Abstract: FR-OR029

Development and Validation of an EHR-Based Computable Phenotype to Rapidly Identify Glomerular Disease in Large Populations

Session Information

Category: Glomerular

  • 1005 Clinical Glomerular Disorders

Authors

  • Denburg, Michelle, The Children's Hospital of Philadelphia, Philadelphia, Pennsylvania, United States
  • Zaritsky, Joshua, Nemours/Alfred I. duPont Hospital for Children, Wilmington, Delaware, United States
  • Flynn, Joseph T., Seattle Children's Hospital, Seattle, Washington, United States
  • Mitsnefes, Mark, Cincinnati Children's Hospital, Cincinnati, Ohio, United States
  • Benton, Maryjane, The Children's Hospital of Philadelphia, Philadelphia, Pennsylvania, United States
  • Mariani, Laura H., University of Michigan, Ann Arbor, Michigan, United States
  • Forrest, Christopher B, The Children's Hospital of Philadelphia, Philadelphia, Pennsylvania, United States
  • Furth, Susan L., The Children's Hospital of Philadelphia, Philadelphia, Pennsylvania, United States
  • Bailey, Charles, The Children's Hospital of Philadelphia, Philadelphia, Pennsylvania, United States
  • Razzaghi, Hanieh, The Children's Hospital of Philadelphia, Philadelphia, Pennsylvania, United States
  • Soranno, Danielle, Children's Hospital Colorado, Aurora, Colorado, United States
  • Pollack, Ari, Seattle Children's Hospital, Seattle, Washington, United States
  • Claes, Donna J., Cincinnati Children's Hospital, Cincinnati, Ohio, United States
  • Dharnidharka, Vikas R., Washington University School of Medicine, St Louis, Missouri, United States
  • Smoyer, William E., Nationwide Children's Hospital, Columbus, Ohio, United States
  • Somers, Michael J., Children's Hospital, Boston, Massachusetts, United States
Background

The objective of this study was to develop and validate a computable phenotype algorithm to identify all patients with glomerular disease (GD) using PEDSnet, a pediatric clinical research network (CDRN) that aggregates and standardizes electronic health record (EHR) data on >5 million children from multiple pediatric health systems.

Methods

A systematic review of EHR data from 231 patients with GD seen at the Children's Hospital of Philadelphia (CHOP) from April-December 2013 was used to develop, iterate, and validate a computerized algorithm comprised of Systematized Nomenclature of Medicine (SNOMED) diagnosis codes, kidney biopsy and transplant procedure codes, and pediatric nephrologist encounters. The algorithm identified 6138 cases of GD from 7 PEDSnet institutions. For validation, non-cases were defined as patients with ≥3 pediatric nephrologist encounters (n=42,186). Random samples of cases (n=694) and non-cases (n=697) were evaluated at each site using a standardized chart review form, and with the reviewer blinded to case status.

Results

When first implemented at CHOP, the computable phenotype identified GD with a sensitivity (SENS) of 100%, specificity (SPEC) of 93%, positive predictive value (PPV) of 92%, and negative predictive value (NPV) of 100%. When implemented across 7 health systems in PEDSnet, the performance characteristics were: SENS 97%, SPEC 80%, PPV 76%, and NPV 97%. One SNOMED code contributed considerably to the false positives, and refining the algorithm to exclude it as a qualifying code improved performance: SENS 96%, SPEC 87%, PPV 83%, and NPV 97%. The most common biopsy-based diagnoses were focal segmental glomerulosclerosis (18%), minimal change disease (17%), lupus nephritis (16%), and IgA nephropathy (13%). The most common non-biopsy diagnosis was nephrotic syndrome (41%).

Conclusion

We developed and validated an EHR-based computerized algorithm that identifies virtually all patients with GD within the PEDSnet CDRN. This method for rapid cohort identification is critcal for population-based comparative effectiveness and outcomes research in GD and other rare diseases across large health system populations.

Funding

  • NIDDK Support