Benchmarking cesarean delivery rates using machine learning-derived optimal classification trees.

Area of Science:

• Healthcare quality improvement
• Clinical informatics
• Machine learning in medicine

Background:

• Cesarean delivery rates vary significantly across hospitals and physician groups.
• Objective performance evaluation tools are needed to identify variations and guide quality improvement initiatives.

Purpose of the Study:

• To develop a machine learning-based tool for case-adjusted, hospital-specific performance evaluation of cesarean deliveries.
• To establish benchmarking standards for physician groups within a hospital setting.

Main Methods:

• Utilized optimal classification trees (OCTs), a machine learning methodology, to predict cesarean delivery rates.
• Developed a prediction model based on 12,841 singleton, vertex, term deliveries from an electronic data warehouse (2015-2018).
• Created case-adjusted benchmarks by predicting outcomes for specific patient populations within different physician groups.

Main Results:

• The overall cesarean delivery rate was 18.6%, with significant variation (13.3%-33.7%) among 22 physician practices.
• The OCT prediction model defined 23 patient cohorts and demonstrated an area under the curve of 0.73.
• Comparisons revealed that some physician groups outperformed the hospital benchmark, while others underperformed.

Conclusions:

• Optimal classification tree (OCT) benchmarking effectively assesses physician practice-specific, case-adjusted performance.
• This methodology serves as a valuable tool for hospital self-assessment and targeted quality improvement in cesarean delivery rates.