Evaluate the accuracy of ICD-10-CM algorithms for identifying CKD in the TriNetX network, using repeated eGFR measurements as the reference standard.
Approach:
Design and population: Real-world cohort study of adults in the TriNetX US Collaborative Network with eGFR measurements from 2021–2025, including a qualifying repeat measurement 90–365 days before the index value.
Reference standard and algorithms: Repeated eGFR values below 60 mL/min/1.73 m² defined CKD stage ≥3. Diagnostic performance was assessed using ≥1 or ≥2 ICD-10-CM codes, with additional analyses restricting code ascertainment to ±3, ±6, or ±12 months of the index eGFR.
Key Findings:
Among 8,239,691 adults with at least two eGFR measurements, 1,328,952 met laboratory criteria for CKD stage ≥3.
For stage ≥3, one ICD-10-CM code had 64.1% sensitivity, 95.6% specificity, 73.8% PPV, and 93.3% NPV.
For stage ≥4, one code had 79.2% sensitivity, 98.0% specificity, 57.3% PPV, and 99.3% NPV.
Requiring at least two codes reduced sensitivity and increased PPV. Temporal restrictions raised PPV to approximately 96% for stage ≥3 and 93% for stage ≥4, while sensitivity declined.
Interpretation:
In this TriNetX cohort, ICD-10-CM algorithms identified repeated eGFR-defined CKD with high specificity and NPV but moderate sensitivity. Code-count and timing choices involve a trade-off between PPV and sensitivity; the authors state that no single phenotype or temporal strategy is universally optimal.
Limitations:
The supplied context does not provide the full Methods or Results, including detailed performance estimates for the ≥2-code and individual temporal-window analyses.
The findings concern the TriNetX US Collaborative Network and may not describe other EHR systems or populations.
Conclusion:
The authors conclude that ICD-10-CM algorithms validly identify repeated eGFR-defined CKD in TriNetX, but incompletely, and suggest that combining diagnosis codes with longitudinal laboratory data may improve identification.
