Optimizing Multivariable Logistic Regression for Identifying Perioperative Risk Factors for Deep Brain Stimulator...

Area of Science:

• Neurosurgery
• Medical Informatics
• Machine Learning

Background:

• Deep brain stimulation (DBS) is a key treatment for Parkinson's Disease (PD), but explantation occurs in 5.6% of cases, incurring significant costs.
• Traditional statistical methods have limitations in identifying complex risk factors for DBS explantation.
• Supervised machine learning offers a promising approach to uncover novel risk factors by analyzing intricate perioperative data interactions.

Purpose of the Study:

• To investigate the efficacy of supervised machine learning in identifying risk factors associated with early deep brain stimulation (DBS) explantation.
• To compare the predictive power of machine learning models against traditional statistical methods for DBS explantation risk.

Main Methods:

• Utilized the Medical Informatics Operating Room Vitals and Events Repository for patient data (n=38) with DBS and at least two years of follow-up.
• Employed Fisher's exact test for initial variable assessment and Recursive Feature Elimination with Cross-Validation (RFECV) for feature selection.
• Trained and evaluated a multivariate logistic regression model using precision, recall, F1-score, and Area Under the Curve (AUC).

Main Results:

• Fisher's exact test identified chronic pain (p=0.0108) and tobacco use (p=0.0026) as preliminary risk factors.
• RFECV selected 24 optimal features, and the logistic regression model achieved high performance (AUC: 1.0, Precision: 0.89, Recall: 0.86, F1-score: 0.86).
• Significant predictors for explantation included tobacco use (OR: 3.64), primary PD (OR: 2.01), ASA score (OR: 1.91), chronic pain (OR: 1.82), and diabetes (OR: 1.63).

Conclusions:

• Supervised machine learning is a viable tool for identifying risk factors contributing to early deep brain stimulation (DBS) explantation.
• The study highlights specific patient factors, such as tobacco use and chronic pain, as significant predictors of DBS explantation.
• Further research with larger cohorts is recommended to validate these machine learning-derived findings in clinical practice.