Overview of design and data sources
To assess anti-osteoporosis medication utilization patterns and to identify baseline factors associated with alternative definitions of discontinuation, we conducted a retrospective cohort study. To identify factors precipitating discontinuation among adherent patients, we conducted a case-crossover analysis, nested within the cohort study [17]. Both the cohort and the case-crossover analyses used 2006–2009 data on a national 5% random sample of Medicare beneficiaries, obtained from the Centers for Medicare and Medicaid Services (CMS) Chronic Condition Data Warehouse [18]. For each beneficiary, data included information on demographic and insurance coverage; claims for inpatient, outpatient, skilled nursing facility, noninstitutional provider, home health, hospice, durable medical equipment services; and claims for prescription drugs. The institutional review board of the University of Alabama at Birmingham approved the study.
Cohort study design
Eligible subjects for the cohort study were Medicare beneficiaries who were 65 years of age or older; lived in the US; were enrolled continuously in traditional Medicare fee-for-service and pharmacy coverage (Parts A, B and D) and not in a Medicare Advantage plan for at least 13 continuous months (12 months baseline and at least one month follow-up time); and were newly treated with a bisphosphonates. Bisphosphonate medications included infusion (IV) zoledronic acid, IV ibandronate, oral ibandronate, oral alendronate, or oral risedronate, during the period 2007 to 2009. We defined new treatment with a particular bisphosphonate as therapy initiated after a baseline period of 12 months during which no bisphosphonate prescription was filled and no administration of IV bisphosphonate occurred. We defined the “original” bisphosphonate as the one taken at initiation. We excluded patients who filled a prescription for raloxifene, calcitonin or teriparatide during baseline and for each beneficiary, follow-up began on the earliest treatment initiation date during the study period and ended on the earliest of first date of losing full Medicare coverage, the death date, or 12/31/2009.
We examined utilization patterns during a follow-up period of 12 months, during which full Medicare coverage was required. We classified patients’ medication status at the end of 12 months of follow-up into four mutually exclusive groups: 1) continued original bisphosphonate use without a treatment gap, 2) discontinued without switching to a different osteoporosis medication or restarting the original bisphosphonate, 3) restarted the same bisphosphonate after a treatment gap or 4) switched to another drug (a different bisphosphonate, raloxifene, calcitonin or teriparatide). We defined a treatment gap as a period of at least 90 days, occurring after the end of the days supplied by a prescription or administration of a particular bisphosphonate drug and during which there was no further prescription fill or administration of the drug. We assigned days supplied as 90 days for IV ibandronate (administered at three-month intervals) and 365 days for IV zolendronic acid (administered annually). Based on the medication status at the end of 12 months of follow-up, we defined discontinuation using two definitions in the analysis evaluating the association between baseline characteristics and discontinuation of bisphosphonate therapy (Figure 1). Definition I classified a beneficiary as discontinued only if s/he had discontinued all anti-osteoporosis drugs as of the end of follow up. Definition II classified a beneficiary as discontinued if s/he had discontinued all anti-osteoporosis drugs as of the end of follow up, or switched to another type of bisphosphonate, or switched to a non-bisphosphonate anti-osteoporosis medication, or restarted the same anti-osteoporosis drug after a treatment gap of ≥ 90 days. Calcitonin, raloxifene, teriparatide were considered as different anti-osteoporotic medications during follow-up. For example, if a patient restarted the same bisphosphonate after a treatment gap, and then switched to calcitonin until the end of follow-up, we classified the patient as not discontinued in definition I, but discontinued in definition II. To assess longer-term adherence and switching, we conducted similar analyses among patients who had at least 30 months of follow-up, classifying their medication status at the end of 30 months of follow-up using the same definitions of discontinuation (Additional file 1).
Case-crossover study design
The case-crossover design enables studying the association between transient exposure, and an acute event by comparing exposure in a period shortly before the event with the patient’s usual frequency of that exposure [19]. Because control information for each subject is based on his or her own past exposure experience in this design, results are not confounded by risk factors that are stable over time [19–21].
To identify and quantify factors that may precipitate discontinuation to anti-osteoporosis medication, we evaluated each cohort member’s adherence to bisphosphonates during the first 12 months of treatment using the medication possession ratio (MPR), calculated as the total days of bisphosphonate exposure during the first 12 months of follow-up divided by 365 days, and then classified a subject as adherent to bisphosphonate therapy if s/he had an MPR ≥ 80%. The case-crossover analysis included subjects from the cohort study who were adherent to their original bisphosphonate during the first 12 months after treatment initiation, without switching, stopping and/or restarting their original bisphosphonate at any time during the first 12 months follow-up, who then discontinued their original bisphosphonate therapy (i.e., experienced a gap of 90 days or longer) during the subsequent follow-up (i.e., > 12 months after treatment initiation). We excluded those who had a hospitalization or skilled nursing home stay in the first 90 days after discontinuation in order to minimize misclassification, because claims data may not accurately measure drugs used during inpatient and skilled nursing facility stays (21). The follow-up period for identifying discontinuation began 12 months after initiation and continued until the earliest of the loss of full Medicare coverage, discontinuation, death, or 12/31/2009. For each eligible subject, we defined the “case” (or “hazard”) period as the 30 days immediately before bisphosphonate discontinuation, and we specified five “control” periods as the five 30-day periods immediately before the hazard period (Figure 2) [17, 21, 22].
Factors possibly related to discontinuation
Factors measured during the baseline period for the cohort study and analyzed for possible association with different definitions of discontinuation included demographics, history of co-morbid conditions, history of preventive service and other medical service utilization, eligibility for a low income subsidy and use of prescription drugs other than anti-osteoporosis medications. Demographic characteristics included age, gender, race, geographic region and residential area-level income, all assessed as of the start of follow-up. We estimated income, as a proxy for socioeconomic status, using beneficiaries’ nine-digit ZIP codes linked to their 2000 Census block group of residence and the corresponding median household income [23]. We identified osteoporosis and history of fractures using International Classification of Disease (ICD)-9th revision diagnosis codes specific to the particular fracture sites from inpatient and physician encounter claims. ICD9 codes for osteoporosis included 733.0, 733.00, 733.01, 733.02, 733.03, 733.09, and for fracture included 800.xx-829.xx, and 733.1x. Other medical conditions, which may associated with fracture and consequently with anti-osteoporosis therapy patterns, included glucocorticoid-related and fall-related (predisposing to falls) conditions, diabetes, chronic kidney disease, depressive illness, acute myocardial infarction, other heart disease, metabolic bone disease and cancer [23]. We identified these conditions using the criterion of at least one diagnosis from inpatient or physician evaluation/management claims during baseline. Medications possibly associated with medication discontinuation or risk of fractures included anticonvulsants, antidepressants, antipsychotics, antihypertensives, lipid-lowering drugs, non-steroidal anti-inflammatory drugs, steroids, proton pump inhibitors, H2-receptor blockers, hormone replacement therapy, thiazolidinediones, and aromatase inhibitors. We identified these medications using the criterion of at least one claim for a prescription filled in Medicare Part D data during baseline.
Factors considered as possibly precipitating medication discontinuation in the case-crossover analysis (see detailed algorithm codes in Additional file 2) included (DXA) testing, skilled nursing home stays, hospitalization for any cause, entering the part D coverage gap [24], having a rheumatologist or endocrinologist visit, having an oncologist visit, eligibility for a low income subsidy, the occurrence of fracture, malignancy, upper gastrointestinal disease, adverse events (including osteonecrosis of the jaw, atrial fibrillation, esophageal cancer, renal disease, subtrochanteric or femoral shaft fractures), total number of physician visits, number of different medications, the Charlson comorbidity index [25], total Medicare costs and out-of-pocket drug costs. The first eight factors were dichotomous, whereas the rest had multiple categories based on their distribution during the last 30 days of the first 12 months of treatment.
For analyses evaluating the effect of baseline factors on discontinuation, the outcomes were the two alternative definitions of discontinuation at the end of 12 months following bisphosphonate initiation. For the analysis of factors precipitating discontinuation of anti-osteoporosis therapy, we selected as cases only those who totally discontinued bisphosphonate use at the end of follow-up.
Analysis
We evaluated new bisphosphonate treatment patterns during 2007–2009 graphically for each bisphosphonate including branded alendronate without vitamin D, branded alendronate with vitamin D, generic alendronate, risedronate, oral ibandronate, IV ibandronate and IV zoledronic acid. We identified the number of new users of each specific bisphosphonate drug, calculated the proportion of each drug among all new users of bisphosphonates for each quarter of the study period, and calculated the proportion by medication status as of the end of 12 months of follow-up period.
We compared the baseline characteristics by medication status of new bisphosphonate users at the end of 12 months follow up and computed the standardized difference scores to evaluate the magnitude of differences between groups. We considered standardized difference scores that were less than 0.1 as clinically unimportant [26]. We conducted logistic regression analyses to identify baseline characteristics associated with the two different definitions of discontinuation at the end of 12 months of follow-up. Finally, we repeated all analyses to identify baseline characteristics associated with the two different definitions of discontinuation for patients who had at least 30 months of follow-up at the end of 30 months. These analyses used the odds ratio (OR) with its 95% confidence interval (CI) as the measure of association. Statistical significance was assessed at the alpha level of 0.05.
In the case-crossover analysis, we compared exposures in a subject’s hazard period (30 days immediately before discontinuation) and control periods (five 30-day periods immediately before the hazard period) [17, 21, 22]. In sensitivity analyses, we compared exposures in the 30 days immediately before discontinuation and four, three, two and one 30-day periods immediately before the hazard period respectively. We also carried out an additional sensitivity analysis in which the hazard and control time windows were extended to 60 days: that is, this analysis included one hazard period, consisting of the 60 days immediately before discontinuation, and two 60-day control periods immediately before the hazard period, so that the study period is as long as the main analysis We used conditional logistic regression models to compute ORs for the association between factors of interest and discontinuation for the case-crossover analysis. All statistical analyses were performed in SAS 9.3 (SAS institute).