Among women participating in FIT who were randomized to placebo, we did not find significant associations between compliance with placebo and fractures. However, we found that high compliance with placebo was associated with reduced total-hip bone loss, and a similar trend was observed for changes in femoral neck BMD. Furthermore, compared with women with lower compliance with placebo, the risk of hip fracture was 33% lower among women with high compliance with placebo; this relationship did not reach statistical significance and was not observed for any other clinical fracture type. These findings suggest that the effect of compliance with placebo on hip BMD, and perhaps hip fracture risk, was not attributable to confounding by the known fracture risk factors collected in FIT. In total, our findings provide some support for the existence of the healthy adherer effect in this population and suggest that medication compliance may be a proxy for behaviors and/or other factors that confer hip BMD (and possibly hip fracture) benefit independent of the effect of the medication.
The possible protective effect on hip fracture may be mediated at least in part by changes in BMD, recognizing that the correlation between change in BMD and fracture benefit is only modest.(14
) Comparing women with high versus lower compliance with placebo, the annualized difference in change in hip BMD between the two groups was −0.15%. To provide some context for this difference, these changes in BMD are comparable with differences in BMD among depressed women compared with those without depressive symptoms(15
) and with the magnitude of BMD loss among women using selective serotonin reuptake inhibitors (SSRIs) compared with nonusers.(16
In analyses that evaluated the dynamic nature of compliance, we observed that compliance varied before and after fractures. At least in a randomized, controlled trial (RCT), the occurrence of a clinical fracture would appear to affect subsequent adherence with study medication, at least for some patients. This finding also has been reported in observational data from a nonclinical trial population.(1
) These results support the conclusion that analyzing compliance in a manner that ignores the time-dependent nature of adherence before and after fractures may yield biased results.
If the healthy adherer effect is generalizable to a nonclinical trial setting, our findings may have implications for future comparative effectiveness research (CER) of osteoporosis medications, at least for hip fracture outcomes. Analyses studying the relative effectiveness of medications that require intravenous administration (eg, zoledronic acid and denosumab) will need to carefully consider how to choose the most valid comparator groups. For example, patients receiving parenteral osteoporosis agents given by a health care provider consist of a mix of patients who would have variable degrees of compliance with oral bisphosphonates. Comparing these patients with patients who are highly compliant with oral bisphosphonates could yield biased results because the compliant oral bisphosphonate users may have better outcomes (including lower hip fracture risk) in part owing to the healthy adherer effect. In contrast, a comparator group of patients starting oral bisphosphonates, without considering whether they remained compliant, would be biased given their lesser exposure to bisphosphonates. These studies could, for example, consider compliance with other medications to at least partially address these concerns. Future work evaluating healthy behaviors and factors that are associated with compliance behavior are necessary to ensure unbiased results from observational comparative effectiveness analyses.
Although our results suggested a trend toward reduced hip fracture risk associated with placebo compliance, we did not observe similar trends for other fracture types, such as wrist or clinical vertebral fractures. Risk factors for wrist fractures are different from those for hip fractures,(17
) and wrist fractures generally occur in younger, healthier, and more active women. Factors associated with medication adherence may be more relevant for older, frailer patients at risk for hip fracture. Given that our crude and fully adjusted results were similar for each analysis, it would seem that the risk factors controlled for within this analysis are not important confounders of fracture risk associated with medication adherence. Furthermore, studies that show a fracture benefit associated with medication compliance may not be confounded if fracture types other than hip fractures are being evaluated.
The strengths of our study include a large population of women with median follow-up of more than 4 years. Fracture outcomes were adjudicated through medical record review, and longitudinal BMD data were available over the course of FIT. Compliance was assessed with high precision through the use of both patient diaries and pill counts. Despite these strengths, our results must be interpreted in light of some limitations. Modest numbers of outcome events and generally high compliance within FIT required dichotomizing compliance and resulted in wide confidence intervals for some of our results. Additionally, FIT was a clinical trial of women who were enrolled to study fracture outcomes, and thus these results may not be generalizable to other settings and patient populations. Last, some factors that may affect fracture risk were not collected in FIT, such as exercise, falls, depression, comorbidities, and measures of frailty. The association between compliance and smoking and poorer health status reported in suggest the possibility that these two factors may be proxies for one or more of unmeasured confounders.
Another potentially important factor that was not systematically measured in FIT was 25-hydroxyvitamin D. An ancillary study to FIT that sampled 1000 women at baseline showed that only a small proportion (2.3%) were deficient in vitamin D [25(OH)D ≤ 10 ng/mL) at baseline, and the response to alendronate was not affected by baseline 25(OH)D status.(18
) However, it is unknown whether low vitamin D at baseline or throughout the study might be associated with low compliance and could have affected for our results.
In conclusion, based on these data from a randomized, controlled trial, we found small but significant differences in the change in hip BMD between women with high versus low compliance with placebo. However, perhaps most important, studies reporting fracture risk reduction associated with high compliance with bisphosphonates do not appear to be confounded by healthy behaviors and factors associated with medication compliance except possibly for hip fracture. Further work is needed to assess the existence of a healthy adherer effect for fracture outcomes in other populations and how best to control for this potential confounder.