Algorithm to Identify Type 2 Diabetes Using Electronic Health Record and Self-Reported Data

Area of Science:

• Biomedical Informatics
• Clinical Research

Background:

• Accurate identification of participants with type 2 diabetes (T2D) is crucial for clinical research.
• Relying solely on electronic health records (EHR) or self-reported data presents limitations in T2D classification accuracy.

Purpose of the Study:

• To develop and validate an algorithm integrating EHR and self-reported data for precise identification of individuals with and without T2D.
• To improve the accuracy of T2D case ascertainment in large biobank cohorts.

Main Methods:

• Utilized data from the Mayo Clinic Biobank, including baseline questionnaires and EHR data (ICD codes, HbA1c, glucose, medications).
• Developed an algorithm to classify participants into T2D, no T2D, "only self-reported T2D," and "only self-reported no T2D" categories.
• Validated the algorithm's performance using manual chart reviews as the gold standard, calculating positive predictive value (PPV) and negative predictive value (NPV).

Main Results:

• The algorithm classified 57,000 participants: 6,238 with T2D, 38,883 with no T2D, 757 with "only self-reported T2D," and 9,759 with "only self-reported no T2D."
• Achieved high performance metrics: PPV of 96.0%, NPV of 100%, and overall accuracy of 99.5%.
• Significant differences observed in age and sex distribution across classification groups.

Conclusions:

• The developed algorithm demonstrates high accuracy and reliability for identifying individuals with and without T2D using combined EHR and self-reported data.
• This validated algorithm offers a robust tool for T2D ascertainment in research settings, potentially applicable to other cohorts with linked EHR data.