Feasibility and Assessment of a Machine Learning-Based Predictive Model of Outcome After Lumbar Decompression Surgery

Area of Science:

• Neurosurgery
• Artificial Intelligence
• Medical Informatics

Background:

• Lumbar decompression surgery outcomes can be difficult to predict.
• Accurate prediction of surgical success is crucial for patient management.
• Developing predictive models can optimize patient selection and outcomes.

Purpose of the Study:

• To develop a virtual patient model for lumbar decompression surgery.
• To evaluate an artificial neural network (ANN) model for predicting surgical outcomes.
• To identify predictors of surgical success and failure in lumbar decompression.

Main Methods:

• Retrospective analysis of 60 complete Electronic Health Records (EHR).
• Creation of a synthetic EHR cohort using 12,000 virtual patients (10,000 for training, 2,000 for testing).
• Classification of patients into 'green zone' (success) and 'orange zone' (partial failure) using an ANN.

Main Results:

• The ANN model demonstrated 72% accuracy and a ROC score of 0.78.
• Sensitivity was 0.885 and specificity was 0.59.
• Patients in the 'orange zone' had a higher average number of positive criteria (10.92) compared to the 'green zone' (8.62).

Conclusions:

• The developed ANN model can predict favorable candidates for lumbar decompression surgery.
• Further development is needed to improve the analysis of patients in the 'failure of treatment' zone.
• The model aids in precise patient health management before spinal surgery.