Predicting Antibiotic Failure in Infectious Intracranial Aneurysms: A Multicenter Study and Ai-based Model

Introduction: Infectious intracranial aneurysms (IIAs) are rare but serious complications of systemic infections. Data on the management for ruptured and unruptured IIAs remain limited. This multicenter study aimed to identify predictors of antibiotic failure in IIAs treatment and to develop an artificial intelligence (AI)-based model to guide treatment decisions.

Methods: We conducted a multicenter cohort study across 11 tertiary care centers in the United States from 6/2018 to 6/2023. Data on patient presentation, aneurysm characteristics, and treatment outcomes were collected. Treatment failure was defined as aneurysm growth, recurrence, or rupture within 3 months. Secondary outcomes included the 90-day modified Rankin score (mRS) and 90-day mortality. Propensity score matching (PSM) was performed to compare outcomes between patients treated with antibiotics alone versus surgical intervention. Various AI models were trained using 70% of the data to predict treatment failure and mortality, with external validation conducted using an independent dataset.

Results: We included 104 patients with IIAs (median age 44 years, 68% presenting with rupture). Medical management was the primary approach for 57%, with 29% undergoing endovascular and 13% undergoing microsurgical treatments. Predictors of antibiotic failure included age (aOR=0.97, p< 0.05), hypertension (aOR = 2.63, p < 0.05), ruptured presentation (aOR=2.99, p< 0.05), and size (aOR=1.13, p< 0.05). After PSM, medical management was associated with higher treatment failure within 6 weeks (Hazard Ratio (HR)=4.7, p< 0.001), and to die within 1 year (HR=4.4, P=0.04). Random forests demonstrated the best performance in predicting treatment failure, with area under the curve (AUC) of 89.5% and a positive predictive value (PPV) of 88%. For 90-day mortality prediction, the naïve Bayes performed best (AUC=87.5%, PPV=93%).

Conclusion : This study identifies critical predictors of antibiotic failure in IIAs and highlights the potential of AI-based models to inform early surgical intervention. An online decision support tool was developed for clinical use.