Developing Laterality-Specific Computable Phenotypes from Electronic Health Record Data, Employing...

Area of Science:

• Ophthalmology
• Health Informatics
• Data Science

Background:

• Diabetic retinopathy (DR) management requires accurate laterality-specific data for treatment-warranted diabetic macular edema (TW-DME).
• Electronic health records (EHRs) contain valuable data but often lack completeness and accuracy regarding disease laterality.

Purpose of the Study:

• To develop a general algorithm using structured and unstructured EHR data to improve the identification of laterality-specific TW-DME.
• Enhance the accuracy of longitudinal eye-level disease assessment within EHRs.

Main Methods:

• A retrospective cohort study involving patients with DR from a safety net system and a university hospital (2013-2023).
• Investigated laterality data completeness and accuracy across five data tiers (documentation, charge codes, medications, procedure codes, diagnosis codes).
• Validated algorithm performance using manual chart review.

Main Results:

• The algorithm significantly improved laterality completeness to 93.6% (safety net) and 99.0% (university).
• Chart review validation demonstrated substantial increases in positive predictive value, negative predictive value, sensitivity, specificity, agreement, and F1 score.
• Performance metrics showed significant improvements across both safety net and university sites, indicating generalizability.

Conclusions:

• The developed algorithm provides a general, reproducible method for accurately extracting laterality-specific data from EHRs.
• This approach is effective across healthcare sites with varying documentation and coding practices.
• The algorithm can enhance the utility of routine clinical data for future ocular disease research, including prevalence, treatment patterns, and costs.