Data source and study population
Administrative medical and pharmacy claims data from the Optum™ Research Database (ORD) were used to identify fracture rates during the period January 1, 2007, to May 31, 2017, in US commercial (private, for-profit) and Medicare Advantage (privately maintained, government-approved) health plan members. Data for approximately 66.3 million commercial and an additional 7.4 million Medicare Advantage enrollees (both men and women) were included in the ORD as of 2017. All data were maintained in a de-identified manner and were accessed following protocols compliant with the Health Insurance Portability and Accountability Act (HIPAA).
Patients aged ≥ 50 years with evidence of a “case-qualifying” fracture in the identification period (January 1, 2007, to May 31, 2017) were included in the analysis, based on a validated claims-based algorithm [14, 15] with a positive predictive value that exceeds 90% . Case-qualifying fractures were those identified during an inpatient stay (in any position on the medical claim) or in an outpatient setting accompanied by a fracture repair procedure code (e.g., surgery, kyphoplasty). Fracture diagnoses were based on primary or secondary International Statistical Classifications of Diseases and Related Health Problems, 9th Revision (ICD-9) or 10th Revision (ICD-10), listed on the same claim. Exclusion criteria included Paget’s disease of bone or malignancy (except non-melanoma skin cancer) at baseline (during the 12 months prior to index fracture) or during the first month of follow-up.
The first fracture during the identification period (January 1, 2007, to May 31, 2017) was considered the index fracture. Fracture of the spine, pelvis, shoulder, radius/ulna, carpal/wrist, hip, femur, tibia/fibula, ankle, and multiple sites in the above grouping were included in the analysis. Fractures that occurred within 30 days of the index date were considered index fractures (i.e., part of the initial episode) and defined as fractures at multiple sites; fractures at different sites occurring more than 30 days after the index date were considered subsequent fractures. Episodes continued until a gap of 90 days was observed between consecutive claims.
An analysis was conducted to examine the trend in fracture rate over time. Age- and sex-adjusted fracture rates were calculated using the age and sex distribution of patients with an index date in 2016. The year 2016 was used because the observation period for 2017 was not a complete year. The fracture rate (number of fractures/1000 years of continuous enrollment) within each year was stratified by age and sex, then weighted to match the age and sex distribution observed in 2016. These stratified rates were condensed by year to generate a yearly age- and sex-adjusted fracture rate. The raw fracture rate for 2016 was equal to the age- and sex-adjusted fracture rate.
Incidence rate was reported as the number of events per 1000 person-years (py) of enrollment during each year of the identification period. The denominator population comprised of members who were ≥ 50 years of age during the year of interest. The number of days the member was enrolled in the health plan during the year was calculated and was used to determine the total person-years (py) of enrollment for the denominator. Incidence rate is reported as number of events per 1000 py of enrollment during each year from 2007 to May 2017. Fracture rates were stratified by age category (50 to 64 vs. ≥ 65 years), sex (male vs. female), and fracture site (ankle, pelvis, radius/ulna, hip, shoulder, carpal/wrist, tibia/fibula, spine, femur, and multiple sites).
In order to test if there was a different trend over time in the rate of fractures between 2007 and 2013 vs. 2014 and 2017, we constructed generalized linear Poisson models with log link. We used three models: (1) females aged 65+ years, (2) males aged 65+ years, (3) all patients adjusted for age category (50–54, 55–59, 60–64, 65–69, 70–74, 75–79, 80–84, and 85+ years) and sex. For each model, data were aggregated for the unique combinations of calendar year, sex, and age category. The dependent variable was the sum of fractures within each of the combinations. The independent variables were calendar year for years 2007 to 2013 and a separate continuous variable for years 2014 to 2017, which allowed for an estimation of the rate ratio for each of the two time periods. The exposure (denominator) was the number of person-years of continuous enrollment in a qualifying medical plan for each of the combinations. Robust standard errors were used to calculate the statistical significance of the rate ratios. A rate ratio less than 1 indicates a reduction in fracture rates over time, while a rate ratio greater than 1 indicates an increase in fracture rates over time. Interpretation for each row is provided with the row.