Did Prescription Bias Affect Outcomes in a Study of the Relative Effectiveness of Osteoporosis Drugs?

IN RESPONSE:

We agree with and appreciate the comments by Drs. Beri and Khattri. They highlight our finding that calcitonin recipients were sicker than alendronate recipients in terms of measured variables. They also point out that bisphosphonate dosing instructions are complex, and thus frail patients may have been preferentially treated with calcitonin, administered by daily nasal spray. We agree that residual confounding could explain part of our findings, and we discussed limitations of administrative claims data in our article. Results from our theoretical sensitivity analysis showed that the higher fracture risk among calcitonin recipients compared with alendronate recipients is probably not due to unmeasured confounding. However, because our analysis was based on the assumption of a single unmeasured confounder (1), multiple unmeasured confounding factors may have collectively introduced bias into the results. Nonetheless, we did adjust for many factors associated with frailty, such as age, history of falls, vertebral fractures, Parkinson disease, dementia, comorbidity score, and medication use. We also completed 7 different subgroup analyses, all with results similar to our main findings: no large difference in nonvertebral fracture rates between risedronate or raloxifene and alendronate recipients, and higher nonvertebral fracture risk among calcitonin recipients than among alendronate recipients. However, we did document more fractures among raloxifene recipients than among alendronate recipients in the subgroup with previous fracture. We therefore emphasize caution in interpreting findings comparing calcitonin and alendronate, as well as results comparing raloxifene and alendronate, because raloxifene recipients were apparently healthier in terms of measured variables. Our results comparing bisphophonates are more compelling because risedronate and alendronate recipients were similar in terms of measured risk factors for fracture; however, we cannot rule out potential differences in unmeasured factors.