Clinical prediction rules for cognitive outcomes post-stroke: an updated systematic review and meta-analysis

Area of Science:

• Neurology
• Epidemiology
• Biostatistics

Background:

• Stroke survivors have an increased risk of cognitive syndromes, including dementia and delirium.
• Early identification of at-risk individuals is crucial for effective clinical management and risk reduction.
• This study systematically reviews and appraises the prognostic accuracy of multicomponent risk models for post-stroke cognitive syndromes.

Purpose of the Study:

• To update and evaluate the evidence on the prognostic accuracy of multicomponent risk prediction models for post-stroke cognitive syndromes.
• To identify existing models and assess their performance in predicting cognitive impairment, dementia, or delirium after stroke.
• To provide an evidence base for the development and implementation of effective risk prediction tools.

Main Methods:

• An updated systematic review and meta-analysis of multidisciplinary electronic databases (November 2019 - May 2025).
• Inclusion criteria: multicomponent risk prediction tools developed in stroke populations (≥18 years), free of baseline cognitive impairment, reporting discriminative performance metrics.
• Risk of bias assessed using PROBAST, certainty of evidence by GRADE, heterogeneity via I-squared statistics. Study preregistered with PROSPERO.

Main Results:

• 20 new studies contributed 31 models for cognitive impairment/dementia and 6 for delirium; most models developed in Asia.
• Pooled c-statistics were 0.81 for cognitive impairment and 0.85 for delirium. External and temporal validation showed good performance (C-statistic: 0.72-0.91, AUC: 0.81-0.82).
• Most studies had low risk of bias, but overall certainty of evidence was low. Development cohorts were small, with limited assessment of model transportability.

Conclusions:

• The development of risk models for post-stroke cognitive syndromes has increased, with promising prognostic accuracy.
• Challenges remain, including small development cohorts, limited external validation, and a need for data pooling and utilization of large datasets.
• Further research focusing on model transportability, stakeholder engagement, and cost-effectiveness is necessary for clinical implementation.