Post-Stroke Epilepsy Prediction with Interpretable ML in a 21,459-Patient Cohort

Lijun Wu's research dataset from 2026 contains a retrospective cohort of 21,459 stroke patients, including 936 who developed post-stroke epilepsy. It was used to develop a dual-tier, interpretable machine learning framework for screening in a severely imbalanced clinical setting. The dataset supports models achieving high performance metrics, such as an area under the precision-recall curve of 0.970.

Use Cases

• Developing imbalance-aware prediction models based on the described cohort of 21,459 stroke patients.
• Implementing a dual-tier clinical screening framework based on the primary and secondary model strategy described.
• Interpreting model predictions for post-stroke epilepsy based on key contributors like neurological severity and D-dimer levels mentioned.
• Benchmarking model performance on imbalanced medical data based on reported metrics like sensitivity of 0.971 and specificity of 0.998.

Strengths

• Dataset is derived from a substantial cohort of 21,459 patients with stroke.
• Includes 936 confirmed cases of post-stroke epilepsy, providing a basis for modeling a rare outcome.
• Research describes specific, high-performing model results, including a macro-AUC of 0.996.
• Framework addresses the specific challenge of severe class imbalance, noted as approximately 21.9:1.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the underlying data tables is unknown, which may limit suitability assessment.
• Data is presented in a DOCX file format (21.3 KB), suggesting limited raw data scope and potential preprocessing requirements.

Provenance

Source

figshare, author Lijun Wu.

Collection Method

Retrospective cohort study, as described in the background.

Freshness

Last updated 2026-05-21 04:19:22; freshness should be verified.