(116) Case-Identifying Algorithms of Chronic Kidney Disease for Real-World Evidence Research: A Targeted Literature Review

Background: CKD is a leading cause of morbidity and mortality, and up to half of the cases go undiagnosed until late in the disease course. Real-world databases are valuable for understanding contemporary epidemiology, treatment, and outcomes among people with CKD. However, effective strategies to robustly identify early-stage CKD patients are needed, to improve early diagnosis, monitoring, intervention and outcomes in these patients. Understanding the availability of stage-specific case-identifying algorithms is key to designing future real-world studies.

Objectives: To summarize and evaluate the performance of CKD stage-specific case-identifying algorithms to inform patient identification in future real-world studies.

Methods: A targeted literature review (TLR) was conducted in Ovid MEDLINE to capture case-identifying algorithms for CKD from 2016 onwards; supplemented by additional targeted searching of PubMed and Google Scholar. Two reviewers independently screened and identified abstracts and extracted study and algorithm details from eligible studies. A case-identifying algorithm was defined as a comparison of a population identified from a claims-based CKD definition to a gold standard definition. Performance estimates of the positive predictive value (PPV), negative predictive value (NPV), sensitivity (SE), and specificity (SP), where available, were summarized, and stratified by algorithm features, including CKD stage.

Results: Fourteen studies yielding 108 CKD case-identifying algorithms were included: 53.5% used diagnosis codes only and 17.6% used a combination of diagnosis codes and clinical data (i.e., diagnosis, procedure, and/or lab data). Overall, approximately 50% of the algorithms identified 'any' CKD (stage 1-5; 23.1%) or more severe CKD (stages 3a-5; 25.9%). Only 1.9% of algorithms identified CKD stages 1 to 4 individually, and none identified algorithms captured early-stage CKD (stage 1-2 combined). Among algorithms with ≥1 performance measure (n=101), SP increased with CKD stage, ranging from 93.0% (stage 1) to 99% (stage 5). Compared to diagnosis codes only, combination algorithms had higher PPV (77.0% vs. 51.6%), NPV (92.4% vs. 89.9%), and SE (61.7% vs. 37.2%), but slightly lower SP (93.0% vs. 95.4%).

Conclusions: Few CKD stage-specific claims-based algorithms were identified, with a particular paucity in early-stage algorithms. Further research examining the feasibility of developing claims-based stage-specific CKD case-identifying algorithms, is warranted, particularly for early-stage CKD. This will help identify the at-risk individuals with CKD who can be adequately screened and targeted for appropriate interventions.