Accelerating Chart Review Using Automated Methods on Electronic Health Record Data for Postoperative Complications.

Area of Science:

• Clinical informatics
• Machine learning in healthcare
• Medical data analysis

Background:

• Manual Chart Review (MCR) is essential for clinical research and quality improvement but is time-consuming.
• Extracting postoperative outcomes from medical records presents a significant bottleneck.

Purpose of the Study:

• To develop an automated application for detecting postoperative complications using structured electronic health record (EHR) data.
• To accelerate the process of extracting critical postoperative outcomes from patient charts.

Main Methods:

• Applied machine learning techniques to identify common postoperative complications, including surgical site infections, pneumonia, urinary tract infections, sepsis, and septic shock.
• Compared the performance of one single-task learning model against five multi-task learning models, including propensity weighted observations (PWO).

Main Results:

• The developed models demonstrated high detection performance for postoperative complications.
• Propensity weighted observations (PWO), a multi-task learning method, achieved the highest detection accuracy.
• Single-task learning models also showed strong performance, closely following the top multi-task approach.

Conclusions:

• Automated detection of postoperative complications using EHR data is feasible and can significantly accelerate MCR.
• Machine learning, especially multi-task learning methods like PWO, offers a promising solution for efficient clinical data analysis.