Post-fracture pharmacotherapy for women with osteoporotic fracture: analysis of a managed care population in the USA

We used de-identified, longitudinal patient information contained in the i3 Invision Datamart database (now known as the Clinformatics™ Data Mart [23]). This administrative claims dataset contains medical and pharmacy claims and laboratory results for a large geographically diverse patient population in the USA including approximately three million patients with osteoporosis. Available data included patient enrollment and administrative claims data for outpatient visits, hospitalizations, laboratory tests, and pharmacy claims. Disease diagnoses and comorbidities were identified in the dataset using the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnosis codes [24]. Medications were identified using the National Drug Code Directory (NDC) codes in the pharmacy claims [25] and J-codes for intravenous drugs in the medical claims.

Osteoporotic fractures were identified by ICD-9-CM diagnosis codes [24, 26] and categorized by site into four groups: vertebral fracture only, hip fracture only, non-hip/non-vertebral, and fractures at multiple sites.

We excluded patients with a record of fracture before the index date, a diagnosis of Paget’s disease (osteitis deformans; ICD-9-CM code 731.0) at any time in the database, or a diagnosis of malignant neoplasm (ICD-9-CM codes 140.xx to 208.xx, 230.xx to 239.xx, or 172.xx) during their 3-year study period.

We assessed patients’ age at the index date, insurance type, the presence or absence of 28 comorbidities (listed in the Online Resource Table S1), and the Charlson comorbidity index score to quantify overall comorbidity [27]. In addition, for the baseline period, we assessed the presence of bone mineral density tests (CPT codes 76070–76071, 76075–76076, 76078, 76977, and 77078–77083), the history of falls resulting in medical services (ICD-9-CM codes E880–E888) and the use of gastroprotective agents (proton pump inhibitors, histamine H2-receptor antagonists, and cytoprotectant agents), corticosteroids, estrogen, and non-steroidal anti-inflammatory drugs (NSAIDs).

We examined whether patients had received osteoporosis therapy during the baseline period, within 90 days post-fracture, and within 1 year post-fracture. Patients who had received baseline therapy were classified into two groups according to whether the days’ supply of medication overlapped the index date (currently treated) or did not overlap the index date (not currently treated). All forms of osteoporosis therapy, oral and parenteral, were considered in the study and included the bisphosphonates (alendronate, risedronate, ibandronate, zoledronic acid), calcitonin, denosumab, raloxifene, and teriparatide.

Dispensed prescriptions for all types of osteoporosis therapy were used to calculate medication possession ratio (MPR), namely, the proportion of days covered by dispensed medications. For the post-fracture period, the MPR was calculated as total days’ supply/365. If the total days’ supply exceeded 365 days, it was truncated to 365 days, yielding a maximum MPR of 1.0. The MPR for pre-index date therapy during the baseline period was calculated as the total days’ supply/730. We defined adherence to therapy as an MPR ≥0.80.

Baseline patient characteristics (including demographic and clinical parameters) were summarized and described overall and by index fracture group (vertebral, hip, non-hip/non-vertebral, and multiple fractures). Continuous variables were summarized by means and standard deviations (SDs), overall and by fracture group; comparisons were made using F-tests or Kruskal-Wallis tests where appropriate. Categorical variables were summarized by the number and percentage of patients, overall and by fracture group; comparisons were made using χ² tests. A statistical significance level of 5 % was used for all aforementioned analyses.

Multivariable logistic regression was used to identify the association between 1-year treatment status (treated vs. not treated within 1 year post-fracture) and patients’ baseline treatment status (not treated vs. previously/currently treated) among index fracture groups, adjusted for baseline patient characteristics. Interactions between the baseline treatment status and index fracture group were considered to estimate differential effects on 1-year treatment status among different treatment-fracture combinations. Using stepwise (forward and backward) variable selection with a tuning significance of 20 %, potentially important baseline patient characteristics were selected from a list of candidate covariates possibly associated with post-fracture treatment status. The covariates included age groups, health plan types, Charlson comorbidity index score, use of concomitant medications, and comorbid conditions, as listed in Online Resource Table S1. Only 13,380 (28.4 %) patients had a bone mineral density assessment, and the results were not available from the database; thus, bone mineral density was not used as an inclusion criterion or included in the model. The final model included the key predictors (baseline treatment status, index fracture group, and their interaction) and the selected factors. The associations with the 1-year treatment status were quantified in terms of odds ratios (ORs) with corresponding 95 % confidence intervals (CIs). All analyses were conducted using SAS software, version 9.3 (SAS Institute, Cary, NC).