Effect of Discordant Hip Bone Density on Incident Fracture Risk: A Registry-Based Cohort Study

An Update on the Fracture Risk Assessment Tool: What Have We Learned over 15+ years?

The Fracture Risk Assessment Tool (FRAX) was launched in 2008 and uses clinical variables to estimate 10-year fracture risk. FRAX has been incorporated into clinical treatment guidelines and is well validated in specific disease states like chronic kidney disease. However, there are risk factors which are not captured by FRAX such as diabetes and falls. The use of race-ethnicity as a factor in FRAX is a source of controversy. Though other risk calculators exist, FRAX is likely to remain the gold standard for fracture risk prediction. An update of FRAX using data from a larger cohort is in development.

Fracture Risk Associated with Different Numbers and Combinations of Lumbar Vertebrae: The Manitoba BMD Registry

Bone mineral density (BMD) is widely used for assessment of fracture risk. For the lumbar spine, BMD is typically measured from L1-L4 as it provides the largest area for assessment with the best measurement precision. Structural artifact often confounds spine BMD in clinical practice, and the International Society for Clinical Densitometry (ISCD) recommends removing vertebrae with artifact when reporting spine BMD. In its most recent position statements, the ISCD recommended against the use of a single vertebra when reporting spine BMD but stated that further studies should be done. The current analysis was performed to compare the performance of BMD from different numbers and combination of vertebral levels on fracture prediction in a large clinical registry of DXA tests for the Province of Manitoba, Canada. The study population comprised 39,727 individuals aged 40 years and older (mean age 62.7 years, 91.0 % female) with baseline DXA after excluding those with evidence of structural artifact. Mean follow-up for ascertaining fracture outcomes was 8.7 years. Area under the curve (AUC) for incident fracture risk stratification was statistically significant regardless of the BMD measurement site or fracture outcome. AUC differences with the various numbers and combinations of vertebral levels including a single vertebral body were small (less than or equal to 0.01). More substantial AUC differences were seen for femoral neck and total hip BMD versus L1-L4 BMD, approaching 0.1 for hip fracture stratification. In summary, we found that using combinations of fewer than 4 vertebrae including individual lumbar vertebrae predicted incident fractures. Importantly, differences between these different combinations were small when compared with L1-L4. Spine BMD was a better predictor of incident spine fracture compared to the hip, whereas the hip was better for hip fracture and overall fracture prediction.

Classification of Fracture Risk in Fallers Using Dual-Energy X-Ray Absorptiometry (DXA) Images and Deep Learning-Based Feature Extraction

Dual-energy X-ray absorptiometry (DXA) scans are one of the most frequently used imaging techniques for calculating bone mineral density, yet calculating fracture risk using DXA image features is rarely performed. The objective of this study was to combine deep neural networks, together with DXA images and patient clinical information, to evaluate fracture risk in a cohort of adults with at least one known fall and age-matched healthy controls. DXA images of the entire body as, well as isolated images of the hip, forearm, and spine (1488 total), were obtained from 478 fallers and 48 non-faller controls. A modeling pipeline was developed for fracture risk prediction using the DXA images and clinical data. First, self-supervised pretraining of feature extractors was performed using a small vision transformer (ViT-S) and a convolutional neural network model (VGG-16 and Resnet-50). After pretraining, the feature extractors were then paired with a multilayer perceptron model, which was used for fracture risk classification. Classification was achieved with an average area under the receiver-operating characteristic curve (AUROC) score of 74.3%. This study demonstrates ViT-S as a promising neural network technique for fracture risk classification using DXA scans. The findings have future application as a fracture risk screening tool for older adults at risk of falls. © 2023 The Authors. JBMR Plus published by Wiley Periodicals LLC on behalf of American Society for Bone and Mineral Research.

Adjusting Trabecular Bone Score (TBS) for Level-Specific Differences Reduces FRAX®-Based Treatment Reclassification in Patients with Vertebral Exclusions: The Manitoba BMD Registry

Trabecular bone score (TBS) is a FRAX®-independent risk factor for fracture prediction. TBS values increase from cranial to caudal, with the following mean differences between TBSL1-L4 and individual lumbar vertebrae: L1 -0.093, L2 -0.008, L3 +0.055 and L4 +0.046. Excluding vertebral levels can affect FRAX-based treatment recommendations close to the intervention threshold. We examined the effect of adjusting for level-specific TBS differences in individuals with vertebral exclusions due to structural artifact on TBS-adjusted FRAX-based treatment recommendations. We identified 71,209 individuals aged ≥40 years with TBS and FRAX calculations through the Manitoba Bone Density Program. In the 24,428 individuals with vertebral exclusions, adjusting TBS using these level-specific factors agreed with TBSL1-L4 (mean difference -0.001). We compared FRAX-based treatment recommendations for TBSL1-L4 and for non-excluded vertebral levels before and after adjusting for level-specific TBS differences. Among those with baseline major osteoporotic fracture risk ≥15 %, TBS with vertebral exclusions reclassified FRAX-based treatment in 10.6 % of individuals compared with TBSL1-L4, and was reduced to 7.2 % after adjusting for level-specific differences. In 11,131 patients where L1-L2 was used for BMD reporting (the most common exclusion pattern with the largest TBS effect), treatment reclassification was reduced from 13.9 % to 2.4 %, respectively. Among individuals with baseline hip fracture risk ≥2 %, TBS vertebral exclusions reclassified 7.1 % compared with TBSL1-L4, but only 4.5 % after adjusting for level-specific differences. When L1-L2 was used for BMD reporting, treatment reclassification from hip fracture risk was reduced from 9.2 % to 5.2 %. In conclusion, TBS and TBS-adjusted FRAX-based treatment recommendations are affected by vertebral level exclusions for structural artifact. Adjusting for level-specific differences in TBS reduces reclassification in FRAX-based treatment recommendations.

FRAX® Adjustment Using Renormalized Trabecular Bone Score (TBS) from L1 Alone may be Optimal for Fracture Prediction: The Manitoba BMD Registry

Lumbar spine trabecular bone score (TBS) used in conjunction with FRAX® improves 10-year fracture prediction. The derived FRAX risk adjustment is based upon TBS measured from L1-L4, designated TBSL1-L4-FRAX. In prior studies, TBS measurements that include L1 and exclude L4 give better fracture stratification than L1-L4. We compared risk stratification from TBS-adjusted FRAX using TBS derived from different combinations of upper lumbar vertebral levels renormalized for level-specific differences in individuals from the Manitoba Bone Density Program aged >40 years with baseline assessment of TBS and FRAX. TBS measurements for L1-L3, L1-L2 and L1 alone were calculated after renormalization for level-specific differences. Corresponding TBS-adjusted FRAX scores designated TBSL1-L3-FRAX, TBSL1-L2-FRAX and TBSL1-FRAX were compared with TBSL1-L4-FRAX for fracture risk stratification. Incident major osteoporotic fractures (MOF) and hip fractures were assessed. The primary outcome was incremental change in area under the curve (ΔAUC). The study population included 71,209 individuals (mean age 64 years, 89.8% female). Before renormalization, mean TBS for L1-3, L1-L2 and L1 was significantly lower and TBS-adjusted FRAX significantly higher than from using TBSL1-L4. These differences were largely eliminated when TBS was renormalized for level-specific differences. During mean follow-up of 8.7 years 6745 individuals sustained incident MOF and 2039 sustained incident hip fractures. Compared with TBSL1-L4-FRAX, use of FRAX without TBS was associated with lower stratification (ΔAUC = -0.009, p < 0.001). There was progressive improvement in MOF stratification using TBSL1-L3-FRAX (ΔAUC = +0.001, p < 0.001), TBSL1-L2-FRAX (ΔAUC = +0.004, p < 0.001) and TBSL1-FRAX (ΔAUC = +0.005, p < 0.001). TBSL1-FRAX was significantly better than all other combinations for MOF prediction (p < 0.001). Incremental improvement in AUC for hip fracture prediction showed a similar but smaller trend. In conclusion, this single large cohort study found that TBS-adjusted FRAX performance for fracture prediction was improved when limited to the upper lumbar vertebral levels and was best using L1 alone.

Trabecular Bone Score Vertebral Exclusions Affect Risk Classification and Treatment Recommendations: The Manitoba Bmd Registry

Lumbar spine trabecular bone score (TBS), a texture measure derived from spine dual-energy x-ray absorptiometry (DXA) images, is a bone mineral density (BMD)-independent risk factor for fracture. Lumbar vertebral levels that show structural artifact are excluded from BMD measurement. TBS is relatively unaffected by degenerative artifact, and it is uncertain whether the same exclusions should be applied to TBS reporting. To gain insight into the clinical impact of vertebral exclusion on TBS, we examined the effect of lumbar vertebral exclusions in routine clinical practice on tertile-based TBS categorization and TBS adjusted FRAX-based treatment recommendations. The study population consisted of 71,209 individuals aged 40 years and older with narrow fan-beam spine DXA examinations and retrospectively-derived TBS. During BMD reporting, 34.3% of the scans had one or more vertebral exclusions for structural artifact. When TBS was derived from the same vertebral levels used for BMD reporting, using fixed L1-L4 tertile cutoffs (1.23 and 1.31 from the McCloskey meta-analysis) reclassified 17.9% to a lower and 6.5% to a higher TBS category, with 75.6% unchanged. Reclassification was reduced from 24.4% overall to 17.2% when level-specific tertile cutoffs from the software manufacturer were used. Treatment reclassification based upon FRAX major osteoporotic fracture probability occurred in 2.9% overall, but in 9.6% of those with baseline risk ≥15%. For treatment based upon FRAX hip fracture probability, reclassification occurred in 3.4% overall, but in 10.4% in those with baseline risk ≥2%. In summary, lumbar spine TBS measurements based upon vertebral levels other than L1-L4 can alter the tertile category and treatment recommendations based upon TBS-adjusted FRAX calculation, especially for those close to or exceeding the treatment cut-off. Manufacturer level-specific tertile cut-offs should be used if vertebral exclusions are applied.