Background/Purpose: Calcium pyrophosphate deposition disease (CPPD) has a spectrum of manifestations, of which pseudogout is the most acute inflammatory phenotype. To facilitate clinical research on CPPD, an ICD-9 algorithm for “definite or probable CPPD” (per Ryan and McCarty’s diagnostic criteria) was developed at a Veterans’ Administration (VA) Medical Center.¹ The algorithm includes ≥1 ICD-9 code 275.49 (“other disorders of calcium metabolism”) or 712.1-712.39 (“chondrocalcinosis” due to dicalcium phosphate crystals, pyrophosphate crystals, or cause unspecified). We hypothesized that the algorithm would be suboptimal for identifying pseudogout, and aimed to enhance its ability to identify pseudogout in electronic medical records (EMR).

Methods: Following the published methods, we applied the ICD-9 algorithm to patients with ≥1 encounter at our non-VA academic medical center over 2 years (1/1/15-12/31/16). 100 patients were randomly selected for EMR review from date of 1st qualifying ICD-9 code through 5/1/17. We evaluated whether patients fulfilled each of 3 clinical definitions: “definite or probable CPPD”—which the algorithm was designed to identify—and 2 pseudogout definitions (Table). We calculated the positive predictive value (PPV) and 95% confidence intervals (CI) of the ICD-9 algorithm for identifying each phenotype. We then modified the ICD-9 algorithm to include text string searching for “pseudogout” or “calcium pyrophosphate crystals” in narrative notes and re-assessed test performance characteristics.

Results: 55% of 100 patients were female; mean age was 68.6 (±13.9) years. Joint pain was present in 86% and synovitis in 34%. The published algorithm had 68% (59-77%) PPV for “definite or probable CPPD” in our sample (Table), compared to 91% (88-94%) PPV in the VA derivation study. The published algorithm identified only 18% (10-26%) of patients with crystal-proven pseudogout. In our sample of 100 patients, 50 had a positive text string search. Of these, 17 had crystal-proven pseudogout per EMR review, producing a PPV of 34% (25-43%) and NPV of 98% (95-100%). Text string searching was 94% sensitive (89-99%) and 60% specific (50-70%) for crystal-proven pseudogout in the sample.

Conclusion: Adding text string searching to a published ICD-9 CPPD algorithm improved PPV for identifying crystal-proven pseudogout from 18% to 34%, with excellent sensitivity and moderate specificity. The accuracy of identifying pseudogout in EMR data might be further improved using more advanced text searching methodology, such as natural language processsing.

1. Bartels C, Singh J, Parperis K, et al. Validation of administrative codes for calcium pyrophosphate deposition. J Clin Rheum 2015;21:189-92

2. Zhang W, Doherty M, Bardin T, et al. EULAR recommendations for calcium pyrophosphate deposition. Ann Rheum Dis 2011;70:563-70