Development and Validation of a Prediction Model of In-Hospital Neurological Deterioration for Patients with Minor Acute Ischemic Stroke
Objective To develop a prediction model of in-hospital neurological deterioration for patients with minor acute ischemic stroke(AIS),and to provide scientific basis for stratified in-hospital management.Methods Patients with minor AIS(defined as NIHSS score≤5)enrolled in the China national stroke registry Ⅲ(CNSR Ⅲ)and arriving within 24 hours from onset while without taking rt-PA intravenous thrombolysis or endovascular treatment were selected as the study subjects.The derivation cohort was consisted of 2256 patients enrolled from 2015 to 2016,and the validation cohort was consisted of 1775 patients enrolled from 2017 to 2018.The predictors were finally determined by LASSO regression and reviewing of previous studies.In-hospital neurological deterioration was defined as 4 points or more increase in NIHSS score at discharge compared with the NIHSS score at admission.A logistic regression model was used to develop the prediction model.Discrimination and calibration were evaluated using C statistic and the Brier score,respectively.Results A total of 4031 patients were included in the study,with 58(2.6%)of 2256 patients from the derivation cohort and 63(3.5%)of 1775 patients from the validation cohort encountered in-hospital neurological deterioration.The population characteristics were similar between the two cohorts.The prediction model was developed based on 9 predictors,including age,gender,smoking,systolic blood pressure,IL-6,hs-CRP,NIHSS score on admission,diabetes mellitus and infarction pattern.The C statistic for the model was 0.69(95%CI 0.62-0.76)in the derivation cohort and 0.73(95%CI 0.67-0.80)in the validation cohort.The Brier score of the model was 0.025 in the derivation cohort and 0.033 in the validation cohort.Conclusions This study developed a prediction model for the risk of in-hospital neurological deterioration for patients with minor AIS based on routine hospitalization data,and the prediction model achieved acceptable levels of discrimination and calibration,yet the extrapolation needs to be further verified by external data.
国家科技期刊平台
NSTL
万方数据
