Developing a Machine Learning Model for Predicting 30-Day Major Adverse Cardiac and Cerebrovascular Events in...

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Preoperative cardiac risk assessment for noncardiac surgery is crucial for patient safety and resource management.
• Current evaluation methods can be excessive for low-risk patients and insufficient for high-risk individuals.
• Development of practical risk prediction tools is essential to optimize preoperative care.

Purpose of the Study:

• To develop and validate machine learning models for predicting major adverse cardiac and cerebrovascular events (MACCE) in patients undergoing noncardiac surgery.
• Utilize the Observational Medical Outcomes Partnership Common Data Model (OMOP CDM) for standardized data analysis.
• Compare the performance of developed models against the Revised Cardiac Risk Index (RCRI).

Main Methods:

• Retrospective observational study using de-identified electronic health records converted to OMOP CDM format.
• Data from 46,225 patients at Seoul National University Bundang Hospital and 396,424 at Asan Medical Center.
• Developed and validated 5 machine learning algorithms, including random forest, for predicting MACCE in patients aged 65+ undergoing noncardiac surgery.

Main Results:

• Machine learning models significantly outperformed the RCRI (AUROC=0.704) in predicting MACCE.
• The random forest model achieved the highest performance with an internal AUROC of 0.897 and external AUROC of 0.817.
• Key predictors included previous diagnoses and laboratory measurements, highlighting their importance in perioperative risk assessment.

Conclusions:

• Developed machine learning models demonstrate superior performance and generalizability in predicting MACCE within 30 days post-noncardiac surgery compared to the RCRI.
• These models can optimize preoperative evaluations, reduce unnecessary testing, and improve perioperative care efficiency.
• Implementation of these predictive models in clinical practice holds significant potential for enhancing patient outcomes and resource utilization.