Should we really compare absolute risk reduction in different trials on osteoporosis: comment on the article by Ringe JD and Doherty JG
- First Online:
- Cite this article as:
- Belaya, Z.E. Rheumatol Int (2011) 31: 1669. doi:10.1007/s00296-010-1626-8
- 53 Views
After reading the article “Absolute risk reduction in osteoporosis: assessing treatment efficacy by number needed to treat” by Ringe and Doherty , many people would probably feel sheer and utter astonishment at the “great idea” of a “simplified approach”… Indeed, how is it possible that relative risk for vertebral fractures 0.59 (95% CI 0.48–0.73) in the SOTI trial (strontium ranelate)  would turn into a nice number needed to treat (NNT)—9 “to reduce one vertebral fracture”, but the relative risk 0.5 (95% CI 0.26–0.66) for intermittent ibandronate in the BONE clinical trial  is 21 postmenopausal women to be treated to avoid one vertebral fracture. Who is responsible for this? The agent or the populations that were involved in the clinical trials?
The number needed to treat entirely depends on the absolute risk of the event in the studied population. For example, NNT = 40, if aspirin is used to prevent one death at 5 weeks in studied patients after myocardial infarction , but if our intention is to treat fever with aspirin, not to prevent death, NNT could be 2. At the same time, NNT to prevent one death with aspirin is expected to be higher in populations with a lower rate of death, for instance, among younger patients without myocardial infarction.
Mean BMD (T-score)
The rate of patients with new vertebral fractures at year 3
Bone  Ibandronate
Femoral neck −2.0
Lumbar spine −2.8
94% had one previous vertebral fracture 44% had two previous vertebral fractures
9.6% in the placebo group, 4.9% in the treatment group
Soti  Strontium ranelate
Femoral neck −2.8
Lumbar spine −3.5
100% had previous vertebral fracture and the mean number of previous vertebral fractures was 2.16 ± 2.18
32.8% in the placebo group, 20.9% in the treatment group
Thus, the risk of vertebral fracture in the SOTI trial was three times higher in the placebo group (the event rate in the placebo group was 32.8% SOTI vs 9.6% BONE during the 3 years), which entirely reflects the least number of women needed to be treated over 3 years. Indeed simplifying our approach, as the authors suggested, to subtraction, we calculate the absolute risk reduction—11.9% for strontium and—4.7% for ibandronate, which leads to a very simple conclusion 11.9 > 4.7 and the higher the absolute figure the less NNT (1/0.119 < 1/0.047) which seems nice and obvious. However, by applying the “advance technology” of division, we could see that ibandronate was effective in preventing the event in half of the patients (9.6% placebo vs. 4.9% treatment), but strontium ranelate was helpful only in 1/3 of the treated patients (32.6% placebo vs. 20.9% treatment). Thus, ibandronate was effective in a larger proportion of patients, but even if we had the agent that prevents 100% of fracture in a population with absolute risk 9.6%, NNT would be just 10… Vice versa suggesting we have the agent X with the fracture rate in the placebo group 32.6 and 27.9% in the treatment line. The absolute risk reduction would be the same 4.7%; therefore, NNT would be also the same, but the agent X is effective only in 14.4% of cases. That raises a question: “Which drug is more desirable for the healthcare system? The one that was effective in preventing the same event in 50% (100%) of the cases or in 30% (14.4%)?” If we still consider that the results of the BONE  and SOTI  trials can be compared using NNT, it would be equally honest to say that ibandronate is four times more effective than strontium ranelate because the fracture rate in the treatment group is 4.9% ibandronte versus 20.9% strontium ranelate. Just a “simplified” approach from the other side….
The second part of the article however is a completely bizarre. The author compares NNT in the subpopulation analysis in the TROPOS trial (strontium ranelate)  with primary analysis in the Horizon trial (zoledronic acid) … “simplifying” the difference between the subpopulation analysis and first endpoint of the study, not even mentioning the difference in statistical analysis and not providing the differences in the studied population with the exception of BMD. At least it should be mentioned that according to the achieved statistical significant result (RR = 0.64; 95% (0.412–0.997); P = 0.046) in the subpopulation analysis, strontium could be effective in preventing hip fracture (NNT = 40), but only for a certain group of women: mean age 80 with mean T-score −3.55 Femoral Neck .
Nevertheless, why do we always use relative risk reduction (RRR)? Is it really because of an “..artificially high..” figure? Obviously, the relative risk lets us be more or less independent of time and the event rate in the population and that is actually why anti-osteoporotic treatments are generally compared using RRR of fracture. The RRR is also appropriate for meta-analyses  where it is necessary to make data less heterogeneous. Therefore, this seems to be the best way to compare the efficacy of antiresorbtive agents and strontium ranelate. Indeed, the Cochrane Collaboration published their results on the efficacy of strontium ranelate for the treatment of osteoporosis that lead to silver level evidence to support the efficacy of strontium ranelate for the reduction of vertebral fractures (RR 0.63 95% CI (0.56–0.71)) and, to a lesser extent, non-vertebral (RR 0.86 95% CI (0.75–0.98) . The cross-design synthesis for results of different randomized controlled clinical trials on bisphosphonates was published recently revealing a statistically significant reduction in vertebral (0R 0.413 (95% CI 0.279–0.612), non-vertebral (0.796 (0.739–0.858)), hip (0.711 (0.616–0.820)) and all fractures (0.762 (0.680–0.855)) with almost the same efficacy in “real life” between complaint and non-complaint patients .
Thus, if we really desire to compare the efficacy of antiresorbtive agents and strontium, it should be −59% RRR bisphosphonates (BP) versus −37% RRR strontium ranelate (SR) to prevent vertebral fractures; −21% BP versus −14% SR for non-vertebral fractures. Interestingly, calcitriol and alfacalcidol, which are not considered as first-line therapy for osteoporosis, proved to be effective in reducing non-vertebral fractures by 49% (six trials OR—0.51 (95% CI 0.30–0.88) .
Considering the evidence for hip fracture reduction as a primary endpoint of the trial, currently only zoledronic acid  and denosumab  proved to be effective. Nevertheless, the results of meta-analyses for several clinical trials on BP proved that all BP are also effective , but not strontium ranelate (not as a primary end-point, nor by meta-analysis). Only in the subpopulation analysis was strontium shown to be effective for hip fracture prevention , and this is certainly a lower level of evidence.
And last, but not least, the authors classified strontium ranelate as a bone-forming agent. However, where is the evidence? The direct comparison of the bone-forming agent teriparatide and strontium ranelate revealed the slight antiresorbtive effects of strontium ranelate −14% PINP at the 3rd month (P < 0.005) and −19% of βCTx at the 6th month, whereas the markers rose significantly by more than 50% after teriparatide treatment . Blake GM et al., suggested that the therapeutic effect of strontium could be explained by the strontium content even without interference in bone remodeling . Indeed, not more than 1.5 g of calcium is usually recommended for postmenopausal women to consume daily. However, 2 g of strontium is advised daily without hesitation. Thus, classifying 2 g of strontium ranelate daily as a bone-forming agent, just like 20 mkg of teriparatide daily, seems lacking in common sense especially since knowing that the impressive image of microCT bone is just the 3D version of the technical distortion of the X-ray method because strontium has a higher atomic number than calcium and its presence is the cause of the huge rise in BMD  (volumetric or square).
Hence, the whole article is misleading. The limitations claimed by the authors are completely inadequate as they are far greater than admitted. The differences in the absolute fracture risk among the populations of the clinical trials analyzed can be immediately noticed just from the event rates for fractures in the placebo groups that differ dramatically. More accurately, NNT = 9 reflects the highest risk of vertebral fractures in the patients enrolled in the SOTI trials and not the best efficacy of the compound. The “simplified approach” for directly comparing NNT in populations with a different absolute risk or NNT in an analysis of different subpopulations is biased, grotesque and can never, ever be recommended either for routine clinical practice or for justifying how the health care budget is spent on the treatment of osteoporosis.