This systematic review and meta-analysis was registered in the international prospective register of systematic reviews (PROSPERO) (CRD42020200276) following the preferred reporting items for systematic review and meta-analysis (PRISMA) statement and the population, interventions, comparisons, outcomes, study design (PICOS) question to these investigations (Moher et al., 2015; Shamseer et al., 2015):

The focused question was “Does LMHFV enhance the implant osseointegration in animal models?” Appendix 1 shows the specific screening criteria based on the PICOS question.

Primary outcome measures: The assessed primary outcomes were bone-to-implant contact (BIC) and peri-implant bone volume relative to tissue volume (BV/TV), which were defined as follows:

Secondary outcome measures: Secondary outcomes included parameter (magnitude, frequency, duration), gene expression and adverse effect of LMHFV treatment.

An electronic search was conducted among four databases (MEDLINE/PubMed, Embase, Ovid and Cochrane Library) from inception until 15th July 2020. The search was not limited by any restrictions on language or publication. Each database search combines concepts and subject headings, as detailed in Appendix 2.

An extensive manual search was also performed through included references and other related systematic reviews. In addition, relevant articles published between July 2000 and July 2020 were also inspected. The journal list was depicted as outlined in Appendix 3. Finally, the “specific theses database” (www.theses.com) and the “Gray Literature” (opensigle.inist.fr) were additionally screened for ongoing studies.

Inclusion criteria: (1) RCTs of animals; (2) Ovariectomized (OVX) or non-ovariectomized (non-OVX) implant osseointegration model; (3) More than one sham-loading control group; (4) the effect of LMHFV on peri-implant bone healing or osseointegration process; (5) Definition of the LMHFV's parameters and application; (6) Mean bone-to-implant contact (BIC, %) and/or mean peri-implant bone volume relative to tissue volume (BV/TV, %); (7) Outcomes related to peri-implant bone morphology/efficacy/adverse effects/gene expressions; (8) Full-text literature published in English before 15, July, 2020.

Exclusion criteria: (1) Included parameters were not consistent with LMHFV definition; (2) Sample size was <3 animals; (3) Treatment period was <2 weeks; (4) Only positive control group was set; (5) Outcomes of the studies remained ambiguous or unavailable; (6) Not animal RCTs (e.g., clinical trials, cell investigations, cross-sectional studies, cohort studies); (7) Other implant osseointegration models of animals (e.g., diabetes mellitus models, bone defect models); (8) Other publication types (reviews, conference reports).

Study selection was conducted by two trained researchers (XY and YB) independently. The reviewers received a professional training prior to the formal evaluation, and the level of agreement between the reviewers of screening was calculated by kappa statistics. All stages (titles, abstract, full-text) were carried out in duplicate to exclude irrelevant papers in initial screening. Then, relevant studies of full-text were retrieved and reviewed based on specific inclusion and exclusion criteria to confirm study eligibility. Study exclusion reasons were recorded as well. We resolved any disagreement by discussion with a third author (SW). Finally, in case of incomplete data, the corresponding authors of included studies were contacted and asked for further information.

Two reviewers (YZ and YZ) extracted the information independently. A third reviewer (XY) moderated any disagreement if needed. Extracted information on study characteristics were presented in Appendix 4. Data was synthesized in a data form which was specifically designed for meta-analysis according to significant category methods, and measurement data was carried out on Microsoft Excel (version 16.38, Microsoft Corp, Redmond, US). For relevant missing data, we have attempted to contact original authors. Only if we failed to contact with authors in 3 times, we would use GetData Graph Digitizer (version 2.26, getdata-graph-digitizer.com, Germany) to obtain the data in the chart.

In this study, the risk of bias was assessed by SYRCLE's tool (Hooijmans et al., 2014; Zeng et al., 2015), while the level of evidence was assessed by GRADE approach (GradeProSetup, the Cochrane Collaboration) (Guyatt et al., 2008; Balshem et al., 2011). A calibration exercise was conducted before the assessment in order to ensure consistency across reviewers. Judgements were made independently by two reviewers (SX and YD) based on the criteria for judging the risk of bias. Disagreements were resolved by discussion and consulting a third author (YG) for arbitration and consensus. Extracted data were combined into a summary Excel (version 16.38, Microsoft Corp, Redmond, US).

A systematic narrative summary was provided with information presented in table to summarize and clarify included studies' characteristics. The data was collected manually into three subgroups (OVX-WBV, non-OVX-WBV, and non-OVX-DLV). The I² statistics was used to assess the statistical heterogeneity in each subgroup. We considered heterogeneity to be statistically significant if P < 0.1 (Du et al., 2020). Due to the statistical heterogeneity, the random effect model was chosen for meta-analysis by RevMan (version 5.3, RevMan, the Cochrane Collaboration). Continuous outcomes (BIC and BV/TV) were analyzed by using standardized mean difference (SMD) with 95% confidence intervals (CI), P < 0.05 was set as significance. Finally, the analysis of the effect on parameters was presented in bubble chart using R package ggplot2 (version 3.3.2, cran.r-project.org).

In this systematic review and meta-analysis, 2,162 studies were searched and selected from MEDLINE/PubMed database (889 studies), Embase database (503 studies), Ovid database (672 studies), and Cochrane Library database (98 studies). After duplicates removed from 2,162 studies, 1,741 studies remained for further analysis. Based on rough evaluation on abstracts, unsatisfied 1,697 studies were excluded due to their topic or content. Afterwards, 30 studies (Figure 1) were left under rigorous evaluation on remained 45 full articles. The most common reasons for exclusion were unsatisfied stimulus types, unacceptable stimulus parameters or absence of animal RCTs. The reasons for excluded articles were recorded detailly in Appendix 1. Besides, no qualified article was found via manual search. Eventually, 15 manuscripts (Akca et al., 2007; Shi et al., 2010; Ogawa et al., 2011a,b, 2014; Chen et al., 2012; Zhang et al., 2012b,c; Chatterjee et al., 2014; Liang et al., 2014; Zhou et al., 2015; Wang et al., 2016, 2018; Ruppert et al., 2018, 2019; Shibamoto et al., 2018) were recruited for qualitative analysis.

Descriptive characteristics of included studies were listed in Table 1. Fifteen selected studies were conducted between 2007 and 2019, including 13 rat studies (6 OVX rat studies and 7 non-OVX studies), 1 rabbit study, and 1 dog study. Among them, 13 studies reported tibia implantation, 1 study revealed femur implantation, while only 1 beagle dog study conducted implant surgical in mandibular bone. In terms of loading modes, 10 studies reported the effect of WBV and the others applied DLV instead. Three studies combined the effect of LMHFV and drugs, in which 2 studies explored the differences and connections due to different biomechanical types. In total, 811 animals, including 785 rats, 20 rabbits, and 6 dogs, were involved in this paper. The implant number is 1,096 in 811 animal studies, we selected 972 for this review.

As a device-driven therapy, the parameters of LMHFV bear an important guiding significance for its application (Zhang et al., 2012b; Gao et al., 2016), which were defined in this paper as follows: Vibration magnitude <1 g (WBV) or 150 μm (DLV), frequency >20 Hz. According to different animal models and vibration patterns, three groups were classified, namely OVX-WBV group, non-OVX-WBV group, and non-OVX-DLV group.

Among studies involving non-OVX-WBV cases (Ogawa et al., 2011a,b, 2014; Ruppert et al., 2018), the frequency, magnitude, and weekly loading time ranged 45–140 Hz (MD: 85.0 ± 34.3 Hz), 0.043–0.6 g (MD: 0.28 ± 0.16 g), and 25–75 min (MD: 47.5 ± 22.2 min), respectively, which were 32.5–140 Hz (MD: 59.6 ± 40.0 Hz), 0.2–0.5 g (MD: 0.35 ± 0.12 g) and 70–300 min (MD: 142.2 ± 92.7 min), respectively in OVX-WBV subgroup (Chen et al., 2012; Chatterjee et al., 2014; Liang et al., 2014; Shibamoto et al., 2018). As OVX magnified defects in bone mass, the loading duration in OVX model was significantly longer than that in non-OVX model.

Generally, the amplitude (μm) was selected to replace DLV as its magnitude was often <0.1 g (Wang et al., 2016). The relationship among frequency, amplitude, and magnitude was expressed as follows:

where g (m/s²) represented the magnitude, A (m) was the amplitude, and f (Hz) was the frequency.

Among studies involving non-OVX-DLV cases (Shi et al., 2010; Zhang et al., 2012a,b; Wang et al., 2018; Ruppert et al., 2019), the frequency, the magnitude and weekly loading time ranged 20–100 Hz (MD: 51.0 ± 41.0 Hz), 7.7–73.8 μm (MD: 19.3 ± 19.2 μm), and 50–210 min (MD: 92.0 ± 62.5 min), respectively. In fact, the results from Equation (1) concluded that the values of magnitude and frequency of DLV were significantly lower than those of WBV treatment. However, the therapeutic effect of DLV was not as good as that of WBV, and lower levels of DLV parameters indicated worse outcomes. Three studies (Shi et al., 2010; Zhang et al., 2012b; Wang et al., 2018) explored the optimal vibration parameters of DLV treatment, and the best parameters were as follows: 20 Hz frequency, 15 μm magnitude, and 210 min weekly loading time in rabbits; 40 Hz frequency, 7.8 μm magnitude and 50 min weekly loading time in rats; 40 Hz frequency, 8 μm magnitude and 50 min weekly loading time in rats. Discrepancies between different parameters may be attribute to different animal models, types of implants, and loading sites.

To clarify the relationship between parameters and effect of LMHFV, we extracted eligible data and results of LMHFV from each study and plotted them into Figure 2. Relevant parameters from a total of 24 groups were extracted from 15 studies, including 6 OVX-WBV group, 9 non-OVX-WBV group, and 9 non-OVX-DLV group. There were 5 and 7 studies demonstrated beneficial effects of WBV on osteointegration enhanced by vibration treatment in OVX animals and non-OVX animals, respectively. One showed that a relative high frequency in early period do not have a significant positive effect on OVX rats, and 2 showed that a relative low frequency with a low magnitude does not have a significant positive effect on non-OVX rats. For 9 studies reporting the effect of DLV on non-OVX animals, 5 showed positive effects on rabbits and rats, and the remaining showed no significant or no positive influence of vibration during osseointegration.

Complexities of pre-clinical researches (animal species, experiment methods etc.) might result in meta-analysis heterogeneity. Hence, subgroup analyses were conducted to eliminate possible sources of heterogeneity. Due to different experimental design, the extracted data were mostly close to the following parameters: 50 Hz, 0.3 g/8 μm, 2 w, medulla.

Table 2 listed the main findings about LMHFV/osseointegration/conclusion of the recruited eligible studies. Though lack of quantitative analyze, the effect of LMHFV on gene expression and adverse effect were discussed as follows:

The kappa test result of the two evaluators was 0.74, which was considered as qualified. Following the standards of SYRCLE's tool, a total of 10 items were evaluated (Figure 5). Specifically, one study (Ruppert et al., 2018) was identified as high risk of sequence generation because there was no clear indication of randomization. Five studies (Ogawa et al., 2011a,b, 2014; Zhang et al., 2012b,c; Wang et al., 2018) were identified as high risk of incomplete outcome data because there existed some missing sample data without an adequate explanation. However, many items were not explained in detail in the included studies, which may lead to great difficulties and deviations in the interpretation of research bias. The GRADE approach was performed to assess the level of evidence body. In detail, the evidence quality of two BIC subgroups (OVX-WBV, non-OVX-DLV) and one BV/TV subgroup (OVX-WBV) were moderate level, while the others were low (Appendix 6).

Focusing on the study results alone, the positive effect of WBV treatment was more definite than that of DLV treatment. Meta-analysis showed that both WBV (P = 0.0006) and DLV (P = 0.01) have positive effects on improving BIC in non-OVX groups. However, WBV (P = 0.01) presented a stronger effect on osteogenesis in comparison to DLV (P = 0.19), manifesting as the improved BV/TV. It is believed that the weak effect of DLV treatment may result in an insufficient loading transfer to surrounding tissues or a massive micro-interface movement, further leading to an insufficient force conduction and a sparse embedding of soft tissues (Shi et al., 2010; Zhang et al., 2012b,c).

In addition, the osteogenic effect of LMHFV was more significant on OVX animal models. Since osteoporosis affects the quality and quantity of whole-body bones, WBV is more suitable for treatment (Marín-Cascales et al., 2018), while DLV has more direct efficiency on peri implant osseointegration. Meta-analysis showed that WBV could improve BIC in both non-OVX-WBV (P = 0.03) and non-OVX-DLV (P = 0.0004) groups. Also, BV/TV was comparable between groups because WBV caused bone growth in both non-OVX (P = 0.01) and OVX (P = 0.02) models. Due to bone mass reduction and bone structure deterioration, the loss of the initial stability of implants in osteoporosis patients is likely to cause an implant failure without a proper treatment (Friberg et al., 2001). The evidence of this meta-analysis suggested that LMHFV could reverse the negative effect caused by osteoporosis partially and enhance bone healing and osseointegration in peri-implant tissues.

The GRADE approach showed that three subgroups (non-OVX-DLV in BV/TV, non-OVX-WBV in BIC and BV/TV) were not conclusive. This might be high risk of bias and considerable heterogeneity within one subgroup. Despite less influence on our result, the evidence was low quality, which needs to be with caution.

In fact, in medical field, systematic reviews are one of the powerful keys to provide high level of evidence (Albelasy et al., 2020; Pound and Ritskes-Hoitinga, 2020), that is essential and formed a basis for pre-clinical assessment for medical devices (Savoldi et al., 2018b) and drugs (Faggion, 2015). Nonetheless, animal trial design in particular RCT, is important because a proper design signifies the success of the translational outcome (Hall and Traystman, 2009; Mak et al., 2014). Thus, this research only selected RCT trials on animal studies. It should be noticed that animal study cannot replace or be a guide for any human clinical trial (Van Norman, 2020), but rather a proof for authority to approve a certain level of safety (Pound et al., 2004). Indeed, all experiments might have certain limitations and it is important to report all factors, no matter contributive or destructive, so that readers or even the “artificial intelligence” can learn the facts genuinely. For the published data, prospective systematic review on animal studies was indeed recently advocated that can produce sound and actual clinical knowledge (Pound and Ritskes-Hoitinga, 2020), under the circumstances proper designs and valid models which can be found in methotrexate (Leenaars et al., 2020) and dental implant (Manzano et al., 2014) studies that were with meta-analysis. Thus, our current study can give a frank, fair, and open opinion to all readers.

Cell experiments have shown that LMHFV regulates the proliferation and differentiation of mesenchymal stem cells by improving gene expressions, growth factor secretion, bone matrix synthesis, energy metabolism etc. (Bacabac et al., 2006; Zhou et al., 2011; Suenaga et al., 2012). This effect was still manifested in implant application by initial stability, biomechanical properties and osseointegration rate of implant, because LMHFV improved promoting bone healing, soft tissue repair and angiogenesis (Shih et al., 2004; Birmingham et al., 2015). However, the transmission mechanism of regulating factors and signaling pathways on LMHFV has not been determined. Further researches on gene expressions are required.

Besides, contact finite element analysis showed different bone responses based on direct forces (i.e., DLV) and indirect forces (i.e., WBV) (Rucci et al., 2007). The main differences between WBV and DLV were shown in Table 3. In particular, WBV had a positive effect during osseointegration process, while DLV only showed a positive effect at the early stage of osseointegration (inflammatory phase and callus formation, for 2 weeks) (Wang et al., 2018; Ruppert et al., 2019).

Moreover, the effect of LMHFV is not as good as that of anti-osteoporosis drugs in OVX animals (Chen et al., 2012). To our interest, the dosage of parathytoid hormonr (PTH) could be reduced appropriately when LMHFV applied simultaneously. This combination could produce synergistic or cumulative effects to some extent (Shibamoto et al., 2018). However, we failed to obtain a pre-conceived better effect of combining LMHFV and alendrinate (ALN), which may be attributed to the inhibited osteoclast activity (Chatterjee et al., 2014). It is suggested that biomechanical interventions can not only improve treatment efficiency, but also reduce drug adverse effects. Thus, researchers can achieve an augmented effect (Shibamoto et al., 2018).

Compared to WBV loading, DLV loading by small device embedded in mice is more appropriate in oral therapy (Wang et al., 2018), which pronouncedly highlights its convenient clinical application. Differently, WBV loading can play a positive role on systemic diseases treatment (Marín-Cascales et al., 2018), while back pain and Raynaud's syndrome may occur when the body is exposed to an extensive loading for a long time (Yung et al., 2018).

At this stage, LMHFV clinical promotion is limited as follows: (Leung et al., 2014; Wong et al., 2020): (1) Lack of large-scale multi-center clinical RCTs; (2) Unknown exact instruction for jaw bone application; (3) Long treatment period; (4) High medical cost; (5) Lack of follow-up from professional guidance.

Several limitations in our paper should be well-concerned: (1) Among included studies, only one was conducted in jaw bone, and our findings should be fully clinically validated; (2) Due to limited sample size and low evidence level of biomechanical index on osseointegration, it was excluded in our study; (3) Insufficient sample size on osseointegration estimation in other bone diseases, such as bone defect and diabetes; (4) The heterogeneity was unavoidable due to differences between implants and animal species even if we had conducted subgroup analysis; (5) The data conducted in one research team and converted via graphs might lead to potential bias; (6) The quality of evidence was medium or low in this study, the exact conclusion needed to be verified by high-quality studies in the future.