Incidence of Leading Causes of Pediatric CKD Using Electronic Health Record-Driven Computable Phenotype.

KEY POINTS

We created computable phenotypes to accurately identify cases of pediatric CKD by underlying diagnosis. Combined annual incidence of five leading causes of pediatric CKD was high, 47.07 (95% confidence interval, 45.96 to 48.20) per 100,000 children. Our computable phenotypes have the potential to be broadly implemented to advance epidemiologic research in pediatric CKD.

BACKGROUND

Incidence data on pediatric CKD are incomplete. We developed electronic health record–based algorithms (e-phenotypes) to identify cases and provide incidence estimates of five leading causes of pediatric CKD.

METHODS

E-phenotypes using common standardized clinical terminology were built and contained utilization, diagnostic, procedural, age, and time-period inclusion and exclusion criteria for autosomal dominant polycystic kidney disease (ADPKD), Alport syndrome (AS), congenital anomalies of the kidney and urinary tract (CAKUT), lupus nephritis (LN), and primary childhood nephrotic syndrome (NS). Cases diagnosed between 2014 and 2023 were identified from a pediatric health care system that is the sole pediatric nephrology provider serving the Atlanta Metropolitan Statistical Area. The performance of the e-phenotypes was tested using a cohort of 1000 pediatric patients. The cases identified were used to estimate incidences using population information from the Georgia Department of Health.

RESULTS

The e-phenotypes demonstrated sensitivity ranging from 0.83 to 0.95, specificity 0.96 to 1.00, positive predictive value 0.81 to 1.00, and negative predictive value 0.98 to 1.00. The positive likelihood ratio was >20, and the negative likelihood ratio was <0.20. The 6814 combined cases of ADPKD (=107), AS (=31), CAKUT (=6120), LN (=161), and NS (=395) had an annual incidence of 47.07 (95% confidence interval, 45.96 to 48.20) per 100,000 children. Annual incidence per 100,000 children (95% confidence interval) for each condition was ADPKD 0.74 (0.61 to 0.89), AS 0.21 (0.15 to 0.30), CAKUT 42.28 (41.22 to 43.35), LN 1.11 (0.95 to 1.30), and NS 2.73 (2.47 to 3.01).

CONCLUSIONS

Our incidence estimates suggest that CKD conditions are common among children. The e-phenotypes require validation for use at other institutions but offer opportunities to examine determinants of CKD detection, management, and outcomes.