Probabilistic Graphical Models for Evaluating the Utility of Data-Driven ICD Code Categories in Pediatric Sepsis.

Area of Science:

• Biomedical Informatics
• Clinical Research
• Data Science

Background:

• Electronic health records (EHRs) are crucial for clinical data management and outcomes research.
• Ambiguous definitions for pediatric sepsis lead to diagnostic delays, necessitating improved patient categorization.
• EHR data, optimized for billing, may lack clinical granularity, impacting research accuracy.

Purpose of the Study:

• To evaluate the utility of probabilistic graphical models (PGMs) for analyzing International Classification of Diseases (ICD) codes in pediatric sepsis research.
• To compare data-driven ICD code categorization using PGMs against traditional chart review.
• To address challenges in EHR data granularity and misclassification for improved sepsis outcomes research.

Main Methods:

• Utilized probabilistic graphical models (PGMs) to handle uncertainty and incorporate prior knowledge in data analysis.
• Compared data-driven ICD code categories derived from EHRs with manual chart review findings.
• Focused on analyzing ICD codes for pediatric sepsis patient categorization.

Main Results:

• Demonstrated the potential of PGMs in analyzing ICD codes for research purposes.
• Showcased the ability of PGMs to manage uncertainty inherent in EHR data.
• Highlighted improvements in patient condition representation compared to standard methods.

Conclusions:

• Probabilistic graphical models (PGMs) show promise for enhancing the analysis of EHR data in clinical research.
• PGMs can mitigate challenges related to data granularity and misclassification in EHRs.
• This approach can lead to more precise patient categorization for pediatric sepsis outcomes research.