Development and validation of machine learning models for predicting operative duration in assisted reproductive...

Area of Science:

• Reproductive Medicine
• Health Informatics
• Surgical Workflow Optimization

Background:

• Accurate operative duration prediction is vital for surgical scheduling and resource management.
• Assisted reproductive technology (ART) procedures are time-sensitive, making precise timing critical to avoid workflow disruptions.
• Electronic medical record (EMR) data offers a potential source for developing predictive models in ART.

Purpose of the Study:

• To develop and validate interpretable prediction models for operative duration in ART procedures.
• To utilize routinely available EMR data for predicting surgical timing.
• To compare different predictive modeling paradigms for their effectiveness in ART.

Main Methods:

• Retrospective analysis of 763 ART operative cases from a South Korean fertility clinic.
• Operative duration determined by EMR start and end timestamps; predictors included procedure type, patient age, reservation characteristics, physician, and day of surgery.
• Evaluation of linear models, tree-based ensembles, and kernel/neural-network approaches using cross-validation, benchmarked against a moving-average heuristic.

Main Results:

• Regularized linear models, specifically ridge and Bayesian ridge regression, showed stable and interpretable performance.
• These linear models achieved a mean absolute error of approximately 3.1 minutes, a 7-8% improvement over the moving-average baseline.
• Procedure type, patient age, and reservation type were identified as the most significant predictors of operative duration.

Conclusions:

• Interpretable linear models utilizing EMR data provide consistent performance improvements for operative duration prediction in ART.
• These findings underscore the utility of transparent and validated modeling strategies in high-throughput clinical environments.
• The study demonstrates the value of EMR data for optimizing scheduling and resource allocation in time-sensitive procedures.