Identifying Key Variances in Clinical Pathways Associated With Prolonged Hospital Stays Using Machine Learning and...

Area of Science:

• Medical Informatics
• Clinical Data Science
• Machine Learning in Healthcare

Background:

• Prolonged hospital stays (PLOS) increase healthcare inefficiencies and resource consumption.
• Identifying factors contributing to PLOS is crucial for optimizing patient care pathways.

Purpose of the Study:

• To develop and validate a machine learning model to identify clinical variances associated with PLOS in lung cancer patients undergoing video-assisted thoracoscopic surgery.
• To pinpoint key predictors of PLOS using real-world data from the ePath system.

Main Methods:

• Analysis of data from 480 lung cancer patients undergoing video-assisted thoracoscopic surgery.
• Development of predictive models using sparse linear regression (Lasso, ridge, elastic net) and decision tree ensembles (random forest, extreme gradient boosting).
• Temporal validation using derivation and testing cohorts, with performance assessed by AUC and Brier score; counterfactual analysis for factor identification.

Main Results:

• Ridge regression model showed optimal performance (AUC 0.84/0.82, Brier score 0.16/0.17).
• Key variables linked to PLOS included abnormal respiratory sounds, postoperative fever, arrhythmia, impaired ambulation, drain removal complications, and pulmonary air leaks.
• Clinical variances were identified as significant contributors to PLOS.

Conclusions:

• A machine learning model effectively identifies critical variances in clinical pathways contributing to PLOS.
• This automated tool can aid clinical decision-making and enhance patient management for lung cancer surgery.