Among women previously compliant with bisphosphonate therapy for at least two years, we found that the incidence rate of subsequent hip fracture was significantly lower than for a non-compliant population. Although we observed a relatively small number of hip fracture events in our primary analysis (n=71), which limited our ability to determine exactly when the rate of hip fracture increased, discontinuation of bisphosphonate for up to approximately 1 year did not appear to be associated with a significantly increased rate of hip fracture. Among women with baseline MPR 66–100% at two years (yielding the largest group of hip fractures), discontinuation of one year or longer was associated with a two to threefold increased relative risk of hip fracture after adjusting for a number of important covariates of interest including age, baseline health services utilization, and medical comorbidities. We also observed that with greater amounts of compliance (e.g., MPR 80–100%) and a longer duration of preceding bisphosphonate therapy, the incidence rate of hip fracture among persons having ceased bisphosphonate therapy was numerically lower than for women somewhat less compliant and having received a shorter prior course of therapy. However, they never reached the nadir in fracture rate achieved by women who remained on bisphosphonates.
The Fracture Intervention Long Term Extension (FLEX) trial evaluated 1099 women who had been treated with alendronate for approximately 5 years (78% of them who remained current users of study medication) and randomized them to placebo versus ongoing treatment with alendronate. Over the next five years, there were no significant differences in the incidence of hip fracture between those randomized to placebo versus active treatment in the intent-to-treat population. However, results from a subgroup analysis showed that those with osteoporotic BMD at the end of the five year initial treatment period had an increased rate of non-vertebral fractures after discontinuation, and those with non-osteoporotic BMD did not. In our study, we found that women in our study who were compliant with MPR 66–100% for 2 years had a significantly increased rate of hip fracture after discontinuation. We also found that women with higher baseline compliance (e.g., MPR≥80%) and those receiving bisphosphonates for at least three years did not have a significantly increased rate of fracture after discontinuation. This may reflect a biologic phenomenon whereby greater cumulative bisphosphonate exposure provides greater fracture protection. Unfortunately, our data source did not permit us to stratify subjects by BMD at the time of discontinuation.
Our results have high relevance to clinical practice since many women have a repeat bone mineral density (BMD) test at 2–3 years after starting bisphosphonate therapy, and the possibility of discontinuation may be considered. Our data suggest that this may be reasonably safe with respect to the risk for hip fracture for at least one year. For women who are compliant for longer periods of time, an even longer period of discontinuation may also be safe, although our data were limited to examine longer discontinuation intervals. However, if discontinuation is advised, the practicalities of when to resume therapy are problematic and potentially arbitrary. Longitudinal changes in BMD likely will be too minor to provide much near-term guidance. Bone turnover markers, although theoretically more useful, are not widely available or adequately reimbursed in most clinical settings. Moreover, there may be medico-legal implications if a patient has a fracture after bisphosphonates have been intentionally discontinued.
The strengths of our study include a large cohort of patients receiving care in routine clinical settings. The incidence rates of hip fracture that we observed are similar to those from RCTs and a large observational study that used similar methods to ours [11
]. Our large sample size permitted us to examine women compliant for two and three years using dynamic definitions of compliance based on MPRs that have been previously shown to be beneficial using bone mineral density (BMD) and fracture endpoints [8
]. Additionally, we evaluated discontinuation in several different ways and reached similar conclusions about associations between bisphosphonate cessation and fractures.
Our results must be interpreted in light of our study design. Perhaps most importantly, patients were not randomized to discontinue bisphosphonate therapy, and the reasons for discontinuation were unknown. It is possible that physicians recommended that patients discontinue based on their perceived low risk for future fracture. However, our results show that discontinuation was not associated with a lower hip fracture rate, as might be expected if predominantly low-risk patients were recommended to discontinue. Moreover, separate analyses of high risk patients (defined as those having a medical diagnosis for osteoporosis or a claim for a recent non-hip fracture) yielded results similar to those for the overall groups of subjects. Finally, we recognize that women who are non-adherent with any therapy are likely different in many ways than women adherent to therapy, and claims data captures only a few of these potential confounders. However, our restriction of the study population to only women that were adherent for at least two or three years should attenuate this concern, given their history of prior adherence.
Despite having more hip fractures in our primary analysis (n=71) than occurred in the recent FLEX trial (n=33 hip fractures), our results were likely underpowered to evaluate the risk of discontinuation in small, discrete increments of time after discontinuation and in the subcohort of women with high adherence for at least 3 years (n=28 hip fractures). This observation is illustrated in the relatively wide confidence intervals shown in and . Additionally, we acknowledge the potential misclassification of the exact time that patients discontinue given our use of pharmacy data that informs us only when medications were filled but not actual medication-taking behavior. However, it is unlikely that this misclassification extended beyond a few months, and we have previously shown high concordance between pharmacy data and actual medication taking behavior for osteoporosis medications [12
]. Additionally, we acknowledge that insurance claims for a hip fracture may misclassify actual events. However, in contrast to some other types of fracture such as vertebral fracture, misclassification of hip fracture in claims data has been previously shown to be uncommon [10
]. Moreover, we conducted a sensitivity analysis restricting outcomes to those with physician E/M (i.e., office visit) claims, which likely increased the specificity of our outcome definition and the corresponding positive predictive value of claims for hip fracture events. Finally, we recognize that the various bisphosphonates may have different skeletal retention times [13
], and the relative near-term safety of drug discontinuation may differ between them. Unfortunately, despite our large sample size, the number of hip fractures that occurred among patients meeting our inclusion criteria were too small to be able to examine individual differences between the bisphosphonates. However, among the oral bisphosphonates represented in our study, most patients were treated with alendronate, which appears to have the longer skeletal retention time of the two bisphosphonates that we studied. Assuming that longer skeletal retention allows for a longer duration of fracture risk reduction after bisphosphonate discontinuation, our results therefore may reflect a ‘best-case-scenario’ with respect to the relative safety of discontinuation. Finally, we evaluated only hip fractures and not other types of fractures. We chose this endpoint because we have previously demonstrated a high rate of misclassification of incident vertebral fractures in administrative claims data [14
], and the magnitude of benefit of bisphosphonate therapy to reduce non-hip, non-vertebral fractures is smaller than for hip fractures; thus, significant differences in fracture rates would likely be more difficult to detect.
We conclude that the risk for hip fracture among women that have previously been compliant with bisphosphonates for two years but whom subsequently discontinue use is increased. However, this risk appears to be attenuated with greater amounts of compliance and longer durations of preceding bisphosphonate therapy. With respect to the timing of this risk increases, our results indicate that this increase in risk appears to begin after approximately one year following discontinuation. This estimate may be overly conservative for women with very high compliance or treated with bisphosphonates for longer periods of time. Although an intentional drug holiday for up to this length of time may be safe, the practicalities of this approach may still outweigh potential benefits.