The present study demonstrated a marked effect of teriparatide given weekly on the occurrence of new vertebral fractures, with a relative risk reduction of 79 %, consistent with previous findings in this cohort , and others relating to both weekly  and daily teriparatide . However, in this current analysis, there was no evidence of a significant interaction between fracture probability and efficacy in the cohort as a whole. There was a small but statistically significant interaction between efficacy and baseline fracture probability calculated with inclusion of BMD, in the sense that marginally lower efficacy was evident at higher fracture probabilities. This appeared to be a consequence of an interaction between efficacy and BMD, driven by the subgroup of participants in the lowest 10 % of BMD and highest 10 % baseline fracture risk.
Earlier studies of raloxifene [13, 14] and strontium ranelate  yielded similar findings with respect to a lack of variation in efficacy according to baseline fracture probability. However, they contrast with results from reanalysis of two other phase III studies [11, 12]. In the first of these, a marked trend toward greater fracture reduction at higher baseline fracture probability was observed in a 3-year prospective, randomized, placebo-controlled trial of oral clodronate in elderly women, identified from general practice registers. When BMD was excluded from the calculation of probability, the interaction was statistically significant , and there was some evidence of a threshold with efficacy apparent at fracture probabilities exceeding 20 %. These data are complemented by findings of a significant effect of bazedoxifene on clinical fractures with fracture probabilities that exceeded 17 % . More recently, evidence of a similar interaction between efficacy and baseline fracture probability has been reported for denosumab .
In the current analysis, there was evidence, albeit not achieving statistical significance, of greater efficacy among younger participants, consistent with other recent findings . Indeed, we also found, in the subgroup in whom DXA BMD measures were included in the FRAX calculation, that efficacy appeared lower in those with lowest BMD (and who tended to be older). Interestingly, the treatment-induced increase in lumbar spine BMD in this trial somewhat lower than that in the pivotal study of the daily medication (6.4 versus 9.7 %) . However, given that our analysis was within one trial and the constraints of a non-head-to-head comparison and different study populations, this is unlikely to explain our results. As the efficacy-BMD interaction was driven primarily by those individuals in the top 10 % of fracture probability and lowest 10 % of BMD, with efficacy in the remainder of the population similar across the distribution and in those with or without BMD measures, we view this observation as of interest but requiring confirmation. Additionally, the definition of incident vertebral fracture differed between the TOWER trial and the pivotal trial of daily teriparatide. Although a similar SQ approach was employed in both trials, in the latter study, new fractures were only considered in previously normal vertebrae, and analyses conducted separately for all and moderate/severe fractures; in the TOWER trial, worsening of pre-existing fractures was considered if the change was by at least one SQ grade and by >20 %. These different approaches may lead to variation in the magnitude of treatment effect demonstrated, but as they are consistent between treatment and placebo groups within each study, there is no reason to suppose that they would influence interactions between baseline risk and treatment .
Patients included in the TOWER trial of weekly teriparatide had relatively high pre-treatment fracture risks. Indeed, only 3.6 % had a baseline probability of major fracture ≤10 %. It is possible, therefore, that any attenuation of efficacy with low fracture probabilities might be seen in patients at much lower risk than those recruited to TOWER. In the present analysis, the mean 10-year probability of a major fracture calculated using FRAX including BMD was 26 %. This compares with 21 % in the raloxifene study , in which no fracture probability-efficacy interaction was documented, but contrasts with the lower value of 10.9 % in the analysis of bazedoxifene , in which an interaction between antifracture efficacy and baseline fracture probability was observed. However, in the clodronate study , where an interaction was found, the mean probability was 18 %, although the overall range of probabilities was greater. Further analyses of phase 3 studies with a mean baseline fracture probability toward the lower part of the overall distribution may help to clarify such findings.
There are a number of limitations to this study that should be considered in the interpretation of the results. First, the TOWER Trial included relatively few incident fractures (n = 44), with the number of vertebral fractures in study participants with a DXA BMD test even fewer (n = 27). Indeed, the criterion for low bone mass was the equivalent of a T-score of −1.67, rather than −2.0 or −2.5, the threshold commonly used in phase 3 studies. However, this approach represents the country norm in Japan, and since we examined relationships across BMD in a continuous fashion, this should not have affected our results; the higher BMD criterion may have reduced our capacity to elucidate relationships between BMD and efficacy at the very lowest T-scores. Furthermore, although the relationship between metacarpal radiogrammetry and vertebral fracture risk is uncertain in this population, the results in the subset who underwent radiogrammetry instead of DXA appeared similar to those in the whole trial cohort. Second, the sole outcome of the TOWER study was vertebral fracture. Thus, it is uncertain whether the same relationship between efficacy and fracture probability would be observed with the inclusion of other fracture outcomes, although congruence between vertebral and nonvertebral fracture outcomes has been documented in other studies [12, 15]. Third, several FRAX risk variables (parental history of hip fracture, current smoking, and alcohol intake) were not recorded in the original TOWER trial. These were set to missing in the current analysis, but, owing to the randomized allocation of patients to treatment or placebo, this is unlikely to have biased baseline risk calculations differentially across the two groups. However, this may have resulted in an underestimation of fracture probability overall, and the possibility of a differential effect across groups remains. Finally, the 10-year probability of a major osteoporotic fracture rather than hip fracture was used as the index of fracture risk, since it was found to be more closely related to the fracture occurrence. However, in a sensitivity analysis using baseline probability for hip fracture, results were extremely similar.
In conclusion, this analysis has demonstrated, consistent with previous findings, that treatment with teriparatide once weekly is associated with a marked decrease in morphometric vertebral fractures compared to treatment with placebo. However, in the cohort as a whole, there was no evidence of an interaction between efficacy and baseline fracture probability. Although the potential interaction between efficacy and DXA BMD might support an initiation strategy predicated on stratification by baseline BMD, these findings require replication in further studies, ideally both of daily and weekly teriparatide, before they can be confidently incorporated into any formal policy. Overall, these findings do not provide support for the stratification of weekly teriparatide prescription according to baseline fracture risk.