Abstract: TH-PO545

CKD Phenotype Validation Using Electronic Health Records (EHR): A Pilot Study

Session Information

Category: Chronic Kidney Disease (Non-Dialysis)

  • 307 CKD: Health Services, Disparities, Prevention

Authors

  • Stanifer, John W., Duke University, Durham, North Carolina, United States
  • Cameron, C. Blake, Duke University Medical Center, Durham, North Carolina, United States
  • Beasley, Cherry M, The University of North Carolina at Pembroke, Pembroke, North Carolina, United States
  • Wang, Daphne W, Duke University, Durham, North Carolina, United States
  • Bhavsar, Nrupen Anjan, Duke University School of Medicine , Durham, North Carolina, United States
  • Boulware, L. Ebony, Duke University School of Medicine , Durham, North Carolina, United States
  • Diamantidis, Clarissa Jonas, Duke University School of Medicine , Durham, North Carolina, United States
Background

Few studies have assessed the accuracy of EHR-based methods for detecting CKD, and phenotypes validated from EHR data abstraction alone are inadequate.

Methods

In a rural healthcare system we piloted a novel EHR-based CKD detection phenotype. Individuals were assigned to one of six mutually exclusive groups according to expected likelihood of CKD from available EHR data. Group 1 had individuals with persistently (>90 days) reduced eGFR (<60ml/min/1.73m2) or albuminuria (≥30mg/g) or an ICD diagnosis of ESRD or kidney replaced by transplant; (2) individuals with 1 reduced eGFR or albuminuria and a comorbidity index ≥4, adopted from the Screening for Occult Renal Disease prediction tool; (3) individuals with 1 reduced eGFR value or albuminuria and comorbidity index <4; (4) individuals with no kidney function measurements and comorbidity index ≥4; (5) individuals with no kidney function measurements and comorbidity index <4; (6) individuals with normal eGFR and no albuminuria. As a reference standard for validation, we randomly sampled individuals from each group and assessed for CKD through measured assessments, with CKD defined by KDIGO 2012 criteria or self-reported transplant or dialysis. We calculated sensitivity and specificity and used ROC curves to assess performance of the CKD phenotype.

Results

Of 59,848 adults in the EHR, 4036 (7%) were classified group 1; 1690 (3%) group 2; 3007 (5%) group 3; 4726 (8%) group 4; 39,349 group 5 (66%); and 7040 (12%) group 6. For validation, we enrolled 110 individuals, of whom 75 (68%) had CKD based on the reference standard. The CKD phenotype showed a range of sensitivity and specificity corresponding with the CKD phenotype groups (table). The ROC curve area was 0.82 (95% CI 0.74-0.90).

Conclusion

Our CKD phenotype showed ability to detect CKD in a small sample identified from the EHR. Additional studies are needed to validate the CKD detection phenotype in other settings and larger samples.

CKD phenotype performance
GroupSensitivitySpecificityLikelihood Ratio
≤6100%0%1.0
≤5100%9%1.1
≤499%17%1.2
≤396%37%1.5
≤284%54%1.8
179%83%4.6

Funding

  • Private Foundation Support