(290) Outcome Misclassification: Differences in Hip Fracture Identification Using Administrative Claims Data

Publication Number: 964

Background: We identified little consistency in fracture outcome definitions in a prior scoping review of studies that utilized administrative claims data to estimate drug effects on fracture risk. Heterogeneity in outcome definitions complicates the comparison of results across studies; outcome misclassification can lead to biased estimates of fracture risk.

Objectives: To compare fracture rates between hip fracture definitions.

Methods: We linked emergency department (ED) and hospitalization (inpatient) claims data for the population of Ontario, Canada to compare differences in hip fracture outcome identification in a cohort of 113,452 long-term oral bisphosphonate users (2003/04 – 2016/12). Six definitions were compared that used diagnostic (ICD-10: S72.0x, S72.1x, S72.2x) only, or diagnostic and procedural (fixation, reduction, replacement) codes. Our definitions, from most to least inclusive were: 1) one ED or inpatient diagnosis, 2) one inpatient diagnosis, 3) one ED diagnosis, 4) ED or inpatient: one diagnosis and one procedure within 7 days of first claim, 5) one diagnosis and one procedure during the same hospitalization, 6) one diagnosis and one procedure in ED. The most inclusive definition (1) served as the referent group for comparisons. Fracture incidence censored on death and end of follow-up (2020/04) was the primary outcome. Incidence rates and number of hip fractures were used to quantify comparisons.

Results: The number of fractures identified ranged from 96 (definition 6) to 9,601 (definition 1). Definitions that only considered diagnoses in either hospital (definition 2) or ED (definition 3) identified 91-93% of fractures identified by definition 1. Interestingly, definitions that required procedural codes in addition to diagnostic codes within hospital only (definition 5), or hospital and ED (definition 4), captured only 87% of fractures relative to definition 1. At the extreme, < 1% of hip fractures identified using diagnostic codes in ED or inpatient were captured by diagnostic and procedural codes in ED (definition 6). Differences in definitions translated into differences in fracture incidence (per 10,000 person-years): 93.3, 89.8, 85.3, 79.5, 79.2 and 0.93 by each respective definition.

Conclusions: Differences in fracture identification impact incidence rates. Definitions that only required a diagnostic code are more inclusive than those requiring both diagnostic and procedural codes. Further research investigating the impact of fracture identification on exposure estimates of fracture risk is warranted.