MySurgeryRisk: Development and Validation of a Machine-learning Risk Algorithm for Major Complications and Death...

Area of Science:

• Medical Informatics
• Surgical Outcomes Research
• Machine Learning in Healthcare

Background:

• Postoperative complications significantly increase mortality and healthcare costs.
• Current methods for predicting surgical risk are limited in precision.
• Accurate preoperative risk assessment is crucial for patient management.

Purpose of the Study:

• To develop and validate a machine-learning algorithm for predicting postoperative complications and death.
• To utilize readily available clinical data from electronic health records for risk prediction.
• To provide patient-level probabilistic risk scores for major adverse events.

Main Methods:

• Developed and validated an automated analytics framework (MySurgeryRisk) on a cohort of 51,457 surgical patients.
• Utilized existing clinical data from electronic health records for model development.
• Assessed model performance using area under the receiver characteristic curve (AUC) and predictiveness curves.

Main Results:

• MySurgeryRisk achieved high discriminatory ability for 8 major postoperative complications (AUCs 0.82-0.94).
• The model accurately predicted the risk of death up to 24 months post-surgery (AUCs 0.77-0.83).
• The algorithm demonstrated strong performance with 99% confidence intervals provided for AUC values.

Conclusions:

• An automated machine-learning framework was created to assess surgical complication and death risk.
• The algorithm effectively uses preoperative electronic health record data for risk stratification.
• Further testing is required to evaluate the feasibility of real-time clinical workflow implementation.