Abstract: FR-PO980
A Comparison of Electronic Health Record (EHR) Phenotype Definitions for CKD
Session Information
- Bioengineering and Informatics
November 03, 2017 | Location: Hall H, Morial Convention Center
Abstract Time: 10:00 AM - 10:00 AM
Category: Bioengineering and Informatics
- 101 Bioengineering and Informatics
Authors
- Cameron, C. Blake, Duke University, Durham, North Carolina, United States
- Stanifer, John W., Duke University, Durham, North Carolina, United States
- Richesson, Rachel, Duke University, Durham, North Carolina, United States
Background
Multiple methods for identifying patients with CKD using EHR-based phenotypes have been proposed. Few studies have systematically compared or prospectively validated these phenotype definitions.
Methods
In a rural, community-based healthcare system, we applied five distinct CKD phenotype definitions (A-E) to the EHR. Phenotype A defined CKD as >2 eGFR values <60mL/min/1.73m2 separated by 90-730 days. Phenotype B was the same but also included individuals with albuminuria >30mg/g. Phenotypes C and D defined CKD by a single eGFR result <60 and <45 mL/min/1.73m2 respectively. Phenotype E defined CKD as having >2 ambulatory encounters associated with an eligible ICD-9-CM or ICD-10-CM diagnosis code. We evaluated inter-rater agreement between each phenotype pair by calculating Chamberlain’s percent positive agreement and Cohen’s Kappa statistic.
Results
We identified 59,848 unique adults with at least one ambulatory encounter over a two-year period, of whom 6,620 (11%) were classified as having CKD by any one of the phenotypes. Only 666 (1%) were classified as having CKD by all five phenotypes. Phenotype C classified the most patients as having CKD (n=5,596; 9%), followed by phenotype B (n=3,837; 6%); phenotype A (n=3,268; 5%); phenotype D (n=2,552; 4%) and phenotype E (n=1,615; 3%). Phenotypes A and B showed the greatest agreement (Kappa=0.915), followed by phenotypes A and C (Kappa=0.718) and phenotypes B and C (Kappa=0.693). Phenotype E showed low agreement with any of the phenotypes.
Conclusion
In a rural, community-based healthcare system, several commonly used phenotype definitions showed poor agreement in classifying CKD. Additional studies using external reference standards that include prospective laboratory assessment of kidney function and albuminuria are required in order to validate performance characteristics of CKD phenotypes. Once validated, one or more CKD phenotypes could be promoted as a standard to define similar populations for clinical research and population health management.
Positive overlap, percent positive agreement (PPA) and Kappa statistic for phenotype pairs
Phenotype A (n=3,268) | Phenotype B (n=3,837) | Phenotype C (n=5,596) | Phenotype D (n=2,552) | |
Phenotype B (n=3,837) | n=3,268 PPA=85.2% Kappa=0.915 | |||
Phenotype C (n=5,596) | n=3,268 PPA=58.4% Kappa=0.718 | n=3,379 PPA=55.8% Kappa=0.693 | ||
Phenotype D (n=2,552) | n=1,934 PPA=49.8% Kappa=0.648 | n=1,967 PPA=44.5% Kappa=0.595 | n=2,552 PPA=45.6% Kappa=0.603 | |
Phenotype E (n=1,615) | n=788 PPA=19.2% Kappa=0.297 | n=848 PPA=18.4% Kappa=0.284 | n=1,010 PPA=16.3% Kappa=0.249 | n=801 PPA=23.8% Kappa=0.363 |
Funding
- Private Foundation Support