Predicting neurosurgical referral outcomes in patients with chronic subdural hematomas using machine learning...

Area of Science:

• Neurosurgery
• Medical Informatics
• Artificial Intelligence

Background:

• Increasing incidence of chronic subdural hematoma (CSDH) and neurosurgery referral rates necessitate efficient triage tools.
• Current referral processes lack automated, evidence-based decision support for CSDH cases.
• Developing reliable machine learning (ML) models is crucial for predicting CSDH referral outcomes.

Purpose of the Study:

• To assess the feasibility of using ML algorithms to predict the outcome of CSDH referrals to neurosurgery.
• To evaluate the reliability of these ML models through external validation.
• To explore the potential for ML-driven decision support in CSDH referral management.

Main Methods:

• A multicenter retrospective case series from 2015-2020 analyzed CSDH referrals at two UK neurosurgical centers.
• Ten independent predictor variables were used to forecast referral acceptance (surgical treatment) or rejection (conservative management).
• Five ML algorithms were developed and externally validated to identify the most robust model.

Main Results:

• An artificial neural network model achieved 96.2% accuracy and an AUC of 0.951 on an internal holdout set.
• External validation on 1713 patients demonstrated an AUC of 0.896 and 92.3% accuracy.
• The developed ML model is publicly accessible for clinical decision support.

Conclusions:

• ML models demonstrate high accuracy in predicting CSDH referral outcomes, showing potential for clinical application.
• These tools can serve as valuable decision-making support systems for CSDH referrals.
• The increasing digitization of health records presents opportunities for ML in complex clinical decision-making scenarios.