The Design and Validation of a New Algorithm to Identify Incident Fractures in Administrative Claims Data

Our study validated a claims-based algorithm for the identification of incident and recurrent fractures in administrative data. We used Centers for Medicare and Medicaid (CMS) claims from 2005 to 2014 linked to the Reasons for Geographic and Racial Differences in Stroke (REGARDS) database. Case qualifying (CQ) fractures were identified among participants with ≥12 months of fee-for-service coverage before first fracture claim and ≥6 months after. Recurrent fractures were defined as the first CQ fracture that occurred following a clean period of at least 90 days from the last claim associated with the preceding incident fracture. We used medical records (discharge summary, imaging, and surgical report) to adjudicate fractures. We calculated positive predictive values (PPVs) for incident and recurrent fractures. Our study was not designed to assess the algorithm sensitivity or negative predictive value. We identified 2049 potential incident fractures from claims among 1650 participants. Record retrieval was attempted for 728 (35.5%) suspected incident fractures (prioritizing more recent CQ fractures associated with osteoporosis, but without explicitly requiring any osteoporosis ICD-9 diagnosis code). Our final sample included 520 claims-identified fractures with medical records, of which 502 (96.5%) were confirmed. The PPVs (95% CI) of the hip, wrist, humerus, and clinical vertebra-all exceeded 95%. We identified 117 beneficiaries with 292 ≥2 CQ fracture episodes at the same site, and attempted retrieval on 105 (36.0%) episodes. Our analytic sample included 72 (68.5%) CQ episodes from 33 participants. The PPVs for identifying recurrent clinical vertebral, hip/femur, and nonvertebral fractures with a 90-day clean period exceeded 95%. Although we could not ascertain sensitivity, our updated fracture identification algorithms had high PPV for the identification of incident and recurrent fractures of the same site. Although medical record review and clinical adjudication remain a gold standard, our claims-based algorithm provides an alternative approach to fracture ascertainment when high PPV is desired. © 2019 American Society for Bone and Mineral Research.