Validation of a Major and Clinically Relevant Nonmajor Bleeding Phenotyping Algorithm on Electronic Health Records

Area of Science:

• Medical Informatics
• Epidemiology
• Health Data Science

Background:

• Bleeding events are critical health outcomes in epidemiological research.
• Accurate identification of bleeding in electronic healthcare data is essential for population studies.

Purpose of the Study:

• To develop and validate rule-based algorithms for identifying major bleeding and all clinically relevant bleeding (CRB) in real-world electronic healthcare data.
• To compare algorithm performance against diagnosis codes alone.

Main Methods:

• Utilized a random sample of 1630 inpatient admissions from Singapore public healthcare institutions (2019-2020).
• Ascertained major bleeding and CRB through chart review by two annotators.
• Developed and validated sensitivity- and positive predictive value (PPV)-optimized algorithms combining hemoglobin test patterns and diagnosis codes.

Main Results:

• Diagnosis codes alone showed low sensitivity (0.16 for major bleeding, 0.24 for CRB) but high specificity and PPV (>0.97).
• Sensitivity-optimized algorithm for major bleeding achieved high sensitivity (0.94) and NPV (1.00) but lower PPV (0.34).
• PPV-optimized algorithm for major bleeding improved specificity (0.96) and PPV (0.52) with minimal impact on sensitivity (0.88).
• Algorithms for CRB events demonstrated lower sensitivities (0.50-0.56).

Conclusions:

• Diagnosis codes alone are insufficient for accurately identifying major bleeding events in electronic health records.
• Developed major bleeding algorithms demonstrate high sensitivity, enabling better ascertainment of bleeding events in large populations.
• These algorithms are valuable tools for epidemiological studies requiring accurate bleeding event identification.