Background: Medicare claims data might provide an efficient source for outcomes research in patients with chronic kidney disease (CKD). However, in the absence of laboratory data, one would need to identify patients with CKD from diagnosis codes associated with health care claims. The validity of this approach to identify patients with CKD has not been sufficiently studied.
Methods: From chart abstraction, we obtained the first serum creatinine measurement of 1,852 elderly Medicare beneficiaries upon hospitalization for myocardial infarction and estimated each patient's glomerular filtration rate. We then searched all Medicare claims of the preceding year for the presence of a diagnosis code for diabetic nephropathy, hypertensive nephropathy, chronic renal insufficiency, acute renal failure, and miscellaneous other renal diseases. Using the gold standard of an estimated glomerular filtration rate less than 60 mL/min/1.73 m2 (<1.00 mL/s/1.73 m2) for definition of CKD, we calculated the sensitivity, specificity, and positive and negative predictive values for each of these diagnoses and combinations of these diagnoses.
Results: The sensitivity of individual diagnosis algorithms ranged from 2.7% for diabetic nephropathy to 17.5% for miscellaneous. However, miscellaneous had a lower specificity (95.5%) than all other individual diagnosis algorithms (all > or =99%). Using combinations of these algorithms improved sensitivity up to 26.6%, but at the cost of lower specificity. Positive predictive values generally were high (85.7% to 97.5%), but negative predictive values were low (32.4% to 37.4%).
Conclusion: High positive predictive values indicate that Medicare claims data can be used to accurately identify patients with CKD for study. However, the utility of such databases for comparison of patients with CKD versus lesser degrees of CKD is limited.