In this issue of JBMR, Curtis colleagues1 use data from the placebo group of a randomized, controlled osteoporosis trial to argue for a possible healthy adherer effect. Our understanding of this effect began when some randomized, controlled clinical trials in cardiovascular disease had cardiovascular outcomes that contradicted findings in prior observational studies.2 These findings highlighted the fact that several sources of bias can occur in observational studies. These biases can be related to patient behaviors, including both healthy user and healthy adherer effects. The best known examples of these biases have appeared in studies of cardiovascular disease. For example, a large observational study, the Nurses' Health Study, showed decreased coronary heart disease in women taking hormone-replacement therapy (HRT) in 1985,3 a finding that was confirmed by other observational studies.4 However, more than 10 years later, two randomized, controlled clinical trials contradicted the observational findings: The Heart and Estrogen/Progestin Replacement Study (HERS) showed no cardiovascular benefit from estrogen supplementation,5 and the Women's Health Initiative showed increased incidence of cardiovascular events in women taking estrogen.6
How can we explain this divergence of results? First, we must identify and understand this divergence by understanding sources of bias.2 The first potential bias, the healthy user effect, occurs when patients who choose to receive one preventive service such as estrogen or a lipid-lowering agent also choose to receive other preventive services such as immunizations and screening tests. For example, patients who adhere to statins more frequently receive vaccinations and screening services.7 These same patients also may have less comorbidity and better functional status,8 increasing the likelihood of other healthy behaviors These patients also may be involved in fewer accidents,9 suggesting possible risk aversion. A second potential bias—the healthy adherer bias—was first demonstrated in 1980 when decreased mortality and fewer cardiovascular events were found in good adherers of placebo in a randomized clinical trial of a cardiovascular agent.10 Subsequent randomized clinical trials of cardiovascular agents supported this finding,11, 12 and it was confirmed again in a meta-analysis of the placebo arms from eight studies containing 19,633 participants.13 These findings of decreased mortality in good adherers of placebo may be explained by the so-called healthy adherer effect. Here, patients who adhere to a specific therapy also may be more likely to engage in healthy behaviors and/or to adhere to other medications.14
Do these same biases occur when we look at patients with osteoporosis and use fracture events as outcomes? The results of two studies—one an observational study and the other a randomized clinical trial—are of interest in answering this question. The recent observational study by Cadarette and colleagues15 showed support for a healthy adherer effect for only one class of osteoporosis medications. The authors found a significant reduction in the risk of nonvertebral fractures for secondary prevention in good adherers with bisphosphonates. However, they did not find effects for good adherers to raloxifene or calcitonin for either primary or secondary prevention. Interestingly, these medications are not associated with nonvertebral fracture efficacy in randomized clinical trials either. Cadarette and colleagues also found that good adherers of osteoporosis medications were greater users of preventive health services, including bone density testing.15 These observational study results,15 however, may have been confounded by selective prescribing.16, 17 Patients who receive different osteoporosis medications likely have different risks of fracture. If the risk of fracture is lower with one medication and there are fewer fracture events, it may be more difficult to see fracture reduction. If patients who are given one medication have lower fracture risk, over time there will be fewer fracture events than in patients given another medication who are at higher risk. If fewer fracture events occur in patients given one medication, it will be more difficult to show statistically significant fracture reduction with that medication.
The second study, which Curtis and colleagues1 report in this issue of JBMR, focuses on placebo participants in a randomized, placebo-controlled clinical trial, the Fracture Intervention Trial (FIT). They defined good adherers as those with more than 80% adherence, as reported in patient diaries. Although they saw a numerical reduction in hip fracture rate among good adherers in the placebo group, this reduction was not statistically significant. Further, they found no reduction in clinical fractures or wrist fractures. If Curtis and colleagues1 are correct in attributing their findings—that placebo participants have lower hip fracture rates because of the healthy adherer effects—then only certain types of fracture events may benefit from the proposed behavioral differences in good adherers. Is this biologically plausible? Are patients who take osteoporosis medications like patients who adhere to statins in that they may take fewer risks and have fewer accidents?9 If this is so, we also would expect fewer nonvertebral nonhip fractures in the good adherers.
If good adherence to placebo treatments in osteoporosis trials does result in reduced fracture rates as Curtis and colleagues1 suggest, is this also related to other healthy bone behaviors, such as exercise and calcium/vitamin D intake? In a study of the impact of patient education on behavioral change in osteoporosis, we found that adherers to one healthy bone behavior such as exercise or calcium or taking an osteoporosis medicine did not necessarily adhere with other related health behaviors.18
In conclusion, the suggestions of Curtis and colleagues1 in this issue that some fracture outcomes in a randomized clinical trial may be influenced by adherence as a possible marker for other healthy behaviors are provocative but need to be confirmed using data from other randomized osteoporosis trials where adherence was measured. This is especially important in relationship to osteoporosis medications because other studies, such as that of Cadarette and colleagues,15 have not shown this effect, and the study of Curtis and colleagues1 had no statistically significant reductions in fractures. Meanwhile, we should continue to critically evaluate results from observational trials in osteoporosis for sources of bias.