Interpretable machine learning accurately reclassifies lobectomy surgical approaches by cost.

Area of Science:

• Thoracic Surgery
• Health Economics
• Machine Learning in Healthcare

Background:

• Robotic lung resection volume is increasing despite higher costs and unproven superiority over video-assisted thoracoscopic surgery (VATS).
• Identifying cost drivers and optimizing surgical approaches is crucial for healthcare efficiency.

Purpose of the Study:

• To evaluate machine learning's ability to identify factors influencing costs in lung resection.
• To determine if high-cost robotic approaches can be reclassified into lower-cost alternatives using predictive modeling.

Main Methods:

• Analysis of Florida and CMS datasets for open, VATS, and robotic lobectomy cases.
• K-means clustering to identify robotic cost groups.
• Development of predictive models using artificial neural networks, SVM, CART, and GBM algorithms.

Main Results:

• Identified 4 distinct cost clusters among 6,618 cases.
• Machine learning models reclassified 35% of robotic cases to potentially lower or higher cost categories if VATS was used.
• Key cost predictors included clinic admission, metastatic cancer diagnosis, urgent admission, and dementia.

Conclusions:

• Machine learning and clustering can identify patient groups suitable for VATS or robotic lobectomy with similar costs.
• Individual patient factors influencing cost can be identified using explainable AI.
• This modeling can stratify high-cost patients to lower-cost approaches, aiding expenditure reduction.