Leveraging Electronic Health Record to Monitor Progression of Kidney Disease in Children

CKD is a growing health problem, affecting 7%–12% of the adult population.1 By contrast, CKD remains relatively rare in children,2 with prevalence estimates difficult to determine based on a lack of population-based studies. The most common causes of CKD progressing to kidney failure in childhood are congenital anomalies of the kidney and urinary tract and acquired glomerular diseases, very different phenotypes than diabetes and hypertension, the most common causes in adults.3 As such, understanding the natural history and risk factors for progression relevant to children requires pediatric-specific data. To achieve sufficient sample sizes, multicenter studies are required. Until now, we have had to rely on large-scale prospective cohort studies, such as the Chronic Kidney Disease in Children (CKiD) study (n=>1000 from 50 centers), to define risk factors and relevant comorbidities unique to children.4 Administrative health care data are another potential population-based data source that has been used for decades for chronic disease surveillance using validated algorithms based on International Statistical Classification of Diseases (ICD) codes from inpatient and outpatient billing data. Validation work has shown excellent sensitivity and specificity for some diseases including diabetes and inflammatory bowel disease.5 Unfortunately, the validity of algorithms to identify CKD in adults and children have been found to be low, with a median sensitivity of only 41% (range 3%–88%) in adults from a systematic review of 19 studies6 and 20%–39% in a recent pediatric study from Canada.7 Under-recognition of abnormal laboratory tests and the lack of adherence to screening recommendations in at-risk populations are hypothesized to contribute to this low sensitivity. Electronic health record (EHR) data now provide a relatively novel, modern data system that has several advantages over traditional administrative health care data sources. EHRs typically contain ICD diagnostic codes from outpatient physician visits, as well as vital signs and anthropometrics collected during visits, prescriptions, and laboratory data. In the case of PEDSNet, inpatient admissions and emergency department visits are also included from site-affiliated hospitals. These additional data greatly improve the potential to capture individuals with CKD and their comorbidities, which are largely based on laboratory testing. Gluck et al.8 present in this issue of CJASN the study titled “Evaluating Kidney Function Decline in Children with Chronic Kidney Disease Using a Multi-Institutional Electronic Health Record Database,” which describes the creation of a large cohort of 11,240 children aged 18 months to 18 years between 2009 and 2020 with CKD in the United States. This large sample was achieved by merging data from six pediatric health systems enrolled in PEDSnet, which is a multispecialty Clinical Research Network in the United States aimed at addressing gaps in knowledge in pediatric care. This innovative infrastructure has the potential to provide access to large pediatric datasets of children with pediatric CKD, among other health conditions. They have identified a prevalence of stage 2–4 CKD of 15.7 per 10,000 children in this population of over 7 million children, as well as evaluated rates of progression, and clinical risk factors for kidney function decline. They have identified lower eGFR category, glomerular disease, malignancy, proteinuria, hypertension, complex chronic comorbidities, male sex, and younger age at cohort entrance as being associated with more rapid progression of CKD. While these findings are not novel, they are in keeping with data from the CKiD study4 and therefore support the validity of the EHR cohort model and the analytic approach. The strength of this study is the demonstration that clinically collected real-world EHR data can provide large population dataset of children with rare disease such as CKD, with extended follow-up periods (median of 5.1 years in this study) that are available for analysis. They use a common data model, which is based on the Observational Medical outcomes Partnership (OMOP) version 5. Codesets are available at https:github.com/PEDSnet/ckd_progression. The PEDSnet infrastructure is invaluable, has provided linkages between multiple systems, and therefore increases the generalizability of the findings. As it is population based, it also decreases the risk of selection bias, which has been a concern in prospective cohort studies. Critically important considerations for this type of research however are the validity of definitions used to identify the target population. The authors have used the following definition for CKD: eGFR values <90 ml/min per 1.73 m2 and ≥15 ml/min per 1.73 m2, separated by ≥90 days without an intervening value ≥90. CKD progression was defined as a composite outcome: eGFR <15 ml/min per 1.73 m2, 50% eGFR decline, long-term dialysis, or kidney transplant. They elected not to include proteinuria in the diagnosis, as has been done in previous studies7 and in keeping with the Kidney Disease Improving Global Outcomes CKD criteria.9 Therefore, children with stage 1 CKD are not included, nor are children with documented proteinuria, but missing serum creatinines in the dataset. In addition, they limited their population to children who have been seen by a nephrologist. While this increases the confidence in the diagnosis of CKD and the specificity, many children not yet referred to a specialist will also be missed. It is highly probable that this cohort of children is not representative of the general pediatric CKD population. There are also several other important considerations. Subcohorts within this study were defined based on ICD codes, stratifying by CKD etiology into glomerular, nonglomerular, or malignancy. The reliability of these diagnoses heavily depends on the accuracy of codes entered by care providers and the specificity of the codes themselves. It is unknown how often these conditions are reliably labeled in the available data fields in the EHR. Validity work on these diagnoses would increase the confidence in the subcohort categories. Consensus-based definitions or those based on previous studies that were not validated are not best practice. Finally, they also included hypertension, defined as ≥2 visits with a hypertension diagnosis code. Previous validation studies have shown that the sensitivity of hypertension diagnoses is quite good. However, the diagnosis of hypertension in children is not straight forward and likely is not fully appreciated by many primary care providers. As such, the sensitivity of hypertension diagnosis is likely also low. While there are challenges with BP measurement in children, they are the basis for the diagnosis of hypertension and consideration should be made to use these measurements in EHR studies to identify children with abnormal blood pressure. This study by Gluck et al. describes the largest multisystem cohort of children with CKD identified using EHR data. This platform will provide sufficient sample sizes of children with CKD to conduct meaningful and important epidemiologist studies and future pragmatic randomized clinical trials. Additional validation work will only strengthen the confidence in these data.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.