Novel predictors and a predictive model of cerebrovascular atherosclerotic ischemic stroke based on clinical databases

ABSTRACT

Background and Purpose

Early identification of cerebrovascular atherosclerotic ischemic stroke is necessary for accurate treatment and clinical research.

Aims

To identify novel predictors and build a predictive model of ischemic strokes due to cerebrovascular atherosclerosis.

Method

MIMIC-IV database was used to search for clinical data of patients with ischemic stroke. Included patients were divided into two groups according to their etiologies. Univariate and multivariate logistic regressions were used to build the predictive model, and the model reliability parameters were calculated. The cut-off value for the model was selected according to the Youden index. Clinical data from the Neurovascular Center of Changhai Hospital were used to verify the predictive model.

Results

Logistical regressions showed a positive correlation between advanced age, peripheral atherosclerosis, history of transient ischemia, and the diagnosis of ischemic strokes due to cerebrovascular atherosclerosis. The history of atrial fibrillation, levels of the National Institutes of Health Stroke Scale, serum potassium, and activated partial thromboplastin time were negatively correlated to the diagnosis of cerebrovascular atherosclerotic ischemic stroke. The predictive model was constructed from logistic regression results, and the area under the curve was 0.764. The cut-off value for the model was set at 0.089 to achieve the highest Youden index, with sensitivity and specificity of 75.9% and 64.1%. Clinical verification of the model revealed that the sensitivity and specificity of the model were 52.5% and 93.0% respectively.

Conclusion

The efficacy of the predictive model was acceptable as an aid in predicting cerebrovascular atherosclerotic ischemic stroke.

KEYWORDS:

• Ischemic stroke
• cerebrovascular atherosclerosis
• MIMIC-IV
• predictive model

Disclosure statement

No potential conflict of interest was reported by the author(s).

Statements and declarations

All the data in this manuscript is accessible by contacting us.

Contribution statement

He Li composed the manuscript, Pei Liu collected the clinical data and modified the manuscript, Hong-Yu Ma performed the analysis with Python, Wei-Long Hua performed the analysis with Stata 15, Yong-Xin Zhang composed Tables 1–3, Lei Zhang composed Table 4 and supplementary table, Yong-Wei Zhang composed Figure 1, Bo Hong composed Figure 2, Peng-Fei Yang raised the idea and provided directions to He Li and Pei Liu, Jian-Min Liu raised the idea with Peng-fei Yang and composed the outline of discussion part.

Additional information

Funding

Our work is funded by the National Natural Science Foundation of China numbered 82071278, and Medical Health Science and Technology Project of Zhoushan City numbered 2022JRC01.