Can Hip Fracture Prediction in Women be Estimated beyond Bone Mineral Density Measurement Alone?

Prediction of osteoporotic hip fracture in postmenopausal women through patient-specific FE analyses and machine learning

A great challenge in osteoporosis clinical assessment is identifying patients at higher risk of hip fracture. Bone Mineral Density (BMD) measured by Dual-Energy X-Ray Absorptiometry (DXA) is the current gold-standard, but its classification accuracy is limited to 65%. DXA-based Finite Element (FE) models have been developed to predict the mechanical failure of the bone. Yet, their contribution has been modest. In this study, supervised machine learning (ML) is applied in conjunction with clinical and computationally driven mechanical attributes. Through this multi-technique approach, we aimed to obtain a predictive model that outperforms BMD and other clinical data alone, as well as to identify the best-learned ML classifier within a group of suitable algorithms. A total number of 137 postmenopausal women (81.4 ± 6.95 years) were included in the study and separated into a fracture group (n = 89) and a control group (n = 48). A semi-automatic and patient-specific DXA-based FE model was used to generate mechanical attributes, describing the geometry, the impact force, bone structure and mechanical response of the bone after a sideways-fall. After preprocessing the whole dataset, 19 attributes were selected as predictors. Support Vector Machine (SVM) with radial basis function (RBF), Logistic Regression, Shallow Neural Networks and Random Forest were tested through a comprehensive validation procedure to compare their predictive performance. Clinical attributes were used alone in another experimental setup for the sake of comparison. SVM was confirmed to generate the best-learned algorithm for both experimental setups, including 19 attributes and only clinical attributes. The first, generated the best-learned model and outperformed BMD by 14pp. The results suggests that this approach could be easily integrated for effective prediction of hip fracture without interrupting the actual clinical workflow.

Assessment of finite element models for prediction of osteoporotic fracture

With increasing life expectancy and mortality rates, the burden of osteoporotic hip fractures is continually on an upward trend. In terms of prevention, there are several osteoporosis treatment strategies such as anti-resorptive drug treatments, which attempt to retard the rate of bone resorption, while promoting the rate of formation. With respect to prediction, several studies have provided insights into obtaining bone strength by non-invasive means through the application of FE analysis. However, what valuable information can we obtain from FE studies that have focused on osteoporosis research, with respect to the prediction of osteoporotic fractures? This paper aims to fine studies that have used FE analysis to predict fractures in the proximal femur through a systematic search of literature using PUBMED, with the main objective of supporting the diagnosis of osteoporosis. The focus of these FE studies is first discussed, and the methodological aspects are summarized, by mainly comparing and contrasting their meshing properties, material properties, and boundary conditions. The implications of these methodological differences in FE modelling processes and propositions with the aim of consolidating or minimalizing these differences are further discussed. We proved that studies need to start converging in terms of their input parameters to make the FE method applicable to clinical settings. This, in turn, will decrease the time needed for in vitro tests. Current advancements in FE analysis need to be consolidated before any further steps can be taken to implement engineering analysis into the clinical scenario.

Homeostatic Expansion of CD4+ T Cells Promotes Cortical and Trabecular Bone Loss, Whereas CD8+ T Cells Induce Trabecular Bone Loss Only

Background

Bone loss occurs in human immunodeficiency virus (HIV) infection but paradoxically is intensified by HIV-associated antiretroviral therapy (ART), resulting in an increased fracture incidence that is largely independent of ART regimen. Inflammation in the bone microenvironment associated with T-cell repopulation following ART initiation may explain ART-induced bone loss. Indeed, we have reported that reconstitution of CD3+ T cells in immunodeficient mice mimics ART-induced bone loss observed in humans. In this study, we quantified the relative effects of CD4+ and CD8+ T-cell subsets on bone.

Methods

T-cell subsets in T-cell receptor β knockout mice were reconstituted by adoptive transfer with CD4+ or CD8+ T-cells subsets were reconstituted in T-cell receptor β knockout mice by adoptive transfer, and bone turnover, bone mineral density, and indices of bone structure and turnover were quantified.

Results

Repopulating CD4+ but not CD8+ T cells significantly diminished bone mineral density. However, micro-computed tomography revealed robust deterioration of trabecular bone volume by both subsets, while CD4+ T cells additionally induced cortical bone loss.

Conclusions

CD4+ T-cell reconstitution, a key function of ART, causes significant cortical and trabecular bone loss. CD8+ T cells may further contribute to trabecular bone loss in some patients with advanced AIDS, in whom CD8+ T cells may also be depleted. Our data suggest that bone densitometry used for assessment of the condition of bone in humans may significantly underestimate trabecular bone damage sustained by ART.

Changes in bone mineral density and bone turnover in patients on 'drug holiday' following bisphosphonate therapy: real-life clinic setting

OBJECTIVES

Treatment discontinuation after long-term bisphosphonate (BP), termed a 'drug holiday', has been proposed to reduce the risk of BP-associated complications. The duration of treatment cessation remains unclear. Changes in bone mineral density (BMD), bone turnover markers (BTMs) and their relationship with FRAX were assessed to help determine the optimum length of a 'drug holiday'.

METHODS

A retrospective analysis of 134 patients (13M, 121F) aged [mean (SD)] 68·4 (8·2) years who discontinued BPs after treatment for 5·9 (3·0) years for osteoporosis was undertaken. BMD at the lumbar spine (LS), total hip (TH), and femoral neck (FN) and biochemical parameters including serum 25 (OH) vitamin D, bone turnover markers (plasma CTX, P1NP) and FRAX scores were determined at discontinuation, 12-18 months and 24-30 months off treatment.

RESULTS

BMD decreased significantly at the LS [% change mean (SD): -0·94 (3·6), P = 0·008], TH [-1·4 (2·4), P < 0·001] and FN [-1·8 (4·4), P < 0·001] after treatment discontinuation for 12-18 months. In the subgroup who remained off treatment for 24-30 months, a progressive decline in BMD was seen at the TH and FN with total % decrease of -2·52 (3·5) and -2·7 (4·76), P < 0·001, respectively. CTX and P1NP increased significantly at 12-18 months after discontinuation [% change CTX: 95 (88), P < 0·001, P1NP: 88 (73), P < 0·001]. FRAX scores were significant predictors of % change in BMD at the FN (P < 0·05), independently of bone turnover and vitamin D status. In summary, our data show that following a 'drug holiday', the use of DEXA scans, BTMs and FRAX may help guide when to resume treatment.

A biomechanical sorting of clinical risk factors affecting osteoporotic hip fracture

Osteoporotic fracture has been found associated with many clinical risk factors, and the associations have been explored dominantly by evidence-based and case-control approaches. The major challenges emerging from the studies are the large number of the risk factors, the difficulty in quantification, the incomplete list, and the interdependence of the risk factors. A biomechanical sorting of the risk factors may shed lights on resolving the above issues. Based on the definition of load-strength ratio (LSR), we first identified the four biomechanical variables determining fracture risk, i.e., the risk of fall, impact force, bone quality, and bone geometry. Then, we explored the links between the FRAX clinical risk factors and the biomechanical variables by looking for evidences in the literature. To accurately assess fracture risk, none of the four biomechanical variables can be ignored and their values must be subject-specific. A clinical risk factor contributes to osteoporotic fracture by affecting one or more of the biomechanical variables. A biomechanical variable represents the integral effect from all the clinical risk factors linked to the variable. The clinical risk factors in FRAX mostly stand for bone quality. The other three biomechanical variables are not adequately represented by the clinical risk factors. From the biomechanical viewpoint, most clinical risk factors are interdependent to each other as they affect the same biomechanical variable(s). As biomechanical variables must be expressed in numbers before their use in calculating LSR, the numerical value of a biomechanical variable can be used as a gauge of the linked clinical risk factors to measure their integral effect on fracture risk, which may be more efficient than to study each individual risk factor.

Assessing bone quality in terms of bone mineral density, buckling ratio and critical fracture load

Background

Bone mineral density (BMD) is used as a sole parameter in the diagnosis of osteoporosis. Due to the ease of acquirement of BMD, clinical diagnosis still involves its usage although the limitations of BMD are quite well-established. Therefore, this preliminary study hoped to reduce the errors introduced by BMD alone by incorporating geometric and structural predictors simultaneously to observe if strength was implicitly dependent on the geometry and BMD. Hence, we illustrated the triadic relationship between BMD, buckling ratio (BR) and critical fracture load (F_cr).

Methods

The geometric predictor was the BR as it involves both the changes in the periosteum and the cortical thickness. Also, structural changes were monitored by finite element (FE) analysis-predicted F_cr. These BR and F_cr measurements were plotted with their respective femoral neck BMD values in elderly female patients (n=6) in a 3-year follow-up study, treated with ibandronate.

Results

In all the three-dimensional plots (baseline, mid and final year), high F_cr values were found at regions containing high BMD and low BR values. Quantitatively, this was also proven where an averaged highest F_cr across the three years had a relatively higher BMD (46%) and lower BR (19%) than that of the averaged lowest F_cr. The dependence of FE predicted strength on both the geometry and bone density was illustrated.

Conclusions

We conclude that use of triadic relationships for the evaluation of osteoporosis and hip fractures with the combination of strength, radiology-derived BR and bone density will lay the foundation for more accurate predictions in the future.

Novel assessment tools for osteoporosis diagnosis and treatment

This review describes new technologies for the diagnosis and treatment, including fracture risk prediction, of postmenopausal osteoporosis. Four promising technologies and their potential for clinical translation and basic science studies are discussed. These include reference point indentation (RPI), Fourier transform infrared spectroscopy (FTIR), Raman spectroscopy, and magnetic resonance imaging (MRI). While each modality exploits different physical principles, the commonality is that none of them require use of ionizing radiation. To provide context for the new developments, brief summaries are provided for the current state of biomarker assays, fracture risk assessment (FRAX), and other fracture risk prediction algorithms and quantitative ultrasound (QUS) measurements.

How accurately can we predict the fracture load of the proximal femur using finite element models?

BACKGROUND

Current clinical methods for fracture prediction rely on two-dimensional imaging methods such as dual-energy X-ray absorptiometry and have limited predictive value. Several researchers have tried to integrate three-dimensional imaging techniques with the finite element (FE) method to improve the accuracy of fracture predictions. Before FE models could be used in clinical settings, a thorough validation of their accuracy is required. In this paper, we try to evaluate the current state of accuracy of subject-specific FE models that are used for prediction of the fracture load of proximal femora.

METHODS

All the studies that have used FE for prediction of fracture load and have compared the predicted fracture load with experimentally measured fracture loads in vitro are identified through a systematic search of the literature. A quantitative analysis of the results of those studies has been carried out to determine the absolute prediction error, percentage error, and linear correlations between predicted and measured fracture loads.

FINDINGS

The reported coefficients of determination (R(2)) vary between 0.773 and 0.96 while the percentage error in prediction of fracture load varies between 5 and 46% with most studies reporting percentage errors between 10 and 20%.

INTERPRETATION

We conclude that FE models, which are currently used only experimentally, are in general more accurate than clinically used fracture risk assessment techniques. However, the accuracy of FE models depends on the details of their modeling methodologies. Therefore, modeling procedures need to be optimized and standardized before FE could be used in clinical settings.

Pitfalls in the external validation of FRAX

SUMMARY

Recent studies have evaluated the performance of FRAX® in independent cohorts. The interpretation of most is problematic for reasons summarised in this perspective.

INTRODUCTION

FRAX is an extensively validated assessment tool for the prediction of fracture in men and women. The aim of this study was to review the methods used since the launch of FRAX to further evaluate this instrument.

METHODS

This covers a critical review of studies investigating the calibration of FRAX or assessing its performance characteristics in external cohorts.

RESULTS

Most studies used inappropriate methodologies to compare the performance characteristics of FRAX with other models. These included discordant parameters of risk (comparing incidence with probabilities), comparison with internally derived predictors and inappropriate use and interpretation of receiver operating characteristic curves. These deficits markedly impair interpretation of these studies.

CONCLUSION

Cohort studies that have evaluated the performance of FRAX need to be interpreted with caution and preferably re-evaluated.

Determinants of fracture risk in a UK-population-based cohort of older women: a cross-sectional analysis of the Cohort for Skeletal Health in Bristol and Avon (COSHIBA)

BACKGROUND

Identification of individuals with high fracture risk from within primary care is complex. It is likely that the true contribution of falls to fracture risk is underestimated.

METHODS

Cross-sectional analysis of a population-based cohort of 3,200 post-menopausal women aged 73 ± 4 years. Self-reported data were collected on fracture, osteoporosis clinical risk factors and falls/mobility risk factors. Self-reported falls were compared with recorded falls on GP computerised records. Multivariable logistic regression was used to identify independent risk factors for fracture.

RESULTS

A total of 838 (26.2%) reported a fracture after aged 50; 441 reported falling more than once per year, but 69% of these had no mention of falls on their computerised GP records. Only age [odds ratios (OR): 1.37 per 5 year increase, 95% confidence interval (CI): 1.23-1.53], height (1.02 per cm increase, 95% CI: 1.01-1.04), weight (OR: 0.99 per kg increase, 95% CI: 0.98-0.99) and falls (OR: 1.49 for more than once per year compared with less, 95% CI: 1.13-1.94) were independent risk factors for fracture. Falls had the strongest association.

CONCLUSION

When identifying individuals with high fracture risk we estimate that more than one fall per year is at least twice as important as height and weight. Furthermore, using self-reported falls data is essential as computerised GP records underestimate falls prevalence.

Official Positions for FRAX® clinical regarding falls and frailty: can falls and frailty be used in FRAX®? From Joint Official Positions Development Conference of the International Society for Clinical Densitometry and International Osteoporosis Foundation on FRAX®

Risk factors for fracture can be purely skeletal, e.g., bone mass, microarchitecture or geometry, or a combination of bone and falls risk related factors such as age and functional status. The remit of this Task Force was to review the evidence and consider if falls should be incorporated into the FRAX® model or, alternatively, to provide guidance to assist clinicians in clinical decision-making for patients with a falls history. It is clear that falls are a risk factor for fracture. Fracture probability may be underestimated by FRAX® in individuals with a history of frequent falls. The substantial evidence that various interventions are effective in reducing falls risk was reviewed. Targeting falls risk reduction strategies towards frail older people at high risk for indoor falls is appropriate. This Task Force believes that further fracture reduction requires measures to reduce falls risk in addition to bone directed therapy. Clinicians should recognize that patients with frequent falls are at higher fracture risk than currently estimated by FRAX® and include this in decision-making. However, quantitative adjustment of the FRAX® estimated risk based on falls history is not currently possible. In the long term, incorporation of falls as a risk factor in the FRAX® model would be ideal.