2073 From Data to Decision: A Multicenter, Externally Validated AI Tool for Managing Infectious Intracranial Aneurysms

View at publisher Summarise with VIRAT
Authors: Ali Mostafa Alawieh, Youssef M. Zohdy, Andrew B. Koo, Hassan Saad, Jonathan A. Grossberg, Charles C. Matouk, C. Michael Cawley, Gustavo Pradilla, Andrew Reisner, Chris Fox, Carlos Perez‐Vega, Jan-Karl Burkhardt
+12 more Mohamed M. Salem, Pascal Jabbour, Kareem Elnaamani, Robert M. Starke, Ahmed Abdelsalam, Richard F. Schmidt, M. Reid Gooch, Victor M. Lu, Michael R Levitt, Alejandro M. Spiotta, Daniel Louis Barrow, Brian M. Howard
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

INTRODUCTION: Infectious intracranial aneurysms (IIAs) are rare but high-risk cerebrovascular lesions that require multidisciplinary physician teams and lack consensus on management. Recent data show that nearly 50% of patients fail antibiotic therapy, often resulting in hemorrhage or requiring surgical intervention. Accurately identifying patients at risk of treatment failure is critical for guiding early intervention and improving outcomes. METHODS: A retrospective cohort of 104 patients with 166 IIAs was compiled across 11 tertiary centers (2018–2023). Patients managed medically were used to train seven classifiers—including random forests, support vector machines, and neural networks—using clinical and radiographic predictors. Model performance was assessed using AUC, PPV, and NPV. The best-performing model was externally validated using published individual patient data curated from a pooled analysis of 1191 patients. The final model was deployed as a clinical decision tool. RESULTS: Medical management failed in 44% of cases, with a median failure time of 18 days. Random forest models best predicted antibiotic failure, with an AUC of 89.5% and PPV of 88% internally. Naïve Bayes best predicted mortality (AUC 87.5%, PPV 93%). External validation showed an AUC of 74% and PPV of 91% for failure of medical management. Most predictive variables included aneurysm rupture, size, and patient age. The model was implemented in an open-access tool to aid treatment stratification. CONCLUSIONS: This study introduces a validated, high-performing machine learning model capable of predicting failure of antibiotic management in IIAs. The tool offers physicians a real-time, evidence-based method to identify high-risk patients, supporting timely escalation to surgical or endovascular therapy and transforming IIA care from reactive to predictive.

Fields of Study & Subjects

Intracranial Aneurysms: Treatment and Complications Infectious Aortic and Vascular Conditions Bacterial Infections and Vaccines

Publication details

Journal / source
Neurosurgery
Volume
72
Issue
Supplement_1
Pages
262–262
Publication date
26 March 2026
Language
EN