The University Institutional Review Board approved the protocol, and all subjects gave informed consent. We recruited postmenopausal women with osteoporosis from a convenience sample of patients who already had dual-energy X-ray absorptiometry and vertebral fracture assessment or lateral spine radiographs. One investigator (NBW) was aware of the subjects’ fracture status but was blinded to the BSA test results; the other investigators were aware of the BSA test results but blinded about the fracture status.

BMD results for spine, femoral neck (FN), and total hip (TH) were obtained using Hologic equipment performed within 12 mo of the study entry. FRAX scores were calculated using version 3.8 without BMD and with lowest ever FN (BMD_FN). Vertebral fractures were assessed from lateral spine images acquired using dual-energy X-ray absorptiometry or lateral spine X-rays; reduction in anterior, middle, and/or posterior vertebral height by ≥20% constituted a fracture (45). Seven of the subjects had 1 or more vertebral fractures, whereas 22 had no vertebral fractures.

CBAM capacity outcome, damping (ζ), was quantified from acceleration measurements at 3 anatomical sites (3,35) (Fig. 1): (1) below the knee at the tibial tuberosity (ζ_BELOWKNEE-R and ζ_BELOWKNEE-L; R for right heel strike and L for left heel strike), (2) above knee at the lateral femoral condyle (ζ_ABOVEKNEE-R and ζ_ABOVEKNEE-L) for each leg, and (3) upper back (T-7) (ζ_UPPERBACK-R and ζ_UPPERBACK-L). Low mass skin-mounted accelerometers were attached to bony prominences, and signals obtained as per our previous publication (35). Each subject performed 5 stationary tasks, lifting the bare foot and placing it down with the heel striking the force platform with a force equivalent (or slightly higher) than used for during natural walking. Our custom BSA software (BSA software 2009–15; University of Cincinnati, Cincinnati, OH) was used to collect data and calculate shock absorption response variables. An average of 5 heel strikes was used for statistical analysis. Reproducibility of BSA test trials was demonstrated in our previous study (46). CBAM damping (ζ) and resonance/dominant frequencies at anatomical sites were calculated as described previously (3,35,46).

Each subject underwent 2 tests: (1) static postural balance standing on a force platform system (this quantifies movement of center of pressure [CP] as a measure of static postural balance) (47) and (2) dynamic FPS test with instrumented timed up and go (iTUG) protocol as described herein:

Static postural balance test: Each subject performed tests addressing vision, proprioception, and the vestibular system in maintaining upright balance using our published protocol (47). Tests were performed with subjects standing on a force platform (1) with eyes open (EO) and eyes closed (EC) standing directly on the platform and (2) with eyes open (FO) and eyes closed (FC) on a 4 inch thick foam pad placed on the platform. Outcome variables of postural balance tests are sway area (SA; cm²), sway length (SL; cm), excursion in mediolateral direction (Excur-ML; cm), and excursion in anterior-posterior direction (Excur-AP; cm), which are significant discriminators of faller status (26). The SA is encompassed by movement patterns of X-Y coordinates of CP movements during postural balance tests. SL is the total distance traveled by the subjects’ CP during postural balance tests. The Excur-ML and Excur-AP are maximum displacements of subject’s CP in the ML and AP directions, respectively, during postural balance tests.

Data are described using mean, standard deviation, or standard error. Because of the small sample size, outcomes were compared between fracture and no-fracture groups using Wilcoxon rank sum test. As a pilot study with expected directions of all outcomes’ responses, a 1-sided p < 0.10 was used for statistical significance. The ROC curve analysis was carried out for damping, static postural balance outcomes (postural sway variables), dynamic FPS (or balance [FPS]; iTUG outcomes), and BMD measures for differentiating facture from no-fracture groups. The results of ROC analysis are summarized with AUC. Spearman rank correlations (r) were obtained among static and dynamic FPS outcomes with damping outcomes, BMD outcomes, and FRAX outcomes. Similarly, Spearman rank correlations were also computed between outcomes for damping BMD and FRAX. Because of smaller sample size for the fracture group (n = 7), correlations among outcome variables were carried out for the no-fracture group (N = 22) only. In addition, exploratory multivariable regression analysis was also done. Two regression models were developed: for model 1, the dependent variable was FRAX score (based on lowest ever BMD) and independent variables were static postural balance outcomes for FO and FC, dynamic FPS, damping, and age. We first assessed the univariate association of cofactors using ordinary linear regression; significant cofactors at 20% level of significance in the univariate regression analyses with known expected directions were included in the multivariable linear regression analysis. Backward elimination was used, starting from the full model, including all independent effects, then deleting effects 1 by 1 until a stopping condition (p = 0.10; 1 sided) was satisfied. Similarly, for model 2, individual regression models were developed for each of the dependent variables of static postural balance and dynamic FPS outcomes. The results of regression model are reported using regression coefficient, standard error of regression coefficient, 1-tailed p value along with model-adjusted coefficient of determination (R²). Statistical software used was SAS, version 9.3 (SAS Institute, Inc., Cary, NC).

Five of 6 variables of dynamic FPS showed statistically significant differences (p values ranged between 0.0488 and 0.088) between fracture and no-fracture groups (Fig. 3A–F). For example, the fracture group took significantly longer (+12%) than the no-fracture group during turning (TD). ROM responses were higher for the fracture group, an unexpected finding. Static postural balance responses to all 4 test conditions were not statistically different between the groups.

We also tested the diagnostic performance of static postural balance and dynamic FPS responses in differentiating fracture and no-fracture groups. Maximum AUC was for PTV (0.71, 95% confidence interval [CI]: 0.51–0.92, p = 0.045) followed by TD (0.70, 95% CI: 0.47–0.94, p = 0.057), left PSV (0.68, 95% CI: 0.41–0.95, p = 0.077), and right PSV (0.68, 95% CI: 0.42–0.93, p = 0.085). In the static postural balance responses, the AUC was statistically significant for Excur-ML (0.66, 95% CI: 0.41–0.91, p = 0.10) in FC test and SL (0.71, 95% CI: 0.53–0.89, p = 0.05) in EC test.

Regarding BSA outcomes, the left heel strike-associated mean damping (ζ) values for the fracture group were 55%, 42%, and 71% lower than for the no-fracture group for sites above knee, below knee, and upper back, respectively (Fig. 4). The right heel strike mean damping values for the fracture group were 50%, 38%, and 61% lower than values for no-fracture group for sites above knee, below knee, and upper back, respectively (Fig. 5). The AUC for damping upper back (ζ_UPPERBACK) was 0.72 (95% CI: 0.52–0.92, p = 0.043), followed by above knee (ζ_ABOVEKNEE [0.65, 95% CI: 0.42–0.87]) and below knee (ζ_BELOWKNEE [0.62, 95% CI: 0.39–0.85]). Although the mean damping values for the fracture group were lower than the no-fracture group for all sites, the AUC of ζ_UPPERBACK was the only site reaching significance, possibly because of small sample size. AUCs of BMD measures were between 0.50 and 0.58. None of the BMD measures had significant AUC for differentiating fracture from no-fracture groups.

Three of 4 postural balance outcomes (SA, SL, and Excur-ML) for the FC test showed significant negative association with damping of the upper back or torso. Damping upper back (ζ_UPPERBACK) was correlated with SA (r = −0.046, p = 0.018), SL (r = −0.39, p = 0.039), and Excur-ML (r = −0.52, p = 0.007). On the other hand, although the Excur-AP was in the expected direction (i.e., negatively associated with damping), the differences between groups was not significant (Excur-AP [r = −0.27, p = 0.12]). Two of 4 postural sway outcomes (SL and Excur-ML) for the EC test showed significant negative association with ζ_UPPERBACK (r: −0.30 to −0.31; p: 0.085–0.09).

There were no consistent association among dynamic FPS outcomes and damping variables.

Higher BMD_TH was significantly associated with decrease in postural sway outcomes in all conditions, suggesting better postural balance. For 3 of 4 test conditions, EO, EC, and FO, all 4 postural sway outcomes (SA, SL, Excur-ML, and Excur-AP) were significantly negatively correlated with BMD_TH. The correlation coefficient ranged from −0.28 to −0.48. For the FC test condition, only 1 of 4 sway outcomes (Excur-AP, r = −0.31, p = 0.08) was significantly negatively associated with BMD_TH. BMD_FN showed significantly negative correlations with 3 of 4 postural sway outcomes (SA, SL, and Excur-ML) for the EC test condition. The correlation coefficient ranged from −0.36 to −0.52.

Higher BMD_TH (r = 0.57, p = 0.003) was significantly associated with an increase in PSVr, an increase in PTV (r = 0.32, p = 0.075), and a decrease in TD (r = −0.28; p = 0.10).

The FPS outcomes correlated negatively with FRAX scores—patients with higher FRAX scores had reduced dynamic functional balance, poorer balance, and increased potential of falling during walking. Of 6 FPS outcomes, 4 (mean PTV, mean PSV right arm, mean ROM right arm, and mean ROM left arm) showed significant negative associations with FRAX scores for major fracture risk (based on lowest BMD) (r = −0.53 to −0.40; p = 0.006–0.031) and hip fracture risk (r = −0.56 to −0.40; p = 0.0036–0.018). Similar correlations with FRAX scores without BMD were found.

Static postural sway was positively correlated with FRAX scores—patients with higher FRAX scores demonstrated higher postural sway, poorer balance, and increased potential of falling, even during static conditions. For the FO test, all sway outcomes demonstrated significant positive associations with FRAX scores (r = 0.55–0.28; p = 0.0041–0.11). For the EO test, except for SL (not significant), Excur-ML, Excur-AP, and SA were significantly associated with higher FRAX scores (r = 0.50–0.34; p = 0.009–0.062). For the FC test, only Excur-AP and SA were significantly associated with FRAX scores (r = 0.36–0.30; p = 0.048–0.085). For the EC test, only Excur-ML showed significant association with FRAX scores (r = 0.30; p = 0.089).

There were no correlations between FRAX scores and damping values, but each was an independent discriminator of fracture groups (Figs. 4 and 5; Table 1).

After covariate adjustment within the regression analysis for static postural sway outcomes, only Excur-ML sway for the FO test reached significance with FRAX score (based on lowest BMD) for major hip fracture risk (Table 2). Age was the only covariate significant with both FRAX scores. After covariate adjustment within the regression analysis for dynamic FPS outcomes, none of the variables reached significance with FRAX scores.

After covariate adjustment within the regression analysis, only ζ_ABOVEKNEE reached significance with only 2 of the dynamic FPS variables, TD and PTV, respectively (Table 3). Table 3 provides 1-tailed p values. BMD_TH was the only covariate that was significant with both TD and PTV. Age was significant in the PTV regression model only.

The regression models were developed only for FC and FO tests as they showed more consistent bivariate associations for multiple postural balance outcomes with damping. After covariate adjustment within the regression analysis, ζ_UPPERBACK-R reached significance with static postural balance variables SL and Excur-ML for FO and FC tests and with SA and Excur-AP for FC test. Table 4 provides 1-tailed p values. In addition, ζ_ABOVEKNEE-R also reached significance with Excur-ML for FO and FC tests, whereas ζ_ABOVEKNEE-L was significantly associated with Excur-AP for the FC test only. Only 1 of 4 damping outcomes, ζ_BELOWKNEE-R, showed an unexpected positive relationship with postural balance outcomes; this may be because of chance and/or small sample size. Covariate of age, as expected, showed significant positive relationships with all postural balance outcomes for all test conditions, consistent with previous studies (55–58).