Predicting lab values for gastrointestinal bleeding patients in the intensive care unit: A comparative study on the...

Area of Science:

• Medical Informatics
• Clinical Laboratory Science
• Data Science in Healthcare

Background:

• Unnecessary laboratory blood tests pose risks and increase healthcare costs.
• Predictive modeling can help identify and reduce superfluous diagnostic procedures.
• Gastrointestinal bleeding patients often undergo extensive laboratory testing.

Purpose of the Study:

• To develop and evaluate a predictive system for identifying unnecessary Calcium and Hematocrit laboratory blood tests.
• To assess the impact of patient comorbidity clustering on the accuracy of laboratory test value prediction.
• To improve patient safety and resource allocation in clinical settings.

Main Methods:

• Utilized the MIMIC-II database, with comprehensive data preprocessing and expert validation.
• Developed a prediction system using zero-order Takagi-Sugeno fuzzy modeling and sequential forward selection.
• Implemented a second system involving patient clustering by comorbidity before feature selection and prediction.

Main Results:

• The initial fuzzy modeling prediction system showed promising results for Calcium and Hematocrit.
• Clustering patients based on comorbidity did not lead to an average improvement in prediction accuracy.
• The dispersion of abnormal laboratory test values across clusters likely hindered performance gains.

Conclusions:

• Predictive modeling offers a viable approach to flagging potentially unnecessary laboratory tests.
• Comorbidity-based patient clustering may not enhance predictive accuracy for rare or dispersed abnormal lab values.
• Further research is needed to refine models for accurate prediction in diverse patient populations.