Development and validation of an electronic phenotyping algorithm for chronic kidney disease

Area of Science:

• Nephrology
• Medical Informatics
• Genomics

Background:

• An estimated 26 million Americans have chronic kidney disease (CKD), increasing risks for cardiovascular and end-stage renal disease.
• CKD is often undiagnosed, delaying crucial interventions and patient identification for research.
• Electronic Medical Records (EMR) offer a valuable resource for improving CKD detection.

Purpose of the Study:

• To develop and validate an automated phenotyping algorithm for identifying diabetic and/or hypertensive CKD cases and controls.
• To improve the accuracy and timeliness of CKD identification from EMR data.
• To compare the algorithm's performance against traditional diagnostic coding methods.

Main Methods:

• An automated phenotyping algorithm was developed using EMR data, including diagnostic codes, lab results, medications, blood pressure records, and clinical notes.
• The algorithm was deployed within the eMERGE (electronic medical records and genomics) Network.
• Validation was performed using positive predictive values (PPV) and negative predictive values (NPV).

Main Results:

• The algorithm achieved high performance with a 96% PPV and 93.3% NPV.
• Independent validation at two eMERGE institutions confirmed these robust results.
• The automated algorithm significantly outperformed identification using ICD-9-CM codes (63% PPV, 54% NPV).

Conclusions:

• Automated phenotyping of CKD from EMRs is accurate and effective.
• This tool can enhance healthcare quality by enabling timely CKD identification and facilitating patient recruitment for research.
• The developed algorithm represents a significant advancement over traditional diagnostic coding for CKD detection.