Automated Type 2 Diabetes Case and Control Identification from the MIMIC-IV Database

Area of Science:

• Biomedical Informatics
• Clinical Research Informatics
• Health Data Science

Background:

• Increasing demand for Type 2 Diabetes Mellitus (T2DM) research necessitates robust phenotyping from electronic health records (EHRs).
• The eMERGE algorithm offers a reliable, rule-based method for identifying T2DM cases and controls in EHR data.
• MIMIC-IV is a large, valuable EHR database for secondary research, but lacks prior T2DM case/control extraction.

Purpose of the Study:

• To adapt and apply the eMERGE algorithm for T2DM phenotyping in the MIMIC-IV database.
• To provide insights into MIMIC-IV's structure and data elements relevant for T2DM research.
• To establish cohorts of T2DM cases and controls from MIMIC-IV for future studies.

Main Methods:

• Explored MIMIC-IV database structure and relevant data elements.
• Adapted the eMERGE rule-based algorithm for T2DM case and control identification.
• Applied the adapted algorithm to the MIMIC-IV dataset.

Main Results:

• Detailed MIMIC-IV data tables and elements utilized in phenotyping.
• Successfully identified 12,735 T2DM cases and 9,828 controls.
• Generated summary statistics for the T2DM cohorts, enabling comparison with other EHR databases.

Conclusions:

• The adapted eMERGE algorithm effectively enables T2DM phenotyping in MIMIC-IV.
• The extracted T2DM cohorts provide a valuable resource for future statistical and machine learning model development.
• This work facilitates advanced research on T2DM using large-scale EHR data.