Prediction models for early neurological deterioration in patients with acute ischemic stroke: A systematic review and critical appraisal

Abstract

Background: Despite the proliferation of risk prediction models for early neurological deterioration (END) in patients with acute ischemic stroke (AIS), significant uncertainties persist regarding their methodological rigor and clinical applicability. Objective: To systematically review and critically evaluate published prediction models for END in patients with AIS. Methods: PubMed, Embase, Scopus, and the Cochrane Library were searched from inception to March 26, 2025. Data extraction was conducted using a standardized data extraction form by two independent reviewers based on the recommendations in the CHecklist for critical Appraisal and data extraction for systematic Reviews of prediction Modelling Studies(CHARMS). The Prediction model Risk Of Bias ASsessment Tool (PROBAST) checklist was used to assess the risk of bias and applicability. A qualitative synthesis was carried out to summarize the main characteristics of the included studies and constructed models. Results: A total of 3682 studies were retrieved, and 45 prediction models from 23 studies were included. Logistic regression and machine learning were utilized to establish END risk prediction models. The reported incidence of END in AIS patients varied from 6.6 % to 43.7 %, depending on the definition and study population. The most frequently used predictors were baseline National Institutes of Health Stroke Scale score and systolic blood pressure. The model's discrimination performance, quantified by the area under the curve or concordance statistic, showed remarkable heterogeneity in predictive accuracy across studies. Critically, all included studies were assessed as having a high risk of bias, mainly owing to inappropriate data sources and poor reporting of the analysis domain. Concerns regarding applicability were generally low across studies. Conclusions: This systematic review provides a comprehensive mapping and critical assessment of existing END prediction models in AIS. The findings reveal a critical gap that current models exhibit high risk of bias, limiting their reliability for clinical adoption. Future research should prioritize prospective model development and validation with pre-specified protocols, rigorous adherence to methodological standards such as the TRIPOD guidelines, adequate sample size estimations, robust external validation, as well as the update and clinical utility of existing predictive models.