Risk Models for Adverse Events in Microsurgery for Intracranial Unruptured Aneurysms.

Area of Science:

• Neurosurgery
• Medical Machine Learning
• Clinical Prediction Models

Background:

• Preventive treatment of unruptured intracranial aneurysms (UIAs) necessitates balancing treatment risks against expected benefits.
• Existing scores for UIA rupture/growth risk lack external validation for microsurgical treatment risks.
• Machine learning models can provide personalized risk assessments for UIA treatment based on patient and aneurysm characteristics.

Purpose of the Study:

• To develop and externally validate clinical prediction models for adverse outcomes following microsurgical treatment of UIAs.
• To assess the models' ability to predict poor neurological outcome, new sensorimotor deficits, and all-cause adverse events at hospital discharge.

Main Methods:

• Utilized data from 20 international centers (3705 patients) for the "prediction of adverse events after microsurgery for intracranial unruptured aneurysms" study.
• Developed and validated models using machine learning on patient and aneurysm characteristics.
• Externally validated models on data from 7 centers (824 patients).

Main Results:

• Models demonstrated good calibration and discrimination for predicting poor neurological outcome (AUC=0.70) and new sensorimotor deficits (AUC=0.69).
• Prediction of all-cause adverse events showed fair calibration with an AUC of 0.59.
• A web application integrating the prediction model is available at https://neurosurgery.shinyapps.io/PRAEMIUM/.

Conclusions:

• Developed models for predicting poor neurological outcome and new sensorimotor deficits show good calibration and fair discrimination upon multinational external validation.
• These models may aid in objectively estimating microsurgical treatment risks for UIAs.
• The predicted probabilities align well with real-world risks, suggesting clinical utility.