Jia Shang,1,2 Zehao Zhang,1 Shifang Ma,1 Hailong Peng,1 Lan Hou,1,3 Fan Yang,1 Pei Wang1,3 

1Department of Neurology, Baoding No 1 Central Hospital, Baoding, People’s Republic of China; 2Graduate School of Hebei Medical University, Shijiazhuang,People’s Republic of China; 3Department of Neurology, Key Laboratory of Neurological Diseases, Baoding, People’s Republic of China

Correspondence: Pei Wang, Department of Neurology, Baoding No 1 Central Hospital, Baoding Great Wall North Street No. 320, Baoding, Hebei Province, 071000, People’s Republic of China, Email w_z_h_01@163.com

Purpose: Early neurological deterioration (END) frequently complicates acute ischemic stroke (AIS), worsening prognosis, particularly in patients with type 2 diabetes mellitus (T2DM), where hyperglycemia accelerates atherosclerosis, increasing both stroke risk and subsequent END. This study aimed to identify predictors of END in minor stroke patients with T2DM and develop a nomogram integrating these factors with intracranial atherosclerosis (ICAS) scores, evaluating its performance against various machine learning (ML) models.
Methods: We retrospectively analyzed clinical data from 473 minor stroke patients with T2DM treated at our hospital between January 2021 and December 2023. Utilizing LASSO and multivariate logistic regression, we identified characteristic predictors. The cohort was randomly allocated into training (n = 331) and validation (n = 142) groups. Six ML algorithms—SVM, LR, RF, CART, KNN, and Naive Bayes—were assessed, and nomograms were used to visualize the predictive model’s performance, evaluated via Area Under the Curve (AUC), calibration plot, and Decision Curve Analysis (DCA).
Results: The ICAS score has been recognized as a pivotal determinant of END, alongside four other significant factors: NIHSS score, low-density lipoprotein cholesterol (LDL-C) levels, presence of branch atheromatous disease (BAD), and stenosis of the responsible vessel ≥ 50%. The model demonstrated robust predictive capabilities, achieving strong performance in training (AUC = 0.795) and validation (AUC = 0.799) sets. This advanced ML model, which integrates biochemical and imaging indicators, enables accurate risk assessment for END in minor stroke patients with T2DM.
Conclusion: By integrating the ICAS score with the NIHSS score, LDL-C levels, presence of BAD, and stenosis of responsible vessels ≥ 50%, we developed a clinical model for predicting END in patients with minor stroke and T2DM. This model provides critical decision support for clinicians, facilitating early identification of high-risk patients, personalized treatment, and improved outcomes.

Keywords: early neurological deterioration, prediction model, type 2 diabetes mellitus, machine learning, nomograms