Abstract
Objectives
Bone grafts are often used to enhance bone volume/quality prior to implantation insertion. This systematic review compares the histomorphometric effectiveness of bone grafts in an evidence-based manner.
Data
Randomized clinical trials comparing histomorphometrically the % of newly-formed bone between two grafts were included. Risk of bias within and across studies was assessed with the Cochrane tool and the GRADE approach, respectively. Random-effects pairwise meta-analyses were conducted, followed by network meta-analysis, network meta-regression and sensitivity analyses.
Sources
Four electronic databases were searched from inception to June 2015 without limitations.
Study selection
A total of 12 trials (5 parallel; 7 cluster) with a total of 231 patients (302 grafted sites) were included. No statistically significant differences were found in the % of new bone from pairwise comparisons between any two bone grafts. Treatment ranking based on the evidence network indicated that autografts presented the highest percentage of new bone, followed by synthetic grafts, xenografts, and allografts. No differences according to patient age, sex, healing time, membrane used or kind of surgical graft use were identified. Our confidence on pairwise comparisons was moderate to very low due to study limitations, inconsistency, and imprecision; our confidence on graft ranking was moderate due to study limitations.
Conclusions
No significant differences were found in the percentage of new bone between any two grafts.
Clinical significance
Synthetic bone substitutes or xenologous bone grafts can be used as an alternative to autologous graft in order to overcome problems of additional surgeries or limited graft availability.
1
Introduction
1.1
Background
Resorption of the edentulous or partially edentulous alveolar ridge frequently compromises dental implant placement in a prosthetically ideal position. Therefore, augmentation of an insufficient bone volume is often indicated prior to or in conjunction with implant placement to attain predictable long-term functioning and an esthetic treatment outcome. Autogenous bone grafts (AUTs) are considered the gold standard in bone regeneration procedures . However, donor site morbidity, transmission of living viruses, unpredictable resorption, limited available quantities, and the need to include additional surgical sites are amongst the autografts-related drawbacks that have intensified the search for suitable alternatives .
Bone-substitute materials have increased in popularity as adjuncts to or replacements for AUTs in bone augmentation procedures to overcome many of their limitations . Bone-substitute materials can be categorized in three groups: (1) allogenic grafts (ALLs), from another individual within the same species; (2) xenogenic grafts (XENs), from another species; or (3) alloplastic, synthetically produced grafts (SYNs). According to contemporary trends, the ideal characteristics of a bone-substitute material include space maintenance, pre-specification of the desired anatomical form, support to the periosteum, acceleration of bone remodeling, osteoconductive guidance, carrier function for antibiotics, growth factors or gene therapy approaches or scaffolds for tissue engineering . It may be too optimistic to expect that a single grafting material fulfills all these functions and will be suitable for all indications.
A large number of systematic reviews with meta-analyses has been published in the last five years , but most were of suboptimal conduct or reporting and/or had methodological limitations , while none performed network meta-analysis to compare directly all existing bone graft alternatives.
1.2
Objective
We conducted a systematic review of parallel and cluster randomized trials (RCTs) including network meta-analysis in order to investigate the comparative effectiveness of bone grafts used in oral and maxillofacial surgery prior to implant placement in humans and to compare all grafts with the current gold standard (AUT).
2
Materials and methods
2.1
Protocol and registration
The protocol for this review was made a priori based on the PRISMA-P statement , registered in PROSPERO (CRD42015023467), and all post hoc changes were noted. This systematic review was conducted according to Cochrane Handbook and reported according to the newly-published PRISMA Extension for network meta-analyses .
2.2
Eligibility criteria and literature search
RCTs on human patients comparing any two natural or synthetic bone grafts were included. No lumping of interventions was performed during the study selection phase. Non-RCTs were excluded due to bias . Both parallel (one graft per patient) and clustered trials (>one graft per patient) were included and assessed appropriately together, by calculating for the latter clustering-adjusted estimates through random-effects regression. The pre-specified eligibiligy criteria can be found in Appendix 1.
Four electronic databases were searched systematically by one author (SNP) without any limitations from inception up to June 15th, 2015 and re-checked in October 2015 for manual additions (Appendix 1). Four additional sources (Scopus, Google Scholar, ClinicalTrials.gov, and ISRCTN registry) were manually searched for additions. Authors contacted for missing data were asked about additional missed trials. No search limitations concerning language, publication year or status were applied, except for studies on humans, where available. The reference/citation lists of the included trials and relevant systematic reviews were manually searched as well.
2.3
Study selection
Titles identified were screened by one author (SNP) with a subsequent duplicate independent checking of their abstracts/full-texts against the eligibility criteria by two authors (SNP, PNP), while conflicts were resolved by a third author (JD).
2.4
Data collection
Characteristics of included trials and numerical data were extracted in triplicate by three authors (SNP, PNP, JD) using a priori constructed and piloted extraction forms. Lumping of identified grafts was performed into four categories: AUT, ALL, SYN, and XEN. In case of combinations of grafts, the graft was categorized according to the graft with over 70% contribution (Appendix 1). Piloting of the forms was performed during the protocol stage until over 90% agreement was reached. Missing or unclear information was requested per e-mail by the trials’ authors.
2.5
Risk of bias in individual trials
The risk of bias of the included trials was assessed using Cochrane’s risk of bias tool after initial calibration by three review authors (SNP, PNP, JD) and any disagreements were discussed with a fourth author (WG). The risk of bias assessment for each trial was based on the primary outcome (% new bone) or, if this was not included in the trial, on the trial’s primary outcome. The risk of bias was incorporated in data synthesis using the framework of Salanti et al. .
2.6
Data synthesis
As the outcome of bone augmentation could be influenced by the bone graft, the technique, the patient’s individual biological response, and post-operative management, a random-effects model according to DerSimonian and Laird was deemed appropriate to encompass this variability . Both pairwise and network meta-analyses were conducted to obtain estimates for primary and secondary outcomes, and presented as Mean Differences (MDs) or Relative Risks (RRs) with 95% Confidence Intervals (CIs). Heterogeneity was conventionally assessed with tau 2 and I 2 (Appendix 2) and 95% Prediction Intervals (PrIs) were calculated to predict effects in a future clinical setting by incorporating heterogeneity. For clustered trials, the raw data were requested from the trial’s authors and clustering-adjusted estimates were calculated with univariable and multivariable regression.
The results of all direct and mixed comparisons were presented in league tables and forest plots. The latter were augmented with contours of effect magnitude based on multiples of the mean standard deviation of the included outcome (10%): 0–10%—clinically-irrelevant effect, 10–20%—moderate effect, 20–30%—large effect, and >30%—very large effect. In order to rank treatments for an outcome, the Surface Under the Cumulative RAnking (SUCRA) probabilities were used, which express as a percentage the effectiveness of every intervention relative to an imaginary intervention that is always the best without uncertainty . Thus, large SUCRA scores indicate a more effective intervention. All analyses were done with Stata version 13 (StataCorp, College Station, TX) by one author (SNP), with the commands xtgee, metan, mvmeta, network and the routines from Chaimani et al. . A two-tailed P-value of 0.05 was considered significant for hypothesis-testing.
The following pre-specified effect modifiers were checked as possible sources of inconsistency/heterogeneity at patient or study level with conventional methods (Appendix 2): (a) characteristics of patients (age, gender), (b) type of graft, (c) surgical procedure conducted, (d) use of membrane, (e) membrane type, and (f) healing time.
2.7
Risk of bias across studies
The overall quality of clinical recommendations (confidence in effect estimates) for each of the main outcomes and for the network was rated using the Grades of Recommendation, Assessment, Development, and Evaluation (GRADE) approach, based on the proposal of Salanti et al. . For this assessment, the risk of bias of each included trial was re-assessed separately at outcome level. The GRADE assessment was performed by one author (SNP) and discussed with the others (PNP, JD, WG).
2.8
Additional analyses
Signs of publication bias were planned to be assessed, if ten or more included studies contributed to an outcome, with a ‘comparison-adjusted’ funnel plot together with an accompanying statistical test ; this was not performed, as no appropriate treatment ordering by bias-related factors was possible. Small-study effects were assessed by network meta-regression according to trial size (effect variance).
Sensitivity analyses were conducted (i) by attempting to form an alternate network geometry and to compare its results, (ii) comparing the design-by-treatment model to the original analysis, and (iii) examining the basic design of the included studies and its influence.
2
Materials and methods
2.1
Protocol and registration
The protocol for this review was made a priori based on the PRISMA-P statement , registered in PROSPERO (CRD42015023467), and all post hoc changes were noted. This systematic review was conducted according to Cochrane Handbook and reported according to the newly-published PRISMA Extension for network meta-analyses .
2.2
Eligibility criteria and literature search
RCTs on human patients comparing any two natural or synthetic bone grafts were included. No lumping of interventions was performed during the study selection phase. Non-RCTs were excluded due to bias . Both parallel (one graft per patient) and clustered trials (>one graft per patient) were included and assessed appropriately together, by calculating for the latter clustering-adjusted estimates through random-effects regression. The pre-specified eligibiligy criteria can be found in Appendix 1.
Four electronic databases were searched systematically by one author (SNP) without any limitations from inception up to June 15th, 2015 and re-checked in October 2015 for manual additions (Appendix 1). Four additional sources (Scopus, Google Scholar, ClinicalTrials.gov, and ISRCTN registry) were manually searched for additions. Authors contacted for missing data were asked about additional missed trials. No search limitations concerning language, publication year or status were applied, except for studies on humans, where available. The reference/citation lists of the included trials and relevant systematic reviews were manually searched as well.
2.3
Study selection
Titles identified were screened by one author (SNP) with a subsequent duplicate independent checking of their abstracts/full-texts against the eligibility criteria by two authors (SNP, PNP), while conflicts were resolved by a third author (JD).
2.4
Data collection
Characteristics of included trials and numerical data were extracted in triplicate by three authors (SNP, PNP, JD) using a priori constructed and piloted extraction forms. Lumping of identified grafts was performed into four categories: AUT, ALL, SYN, and XEN. In case of combinations of grafts, the graft was categorized according to the graft with over 70% contribution (Appendix 1). Piloting of the forms was performed during the protocol stage until over 90% agreement was reached. Missing or unclear information was requested per e-mail by the trials’ authors.
2.5
Risk of bias in individual trials
The risk of bias of the included trials was assessed using Cochrane’s risk of bias tool after initial calibration by three review authors (SNP, PNP, JD) and any disagreements were discussed with a fourth author (WG). The risk of bias assessment for each trial was based on the primary outcome (% new bone) or, if this was not included in the trial, on the trial’s primary outcome. The risk of bias was incorporated in data synthesis using the framework of Salanti et al. .
2.6
Data synthesis
As the outcome of bone augmentation could be influenced by the bone graft, the technique, the patient’s individual biological response, and post-operative management, a random-effects model according to DerSimonian and Laird was deemed appropriate to encompass this variability . Both pairwise and network meta-analyses were conducted to obtain estimates for primary and secondary outcomes, and presented as Mean Differences (MDs) or Relative Risks (RRs) with 95% Confidence Intervals (CIs). Heterogeneity was conventionally assessed with tau 2 and I 2 (Appendix 2) and 95% Prediction Intervals (PrIs) were calculated to predict effects in a future clinical setting by incorporating heterogeneity. For clustered trials, the raw data were requested from the trial’s authors and clustering-adjusted estimates were calculated with univariable and multivariable regression.
The results of all direct and mixed comparisons were presented in league tables and forest plots. The latter were augmented with contours of effect magnitude based on multiples of the mean standard deviation of the included outcome (10%): 0–10%—clinically-irrelevant effect, 10–20%—moderate effect, 20–30%—large effect, and >30%—very large effect. In order to rank treatments for an outcome, the Surface Under the Cumulative RAnking (SUCRA) probabilities were used, which express as a percentage the effectiveness of every intervention relative to an imaginary intervention that is always the best without uncertainty . Thus, large SUCRA scores indicate a more effective intervention. All analyses were done with Stata version 13 (StataCorp, College Station, TX) by one author (SNP), with the commands xtgee, metan, mvmeta, network and the routines from Chaimani et al. . A two-tailed P-value of 0.05 was considered significant for hypothesis-testing.
The following pre-specified effect modifiers were checked as possible sources of inconsistency/heterogeneity at patient or study level with conventional methods (Appendix 2): (a) characteristics of patients (age, gender), (b) type of graft, (c) surgical procedure conducted, (d) use of membrane, (e) membrane type, and (f) healing time.
2.7
Risk of bias across studies
The overall quality of clinical recommendations (confidence in effect estimates) for each of the main outcomes and for the network was rated using the Grades of Recommendation, Assessment, Development, and Evaluation (GRADE) approach, based on the proposal of Salanti et al. . For this assessment, the risk of bias of each included trial was re-assessed separately at outcome level. The GRADE assessment was performed by one author (SNP) and discussed with the others (PNP, JD, WG).
2.8
Additional analyses
Signs of publication bias were planned to be assessed, if ten or more included studies contributed to an outcome, with a ‘comparison-adjusted’ funnel plot together with an accompanying statistical test ; this was not performed, as no appropriate treatment ordering by bias-related factors was possible. Small-study effects were assessed by network meta-regression according to trial size (effect variance).
Sensitivity analyses were conducted (i) by attempting to form an alternate network geometry and to compare its results, (ii) comparing the design-by-treatment model to the original analysis, and (iii) examining the basic design of the included studies and its influence.
3
Results
3.1
Study selection
The systematic electronic and manual search identified 283 and 7 reports, respectively ( Fig. 1 ). From these a total of 104 and 167 reports were excluded through screening and full-text assessment, respectively (Appendix 3). Finally, a total of 45 papers (pertaining to 41 unique trials) were included in the systematic review.
3.2
Study characteristics
The characteristics of the trials included in the qualitative part of this study (i.e. systematic review) can be seen in Appendices 4–6. From the 41 included trials 6 (15%) were multicenter and 37 (90%) took place in a university environment from 15 different countries. A total of 852 patients (47% male and 53% female) were included with an average of 20.8 patients per trial and an average age of 50.6 years. From the included trials, 17 (41%) were of parallel design and 24 (59%) of clustered design (i.e. more than one grafted site per patient).
As far as surgical procedures were concerned, 54% of the trials assessed sinus lift, 32% preservation of extraction sockets, 12% ridge augmentation, and the last 2% both ridge augmentation and sinus lift with a total of 1164 surgical sites being grafted. A wide variety of bone grafts were used, which were categorized as AUT (harvested mostly intraorally or from the iliac crest), ALL, SYN (based on hydroxyapatite or calcium sulphate), and XEN (mostly represented by Bio-Oss; Geistlich, Wolhusen, Switzerland). The grafted region was covered additionally by a membrane (mostly collagen ones) in 56% of the trials, by a collagen sponge or fibrin glue in 4% of the trials or without additional means in 37% of the trials.
In all trials that assessed histomorphometrically the grafted regions, bone samples were collected after a healing phase of 3–9 months during a subsequent implant insertion. Among the 25 out of 41 included trials that adequately reported the subsequent implant insertion, a total of 1261 implants were inserted (mean of 50.4 implants per study).
3.3
Risk of bias within studies
The risk of bias assessment for the 41 included trials can be seen in Appendices 7–9. High risk of bias arising from problematic generation of the random sequence was seen in 10% of the trials, bias from incomplete outcome data in 7% of the trials, and bias from selective outcome reporting in 5% of the trials. The main source of bias however, was residual bias due to other reasons, including poor trial design, inadequate data handling for cluster trials or failing to take into account confounding introduced from smoking or systemic diseases. It is however important to state, that extreme uncertainty exists on the bias assessment of random sequence generation, allocation concealment and blinding, as these were very poorly reported, making a formal assessment of their appropriateness impossible (“unclear” categories in the risk of bias summary).
3.4
Results of individual trials
The results of the 41 individual trials that were included in the systematic review are expressed as MDs or RRs in Appendix 10. In the case of available individual patient data, these are based on univariable or multivariable generalized estimating equations accounting for clustering within patients and any possible confounders reported in the original trial (patient sex, age, healing time, membrane type).
3.5
Network structure and synthesis of results
Out of the 41 trials included in the qualitative part of this study (i.e. systematic review), only 12 were included in the quantitative synthesis (i.e. meta-analyses) for various reasons ( Fig. 1 ). These pertained to 5 pairwise comparisons (AUT vs ALL, AUT vs SYN, AUT vs XEN, ALL vs SYN, and SYN vs XEN), with most of them being statistically insignificant, although high heterogeneity was found ( Table 1 ; Appendix 11).
