Effect of whole-body vibration on BMD: a systematic review and meta-analysis

Abstract

Summary

Our systematic review and meta-analysis of randomized controlled trials (RCTs) examining whole-body vibration (WBV) effect on bone mineral density (BMD) found significant but small improvements in hip areal BMD (aBMD) in postmenopausal women and in tibia and spine volumetric BMD in children/adolescents, but not in other BMD measurements in postmenopausal women and young adults.

Introduction

Animal experiments report anabolic bone changes in response to WBV, but data in humans are limited. Our objective is to conduct a systematic review and meta-analysis of RCTs examining WBV effect on BMD.

Methods

Eligible RCTs included randomized or quasi-randomized trials, with follow-up of ≥6 months, examining WBV effects on BMD in ambulatory individuals without secondary causes of osteoporosis. The weighted mean differences between WBV and control groups in absolute pre-post change in spine and hip aBMD, and in spine and tibia trabecular volumetric BMD (vBMD) were calculated.

Results

Eight RCTs in postmenopausal women (five RCTs), young adults (one RCT), and children and adolescents (two RCTs) were included. The regimens were heterogeneous, study durations were relatively short, and available data was mostly per-protocol. In postmenopausal women, WBV was found to significantly increase hip aBMD (0.015 g cm⁻²; 95% confidence interval (CI), 0.008–0.022; n = 131) versus controls, but not spine aBMD (n = 181) or tibia trabecular vBMD (n = 29). In young adults, WBV did not increase spine or hip bone mineral content, or tibia trabecular vBMD (n = 53). In children and adolescents, WBV significantly increased spine (6.2 mg cm⁻³; 95% CI, 2.5–10.0; n = 65) and tibia (14.2 mg cm⁻³; 95% CI, 5.2–23.2; n = 17) trabecular vBMD.

Conclusions

We found significant but small improvements in BMD in postmenopausal women and children and adolescents, but not in young adults. WBV is a promising new modality, but before recommendations can be made for clinical practice, large-scale long-term studies are needed to determine optimal magnitude, frequency, and duration.

Appendix

Study bias determination

The methodological quality assessment involved identifying the presence of different types of study biases based on the information provided in the included trials (i.e., as published or as reported via email communications). The following types of study biases were identified using an invalidated checklist:

□Selection bias present (check as present if at least one “no” or “unclear” is present)
Yes	No	Unclear	True random study-arm allocation performed
Yes	No	Unclear	Concealed study-arm allocation performed
□Performance bias (check as present if at least one “no” or “unclear” is present)
Groups were matched baseline based on minor confounders
Yes	No	Unclear	Calcium intakes
Yes	No	Unclear	Age at baseline
Yes	No	Unclear	Menstrual status at baseline
Yes	No	Unclear	BMD at baseline
Yes	No	Unclear	Body mass at baseline
□Detection bias present (check as present if at least one “no” or “unclear” is present)
Yes	No	Unclear	Blinding of those assessing BMD outcomes
□Attrition bias present (check as present if at least one “no” or “unclear” is present)
Yes	No	Unclear	Only intention-to-treat analysis performed

Handling of BMD outcomes

Contacting the original authors

We contacted the original authors of all of the included trials to obtain missing information, and we obtained 100% response rate. The mean absolute change in BMD was not reported in a number of original publications [26–30]. After contacting all of the original authors, Iwamoto et al. [26] and Verschueren et al. [30] provided us with the missing mean absolute change in BMD. However, since the original data were no longer available to Rubin et al. [27], Russo et al. [28], and Torvinen et al. [29], some estimations and statistical manipulations were performed to obtain the correct BMD outcome.

Estimations

Where BMD data were reported only as the baseline and final means [28, 29], the mean absolute change in BMD was obtained by subtracting the final from the baseline mean. The standard deviation (SD) corresponding to the absolute BMD change was estimated using the Follmann's method as described in the Cochrane Handbook for Systematic Reviews of Interventions Version 5.0.1 [19]. The correlation coefficient value (r) of 0.95 was used in all Follmann's calculations. Based on the 12-month follow-up data collected in our laboratory, we obtained r values ranging from 0.94 to 0.97 for the total hip, femoral neck, and L1–L4 aBMD change in 440 postmenopausal women receiving either placebo or vitamin K treatment [38]. Prior meta-analysis utilized an even higher correlation coefficient value (r = 0.99) in their Follmann's calculations [39]. Therefore, an r of 0.95 seemed to be a conservative value to use in our calculations.

A trial of Rubin et al. [27] required special attention in the determination of the aBMD outcomes. Neither the absolute mean change in aBMD nor the final aBMD was available to us. Instead, baseline aBMD was reported for 56 participants with all follow-up data out of the 70 originally enrolled participants. Also, the mean percent change in aBMD for the ITT (n = 70) and per-protocol data was reported. The mean percent change for the per-protocol data was only reported for different adherence groups but not for the 56 participants with the baseline aBMD. The adherence group with the largest sample size (n = 33) consisted of at least 60% adherent participants. Therefore, for our primary analysis of the per-protocol data, we utilized the baseline aBMD data (n = 56) and the percent change aBMD data (n = 33) to estimate an absolute change in BMD. Alternatively, for our sensitivity analysis of the ITT data, we utilized the baseline BMD data (n = 56) and the percent change aBMD data (n = 70) to estimate an absolute change in aBMD. Finally, the SD corresponding to the mean absolute change in aBMD was imputed from another trial [25] for each study group (i.e., control and vibration) and each measurement site (i.e., L2–L4 and femoral neck). This trial [25] was used, because it is the most similar to the Rubin's et al. [27] trial in terms of its study sample size, population type, and measurement type.

Other statistical manipulations

When CIs [25] or standard errors [28] were reported only, statistical formulas were used to convert these measurements to SDs. When BMD data for more than one arm were pooled into one arm [30], the mean absolute change was obtained using the “weighted mean” formula, and the corresponding SD was obtained using the “pooled or weighted SD” formula. When raw data were provided in the original publication [31] or in our email communication with the original authors [26], appropriate descriptive statistics were used to obtain the mean absolute change and the corresponding SD.