Development of Multivariable Models to Predict and Benchmark Transfusion in Elective Surgery Supporting Patient Blood...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Patient Blood Management

Background:

• Blood transfusions are common but often unnecessary, leading to adverse outcomes and increased costs.
• Patient Blood Management (PBM) strategies aim to reduce transfusions and optimize their use.
• Predictive modeling offers a potential solution for optimizing transfusion decisions.

Purpose of the Study:

• To evaluate predictive modeling and machine learning for forecasting red blood cell (RBC) transfusion volumes.
• To prospectively optimize blood ordering schedules using predictive models.
• To benchmark hospital transfusion patterns using data-derived insights.

Main Methods:

• Analysis of 6,530 case records from elective surgeries across 16 centers (2004-2005 and 2009-2010).
• Prediction of transfused RBC volume using random forests.
• Development of separate models for overall data, individual centers, and study periods.

Main Results:

• Predictive modeling achieved higher accuracy (cc=0.61) in predicting RBC volume compared to existing algorithms (cc=0.39).
• Significant variations in feature importance were observed across different hospitals.
• Distinct patterns of feature importance were identified between the two study periods.

Conclusions:

• Predictive modeling effectively forecasts RBC transfusion volumes, enhancing PBM.
• Model insights can benchmark hospital practices and identify areas for process optimization.
• The approach has potential applications beyond PBM for improving healthcare processes.