I’ve been skeptical about the value of the nationwide effort for osteoporosis screening. So I went back to the initial studies on the treatment of osteoporosis with alendronate. More correctly, it’s not treatment of osteoporosis, but rather medications to prevent osteoporotic fractures.
I haven’t looked at all the studies, but focused on double blinded randomized controlled trials, that are multicenter with large sample sizes. I found these studies by looking at the citations in the Cochrane review.
I did this to fulfill my curiosity, but also to practice critically appraising the literature. I’m not an expert, but have been using the advice and checklist provided by Dr. Vinay Prasad. It’s a simple checklist:
Triage:
Does the study pertain to my practice/interest?
Is the study randomized or observational?
Is it multicenter or single center?
Is the sample size large or small?
Is the primary outcome measuring a clinical endpoint or surrogate marker?
If the paper passes the triage questions, it warrants a closer look to answer these questions:
What was the intervention?
Is the control arm what you would have done?
What was the effect size?
Clinical or surrogate endpoint?
What happened after the trial ended?
Any games with patient selection?
This checklist may be limited to high level evaluation of studies, and I may be at risk of missing nuance. But I think it’s effective for getting a sense of whether or not the study is good and potentially practice changing. It also helps to quickly dismiss studies that are of lower quality.
The study I looked at first is:
Black DM, Cummings SR, Karpf DB, Cauley JA, Thompson DE, Nevitt MC, Bauer DC, Genant HK, Haskell WL, Marcus R, Ott SM, Torner JC, Quandt SA, Reiss TF, Ensrud KE. Randomised trial of effect of alendronate on risk of fracture in women with existing vertebral fractures. Fracture Intervention Trial Research Group. Lancet. 1996 Dec 7;348(9041):1535-41. doi: 10.1016/s0140-6736(96)07088-2. PMID: 8950879.
Does the study pertain to my practice/interest? Yes
Is the study randomized or observational? Randomized
Is it multicenter or single center? Multicenter
Is the sample size large or small? Large
Is the primary outcome measuring a clinical endpoint or surrogate marker? Surrogate marker
This study didn’t pass the initial triage checklist. This study seemed like it would be one of the “landmark” practice changing studies for the treatment of osteoporosis. Yet, it didn’t pass triage.
I proceeded anyway for practice. As I’ll eventually find out I wouldn’t be able to find any studies that show bisphosphonates (or any medication indicated for osteoporosis) effectively reduce the risk of clinical fractures in patients at risk.
What was the intervention? Alendronate 5 mg, increased to 10 mg at 24 months
Is the control arm what you would have done? Yes
What was the effect size? New vertebral fractures 145 (15%) women in placebo vs 78 (8%) women in the alendronate group. P-value <0.001
Clinical or surrogate endpoint? Surrogate
What happened after the trial ended? N/A (I don’t quite understand this one)
Any games with patient selection? N/A (I don’t quite understand this one)
Relative to other RCTs supporting the use of bisphosphonates for the primary prevention of osteoporotic fractures, I thought this was one of the more convincing ones despite its limitations. This study showed a decreased risk of vertebral fractures with fair effect size ( RRR 47% , ARR 7%). In the secondary analyses, there was also a significant decrease in clinical fractures.
The main limitation of this study is the primary endpoint was vertebral fractures found on screening, not on clinical presentation. Fractures found on x rays done at predetermined time intervals, which are presumably asymptomatic fractures, make this metric a surrogate rather than a clinical one.
Patients may not care about slight decreases in the vertebral height that don’t cause pain. Patients care more about decreasing clinical fractures e.g. fractures that lead to pain, surgery, disability, or death.
The study did find a significant decrease in clinical fractures, but that was in their secondary analyses. My understanding is that significant differences in secondary endpoints can be hypothesis generating, but not practice changing.
Using these data to support nationwide screening and pharmacotherapy would rely on the assumption that decreasing asymptomatic vertebral fractures can be extrapolated to decreasing risk of clinical fractures. It remains unclear if patients taking alendronate are better off compared to those who don’t.
Further, one of the inclusion criteria were women with radiographic evidence of vertebral fracture. In practice, screening for vertebral fractures is not standard of care. This study may not generalize to treating women who have been screened only with a DXA scan.
Despite the impressive effect size, I’m not sure how I would apply these findings to my practice. For a study to change practice, it would need to compare an intervention to placebo in women who have screened positive on DXA. Then the primary endpoint would measure differences in clinical fractures i.e. osteoporotic fractures that presented as pain or disability. I haven’t been able to find this study yet. Therefore, I remain skeptical that incentivizing DXA screenings is valuable.
I’m not an expert, and I haven’t spent hours combing through the literature. But I was surprised at how useful a simple checklist for appraising a medical trial could be. It allowed me to quickly triage a long list of trials and prioritize the ones that are likely to matter. And in those trials, I could quickly see the limitations and draw my own conclusions. Review of the scientific literature is no longer as intimidating, but engaging.
If I find a vertebral fracture on X-ray spine in a post menopausal woman, even asymptomatic. I will not do DEXA, I shall start with Therapy for osteoporosis, as the evidence of osteoporosis is sufficient for me to counsel and initiate therapy.
Amazing work! I appreciate seeing you apply his framework and applying it to something original. Inspires me to try the same.