Development and Validation of a Multi-Modal Algorithm for Chronic Kidney Disease Detection in a Hospital Clinical...

Area of Science:

• Medical Informatics
• Clinical Data Science
• Nephrology Research

Background:

• Chronic kidney disease (CKD) is frequently underdiagnosed in hospitalized elderly patients.
• Incomplete medical coding and limited laboratory data contribute to CKD underdetection.

Purpose of the Study:

• To develop and validate an interpretable, multimodal algorithm for CKD phenotyping in hospitalized older adults.
• To leverage electronic health records, combining laboratory and textual information for improved CKD detection.

Main Methods:

• Utilized eHOP, a clinical data warehouse, to develop a rule-based algorithm.
• Combined laboratory data with clinical text information for multimodal phenotyping.
• Validated the algorithm against ICD-10 codes and a clinician-reviewed subset.

Main Results:

• The algorithm achieved 92.1% accuracy, 96.5% recall, and 86.6% precision in CKD detection.
• Clinical text analysis was crucial for identifying the majority of CKD cases.
• Validation confirmed the reliability and accuracy of the algorithm's classifications.

Conclusions:

• Transparent multimodal phenotyping using hospital data is feasible.
• This approach enhances CKD detection in hospitalized older adults.
• The developed method provides a foundation for scalable decision-support systems for chronic disease surveillance.