant difference in age(t=-1.364,P=0.212)and gender(χ~2=1.981,P=0.159)between the two groups.The baseline NIHSS score(Z=-3.526,P<0.01),MCA stenosis(χ~2=4.781,P=0.029)and carotid artery stenosis≥50%(χ~2=8.873,P=0.003)were statistically different Significant level(P<0.05).The prediction model has an AUC value of 0.753(95%CI:0.629,0.864)in the validation population,and there is a good agreement between the predicted probability and the actual probability.The calibration graph reflects the good agreement between the predicted END probability of the model and the actual observed END probability,and the H-L test P=0.195.Conclusion1.The END prediction model based on IScore score,which included Baseline IScore score,white blood cell,MCA stenosis and carotid artery stenosis≥50%has good prediction performance.Internal verification shows that the model has good stability.2.The END prediction model,which included Baseline NIHSS score,MCA stenosis and carotid artery stenosis≥50%,has good predictive performance.and it was expected to become an important tool for AIS patients to predict END,and it can also be used as a basis for individualized prevention programs.3.The END prediction model has a good degree of discrimination and calibration in the verification population of this study. |
PDF Full Text Request