Clinically oriented dual-tier screening for post-stroke epilepsy with interpretable machine learning in a severely imbalanced cohort - Summary - MDSpire
Advertisement
Clinically oriented dual-tier screening for post-stroke epilepsy with interpretable machine learning in a severely imbalanced cohort
To develop a dual-tier screening framework for post-stroke epilepsy that addresses severe class imbalance and enhances clinical interpretability, thereby improving early identification.
Key Findings:
Marked class imbalance of approximately 21.9:1 in the cohort.
Primary model achieved macro-AUC of 0.996, precision-recall AUC of 0.970, F1-score of 0.931, sensitivity of 0.907, and specificity of 0.998.
Secondary model yielded sensitivity of 0.971 but lower F1-score of 0.854, indicating its role as a high-sensitivity tool.
Interpretation:
The dual-tier framework effectively balances the need for risk stratification and high sensitivity in detecting post-stroke epilepsy, aiding clinical decision-making and follow-up care.
Limitations:
External validation of the models is necessary to confirm findings.
The retrospective nature may introduce biases inherent to observational studies, potentially affecting generalizability.
Conclusion:
The proposed dual-tier screening framework offers a promising approach for identifying post-stroke epilepsy in a clinically meaningful way, despite the challenges posed by class imbalance.
