Prognostic Modeling Based on Post-Endovascular Thrombectomy Systolic Blood Pressure Trajectories Using Explainable Artificial Intelligence: A Secondary Analysis of the OPTIMAL-BP Trial
Clinical Snapshot
PICO Framework
| P — Population | Adults who underwent successful endovascular thrombectomy for acute ischemic stroke (n=288, median age 75 years, 61.1% male) |
| I — Intervention | Machine learning models incorporating systolic blood pressure metrics with clinical variables |
| C — Comparator | Machine learning models using clinical variables only |
| O — Outcomes | Functional independence at 90 days (modified Rankin Scale 0-2), model performance measured by area under the curve |
Bottom Line
This secondary analysis of the OPTIMAL-BP trial demonstrates that incorporating systolic blood pressure trajectory data significantly improves machine learning prediction of functional outcomes after successful endovascular thrombectomy. The enhanced model achieved an AUC of 0.86 compared to 0.80 for clinical variables alone (P=0.037). Explainable AI identified time rate of blood pressure change and minimum systolic pressure as key predictors, with differential importance between intensive and conventional management groups. While the methodology is sound and results clinically meaningful, the relatively small sample size and lack of external validation limit immediate clinical implementation. The findings support the importance of detailed blood pressure monitoring post-EVT and suggest potential for personalised management approaches, though larger validation studies and cost-effectiveness analyses are needed before routine clinical adoption.
Key Findings
P Value: P = 0.037 for model comparison
Effect Size: AUC improvement from 0.80 to 0.86 when adding SBP metrics to clinical variables
Primary Outcome: Functional independence at 90 days (mRS 0-2)
Nnt Or Sensitivity: Time rate of SBP and minimum SBP identified as most important predictive features via SHAP analysis
Confidence Interval: 95% CI 0.76-0.92 for enhanced model vs 0.69-0.88 for clinical variables only
Clinical Application
Requires continuous blood pressure monitoring and computational infrastructure for real-time prediction Relevant to Australian comprehensive stroke units with EVT capability; aligns with National Stroke Foundation guidelines for post-EVT care monitoring Adult stroke patients undergoing successful endovascular thrombectomy in comprehensive stroke centres
Abstract
Blood pressure (BP) management following successful reperfusion after endovascular thrombectomy (EVT) is critical in achieving favorable clinical outcomes. Individualized BP management using predictive modeling by machine learning may further improve prediction of functional outcomes. This study was a retrospective analysis of data from the Outcome in Patients Treated with Intra-Arterial Thrombectomy-Optimal Blood Pressure Control (OPTIMAL-BP) trial, a randomized controlled trial comparing between intensive and conventional BP management during the 24 h after successful recanalization by EVT from June 18, 2020, to November 28, 2022. The trial was conducted across 19 centers in South Korea. Machine learning models were developed to predict functional independence (90-day modified Rankin Scale 0 to 2). Model performance was compared between clinical variables only and systolic blood pressure (SBP) metrics in addition to clinical variables. In addition, the Shapley additive explanations (SHAP) analysis was performed to provide model explanation and understand the importance of SBP metrics. A total of 288 patients (61.1% men, median age 75 years [interquartile range, 65-81]) were included. Among the six algorithms, the deep neural network model incorporating SBP metrics performed best on validation, achieving an area under the curve of 0.86 (95% confidence interval, 0.76-0.92) which was significantly better than the model using only clinical variables (area under the curve 0.80 [95% confidence interval, 0.69-0.88], P = .037). Among SBP metrics, SHAP analysis identified time rate of SBP and minimum SBP as important features, with time rate showing greater influence in the intensive group and minimum SBP in the conventional group. Integrating SBP metrics with clinical variables significantly improved machine learning performance in predicting functional outcomes after successful EVT. Explainable artificial intelligence (AI) identified time rate and minimum SBP as key predictors of outcome. Trial Registration Information: ClinicalTrials.gov (NCT04205305; registered December 17, 2019).
References
- 1.Yu, R., Heo, J., Park, E., Joo, H., Jung, J. W., Kim, K. H., Yun, J., Lee, H., Choi, J. K., Lee, I. H., Lim, I. H., Hong, S.-H., Baik, M., Kim, B. M., Kim, D. J., Shin, N.-Y., Cho, B.-H., Ahn, S. H., Park, H., Sohn, S.-I., Hong, J.-H., Song, T.-J., Chang, Y., Kim, G. S., Seo, K.-D., Lee, K., Chang, J. Y., Seo, J. H., Lee, S., Baek, J.-H., Cho, H.-J., Shin, D. H., Kim, J., Yoo, J., Jung, Y. H., Hwang, Y.-H., Kim, C. K., Kim, J. G., Lee, C. J., Park, S., Lee, H. S., Kwon, S. U., Bang, O. Y., Heo, J. H., Kim, Y. D., & Nam, H. S. (2026). Prognostic modeling based on post-endovascular thrombectomy systolic blood pressure trajectories using explainable artificial intelligence: A secondary analysis of the OPTIMAL-BP trial. *Journal of Medical Systems*. https://doi.org/10.1016/S1474-4422(23)00076-5
Related Research
Sleep medicine reviews
Machine and deep learning in REM sleep behavior disorder: a scoping review and analysis of reporting quality
3 Aug 2026
Neurological research
Federated deep learning model for epilepsy seizure detection using electroencephalogram (EEG) signal
2 Aug 2026
Journal of psychiatric research
Systematic review of machine learning and deep learning models for EEG-based detection of depression.
2 Aug 2026
This content is for educational purposes for healthcare professionals only and does not constitute clinical advice. Clinical decisions should be based on individual patient assessment, current guidelines, and appropriate specialist consultation. Editorial Standards · Privacy Policy · Terms of Service