Concerns regarding the critical evaluation of 3D-Shaper

We read with interest the recent review by Dr Whitmarsh1; however, we believe that it presents an unbalanced and, at times, potentially misleading analysis of the 3D-Shaper® software. First, the author’s description of his role as “the inventor and developer” of the methods discussed in the review reflects an overstated and misleading representation of the collaborative and multi-institutional nature of the original academic research.2 Furthermore, the author has not been involved in subsequent technological developments since 2012, including the cortical and trabecular analysis methods developed by the 3D-Shaper Medical team,3 which are extensively discussed in the review. The author hypothesizes that the number of modes underlying the 3D-Shaper’s statistical model may be insufficient, noting that this information was not specified in prior publications. We clarify that, consistent with our earlier developments,4 the number of retained modes was determined using Cattell’s scree test,5 resulting in 12 shape modes and 11 density modes. The author expresses concerns that “cortical parameters are not directly measured”. This reflects a misunderstanding of model-based reconstruction approaches, as all 3D-Shaper parameters are estimated. The critical question is how accurate these estimates are. Agreement between 3D-Shaper and QCT has been assessed in multiple studies.6,–11 Additional pooled analyses of data from several studies6–8 are currently in preparation for publication, and they suggest agreement between 3D-Shaper and QCT. Collectively, these studies included data from 742 adults and 77 cadaver specimens. Standard statistical metrics were used to evaluate both systematic and random errors, as well as the correlation between 3D-Shaper- and QCT-derived parameters. At the total hip, correlation coefficients (R) ranged from 0.84 to 0.96 for integral vBMD,6–9,11 from 0.79 to 0.93 for trabecular vBMD,6–8,11 from 0.84 to 0.93 for cortical sBMD,7,8 from 0.87 to 0.92 for cortical thickness,6 and from 0.91 to 0.94 for cortical vBMD.6 The moderate correlations (from 0.48 to 0.62) reported for cortical vBMD in the STRUCTURE trial likely reflect methodological differences, “QCT-based MIAF software and 3D-SHAPER software using different approaches for cortical segmentation, region of interest definition, and the definition of the trabecular bone”.11 The author relies on one of these accuracy studies6 to suggest that, “if an individual’s cortical bone appears to change, it remains unclear whether this arises from genuine physiological change, or from the inherent 3D-DXA measurement error”. Assessing longitudinal changes requires consideration of both accuracy and precision. However, the review omits discussion of the 3D-Shaper’s precision data. A key publication reported that 3D-Shaper and DXA demonstrate comparable performance in detecting longitudinal changes, with trend assessment intervals at the total hip of 2.9 yr for integral vBMD, 2.6 yr for trabecular vBMD, and 3.5 yr for cortical sBMD, compared with 2.7 yr for areal BMD (aBMD).12 The precision of 3D-Shaper measurements was further confirmed in a more recent study.13 Throughout the article, the author repeatedly frames the correlation between aBMD and 3D-Shaper parameters as a point of concern, a view similarly reported in previous research.14 However, this correlation is entirely expected because of a well-established physiological principle: individuals with higher overall bone density, as indicated by aBMD, also tend to have higher cortical and trabecular bone densities. Building on this point, the review states: “if aBMD increases, the 3D-DXA analysis will show simultaneous increases in trabecular BMD, cortical BMD, and cortical thickness”. Multiple key publications, however, have demonstrated that 3D-Shaper cortical and trabecular parameters can behave differently—for example, one increasing while the other decreases or remains unchanged.11,15–18 These findings show that 3D-Shaper provides information beyond aBMD, thereby contradicting the review’s assertion. In the 3D-Shaper sub-analysis of the ACTIVE trial,16 subjects receiving 18 mo of abaloparatide or teriparatide experienced a similar and significant increase in trabecular vBMD versus baseline, but “only abaloparatide significantly increased cortical vBMD”. No significant change in cortical vBMD was observed with teriparatide. Two other 3D-Shaper studies from Germany and Japan confirmed that teriparatide significantly increased trabecular vBMD, while cortical vBMD either significantly decreased or showed no significant change.17,18 In the 3D-Shaper sub-analysis of the STRUCTURE trial,11,15 patients previously treated with bisphosphonates and receiving 12 mo of teriparatide showed a significant increase in trabecular vBMD (p < .001) whilst simultaneously showing a significant decrease in cortical vBMD from baseline (p < .001). Although the statistical analysis in the original article focused on group differences, we report here the significance from baseline, assessed using repeated measures models adjusting for treatment, visit, baseline serum β-isomer of the CTx (β-CTx) value, baseline value, DXA-machine type, treatment-by-visit interaction, and baseline value-by-machine type interaction. These findings align with the QCT analysis of a comparable group of subjects from the STRUCTURE trial, which likewise showed a significant increase in trabecular vBMD and a significant decrease in cortical vBMD. Although differences in the magnitude of these changes were observed, they may be attributable to the “QCT-based MIAF software and 3D-SHAPER software using different approaches for cortical segmentation, region of interest definition, and the definition of the trabecular bone”.11 Collectively, these studies consistently reported significant increases in trabecular vBMD, accompanied by either significant decreases or no significant changes in cortical vBMD in patients receiving teriparatide. These findings demonstrate that 3D-Shaper evaluates cortical and trabecular bone changes beyond aBMD, thus invalidating the hypothesis presented in this review that changes in 3D-DXA cortical and trabecular parameters merely mirror those in aBMD. Finally, although findings may vary across studies evaluating the same treatments, such differences often reflect study design or subject characteristics. Notably, significant differences have also been observed in QCT-based studies evaluating the same treatments, where additional sources of variability arise from the specific QCT calibration phantom and image analysis software employed,19 thereby challenging the notion of QCT as a gold standard for cortical and trabecular measurements. The review’s description of the available versions of 3D-Shaper is inaccurate or outdated. There are 2 independent products available. 3D-Shaper Research (version 2.14) provides a comprehensive set of 71 bone parameters and is intended exclusively for investigational use. The color-coded 3D representations and cross-sectional images mentioned are for group-level analyses, not individual patient assessment. As such, statistically significant changes can be detected and interpreted across cohorts—even when the effect size is smaller than the absolute measurement precision—provided that the sample size is sufficiently large. In contrast, the clinical version 2.14 of 3D-Shaper computes only 3 key parameters at the total hip: integral vBMD, trabecular vBMD, and cortical surface BMD (sBMD). These 3 measurements are the only ones intended to support bone health evaluation in clinical settings. Furthermore, the review’s description of regulatory requirements does not reflect current regulatory processes worldwide, suggesting that agencies such as the FDA would only require “a correlation study” to clear 3D-Shaper as a medical device. Regulatory clearance requires the submission of comprehensive documentation, including software validation and verification data, supporting scientific evidence, and quality management records, which, in many territories, must also be upheld through periodic reviews. 3D-Shaper has consistently undergone rigorous quality and regulatory assessments, earning a CE marking (Europe) and regulatory clearances from authorities such as the US FDA, Japan’s PMDA, and Brazil’s ANVISA. We are confident that this letter addresses the main concerns raised and affirms that the 3D-Shaper software is a safe, effective, and validated tool suitable for use in clinical practice—within its intended use and where cleared for use—and in clinical research. Ludovic Humbert (Conceptualization, Methodology, Project administration, Resources, Supervision, Validation, Writing—original draft, Writing—review & editing), Sam Wilson (Conceptualization, Formal analysis, Funding acquisition, Methodology, Project administration, Resources, Supervision, Validation, Writing—original draft, Writing—review & editing) None declared. L.H. and S.W. are employees and shareholders of 3D-Shaper Medical. Data sharing is not applicable to this letter, as no datasets were generated or analyzed during the current study.