Regenerative agents versus autogenous bone graft for alveolar cleft reconstruction: a systematic review and network meta-analysis
Article information
Abstract
Background
Autogenous iliac crest bone graft is the standard for alveolar cleft reconstruction, but donor-site morbidity has prompted interest in regenerative agents. This review compared and ranked these agents against autogenous bone and one another.
Methods
Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for network meta-analyses, six databases were searched from January 2000 to June 2026 for controlled studies comparing a regenerative agent with autogenous bone graft. Risk of bias was assessed using version 2 of the Cochrane risk-of-bias tool for randomized trials and the ROBINS-I tool; certainty was assessed using Grading of Recommendations Assessment, Development and Evaluation. A frequentist random-effects network meta-analysis used autogenous bone as the reference and P-scores to rank treatments for new bone formation, bone density, graft volume, and graft success.
Results
Twenty-one studies (23 reports) were included. Platelet-rich fibrin (risk ratio [RR], 1.46; 95% confidence interval [CI], 1.07–2.01) and concentrated growth factor (RR, 1.21; 95% CI, 1.02–1.43) were superior to autograft for graft success; platelet-rich fibrin ranked first (P-score, 0.87) and exceeded recombinant human bone morphogenetic protein-2 (RR, 1.44; 95% CI, 1.04–1.99). In one trial, bone marrow aspirate concentrate increased graft volume (standardized mean difference, 2.36; 95% CI, 1.51–3.21). No agent increased new bone formation or bone density. Certainty was moderate for one comparison, low for six, and very low for nine.
Conclusion
Autogenous bone remains the volumetric reference standard. The improvement in graft success with platelet-rich fibrin was supported by moderate-certainty evidence and persisted across all sensitivity analyses, including restriction to randomized evidence. The concentrated growth factor finding had low certainty, relied on one retrospective study, and did not persist when analyses were restricted to radiographic outcomes. Adequately powered randomized trials with standardized methods are needed.
INTRODUCTION
Cleft lip and palate affects about 1 in 700 live births globally, with higher prevalence in Asian populations and males [1,2]. Its multifactorial etiology includes genetic susceptibility and maternal exposures such as smoking, whose association with nonsyndromic clefts is particularly pronounced in Asian populations [3]. About three-quarters of affected patients have an alveolar cleft [4]. Reconstruction aims to restore maxillary arch continuity and stability, separate the nasal and oral cavities, and support adjacent teeth and the nasal base [4]. Secondary alveolar bone grafting, introduced by Boyne and Sands, is standard care during mixed dentition before permanent canine eruption; timing is guided by canine root development rather than age [4-6]. This timing supports spontaneous canine eruption into the graft and subsequent orthodontic space closure [5].
The anterior iliac crest is the most common donor site and the gold standard for alveolar cleft repair, supplying abundant osteogenic, osteoinductive, and osteoconductive cancellous bone [7,8]. Donor-site morbidity is its main drawback: in a longitudinal study, 87% of patients reported pain at 1 week, 18% had hematoma, and 22% had sensory nerve injury; chronic pain persisted in 60% at 1 month but subsided by 3 months [9]. Cleft series report donor-site complication rates of roughly 5%–10% [7]. Bone availability is finite, and iliac cancellous grafts can undergo 30%–55% resorption within 6 months, sometimes requiring regrafting [10]. Bergland-scale failure rates range from about 8% to 18% in contemporary cohorts [11], and the reconstructed cleft side retains significantly less bone than the non-cleft side long term [8]. Additional operative time further motivates the search for alternatives.
Regenerative agents augment or replace autogenous bone by supplying osteoinductive signals or osteogenic cells. Recombinant human bone morphogenetic protein-2 (rhBMP-2) recruits osteoprogenitor cells and promotes their differentiation into bone. Pediatric series report bone formation and clinical success comparable to iliac crest grafting without a second surgical site, but no significant advantage over autograft [12]. Platelet-rich plasma (PRP), platelet-rich fibrin (PRF), and concentrated growth factor (CGF) are autologous centrifuged blood products that release bone-healing growth factors. Added to autogenous bone to accelerate healing, they have not consistently increased bone formation or density in comparative trials [13]. Bone marrow aspirate concentrate (BMAC) and mesenchymal or adipose-derived stem cells, including the stromal vascular fraction (SVF), supply osteogenic cells that differentiate toward osteoblasts; early trials yielded bone volumes comparable to autograft [14,15]. Tissue engineering combines these signals and cells with bioactive or three-dimensional (3D) printed scaffolds [16].
Most studies compare one agent with autogenous bone, and meta-analyses have evaluated single materials or groupings: one found no significant graft-volume difference between autogenous bone and alloplastic substitutes [17]; another assessed autogenous bone with substitutes, substitutes alone, and iliac crest graft against the iliac crest reference [18]. None has simultaneously compared and ranked the full range of regenerative agents. Network meta-analysis combines direct and indirect evidence through a common comparator to compare treatments never tested head to head [19] and rank them [20,21]. This review compared regenerative agents with one another and autogenous bone, ranking them for new bone formation, bone density, graft volume, and graft success in alveolar cleft reconstruction.
METHODS
Protocol and registration
This review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement [22] and network meta-analysis extension [23]. The protocol was registered in PROSPERO (CRD420261432210).
Eligibility criteria
Eligibility was defined a priori using the population, intervention, comparison, outcomes, and study design (PICOS) framework. Eligible studies enrolled patients undergoing secondary alveolar cleft bone grafting, compared a regenerative agent alone or with autogenous bone against autogenous bone graft, and reported at least one bone-formation or graft-success outcome for two or more groups (Table 1). Four mutually exclusive outcomes were defined: new bone formation (percentage bone fill or bone formation ratio), graft volume (absolute volume in mm³ or cm³), bone density (Hounsfield units or equivalent radiographic density), and graft success (Bergland or Chelsea scale, or equivalent clinical definition). When multiple metrics were eligible, a prespecified hierarchy assigned a single metric to a single network; networks were analyzed separately. Analyses used absolute values at the follow-up closest to 6 months, or change-from-baseline values when exclusively reported. Supplementary Table 1 lists network assignments. For Yadav et al. [24], bone density was averaged across the coronal, middle, and apical thirds, as defined by the original authors.
Information sources and search
PubMed, Embase, the Cochrane Library, Scopus, ScienceDirect, and Web of Science were searched from January 2000 to June 30, 2026, for English-language human studies. The start date reflected the agents’ introduction into clinical use. Searches combined cleft population, alveolar bone grafting, and regenerative agents with AND, omitting outcomes to preserve sensitivity; reference lists of included studies and relevant reviews were hand-searched. Before deduplication, 1,228 records were retrieved: PubMed, 244; Embase, 188; Cochrane Library, 68; Scopus, 272; ScienceDirect, 208; and Web of Science, 248.
Study selection
Two reviewers independently screened titles, abstracts, and then full texts against eligibility criteria, resolving disagreements through discussion or a third reviewer. The PRISMA flow diagram summarizes selection (Fig. 1).
Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020 flow diagram of study identification, screening, and inclusion. Six databases yielded 1,228 records, of which 21 studies (23 reports) met the eligibility criteria. Sixteen studies were eligible for quantitative synthesis, and five were reserved for prespecified sensitivity analyses.
Data extraction
Two reviewers independently extracted study and patient characteristics, interventions, comparators, outcomes, and followup using a piloted standardized form. Discrepancies were checked against source articles and resolved by consensus.
Risk of bias assessment
Two reviewers independently assessed bias using version 2 of the Cochrane risk-of-bias tool (RoB 2) [25] for randomized trials and the Risk Of Bias In Non-randomized Studies of Interventions (ROBINS-I) tool [26] for nonrandomized comparative studies, resolving disagreements through discussion.
Data synthesis and network meta-analysis
Estimates were reported with 95% confidence intervals (CIs). Continuous outcomes (percentage bone fill, graft volume, and bone density) were expressed as mean differences or standardized mean differences (SMDs) when scales differed; dichotomous outcomes (graft success and complications) were expressed as odds ratios or risk ratios (RRs). Outcomes without a common metric were summarized narratively.
Pairwise random-effects meta-analysis using restricted maximum likelihood was performed for each comparison with at least two studies, with heterogeneity quantified by I² and the between-study variance τ². A frequentist random-effects network meta-analysis combined direct and indirect evidence using autogenous bone graft as the common reference. Agents were ranked by P-score, the frequentist analogue of the surface under the cumulative ranking curve, using the NetMetaEasy module of MetaAnalysisOnline.com [27,28]. For three-arm trials contributing a shared control group to two active-treatment comparisons in one network, the standard method to avoid double-counting was applied: the shared group was divided proportionally according to the original per-arm allocation before pooling. In the study by Yadav et al. [24], the autograft arm (n=25) was split between the PRF comparison (n=13) and the PRP comparison (n=12), matching the active arms’ respective original sample sizes (n=13 and n=12). Transitivity was assessed by tabulating potential effect modifiers across studies in each network: age, dentition stage, cleft type, donor site, graft form, imaging modality, follow-up interval, and unit of analysis (Supplementary Table 2). Because most contrasts relied on only one or two trials, transitivity was assessed clinically and could not be tested statistically. Where a closed loop was present, consistency was assessed by node-splitting [29] and the design-by-treatment interaction model, although the largely star-shaped networks limited formal inconsistency testing.
Primary networks included only studies eligible for quantitative synthesis. Overlapping Children’s Hospital Los Angeles (CHLA) reports [30-32] were treated as one cohort to avoid double-counting. Five studies outside the agent-versus-autograft comparison [33-37] were held out. Four sensitivity analyses were prespecified: (1) randomized controlled trials only, excluding studies at serious ROBINS-I risk; (2) reintroduction of studies [33,34]; (3) reintroduction of studies [35-37] as a substitute-plus-concentrate subgroup; and (4) replacement of the overlapping cluster with its largest report. Comparison-adjusted funnel plots explored small-study and publication bias when sufficient studies were available.
Certainty was assessed using the Grading of Recommendations Assessment, Development and Evaluation approach for network meta-analysis and six Confidence in Network Meta-Analysis domains: within-study bias, reporting bias, indirectness, imprecision, heterogeneity, and incoherence (Table 2, Supplementary Table 3). Incoherence was assessable only for the rhBMP-2 contrasts in the new-bone-formation network, containing the only closed loop (node-split p=0.69); unassessable domains did not trigger downgrading. Reporting bias was not formally tested because each comparison included fewer than 10 studies. Three post hoc sensitivity analyses requested during peer review were exploratory: restriction of graft success to radiographic outcomes, exclusion of Chen et al. [38] from graft volume, and exclusion of Eid et al. [39] from bone density.
RESULTS
Study selection
The six-database search identified 1,228 records, and hand-searching reference lists or trial registers yielded no additional records. After removal of 696 duplicates, 532 records underwent title and abstract screening, and 481 were excluded as clearly irrelevant. Fifty-one reports were sought; two could not be retrieved, leaving 49 for full-text eligibility assessment. At this stage, 26 reports were excluded: 11 were noncomparative, nine had an ineligible intervention or outcome, and six described overlapping cohorts or conference abstracts. The remaining 23 reports represented 21 included studies because three reports from the CHLA series [30-32] were treated as one study to avoid double-counting patients. Of these studies, 16 were eligible for quantitative synthesis (Table 3), and five were reserved for prespecified sensitivity analyses (Table 4). Three of the 16 could not connect to any outcome network: Kahan et al. [40] defined success as the clinical need for regrafting rather than using the Bergland or Chelsea scale and reported volumetric outcomes as medians with interquartile ranges; Desai et al. [41] included BMAC in both arms and had no autograftalone comparator; and Kurimori et al. [42] reported SVF outcomes that could not be expressed on a common scale with other trials. Fig. 1 summarizes selection in the PRISMA 2020 flow diagram.
Study characteristics
The studies were published from 2011 to 2026 and conducted in several countries, most commonly Egypt, India, the United States, and China. Designs included randomized controlled trials and nonrandomized comparative cohorts, most of the latter retrospective. Participants were predominantly children and adolescents with mixed dentition and mean ages of approximately 8–13 years; one trial enrolled adults, and both unilateral and bilateral alveolar clefts were represented. Agents included PRF, PRP, CGF, rhBMP-2, BMAC, SVF, and tissue-engineered or stem-cell constructs, each compared with autogenous bone graft, usually from the anterior iliac crest. Follow-up ranged from 3 months to several years, with bone outcomes assessed by cone-beam computed tomography (CBCT), conventional computed tomography, or two-dimensional (2D) radiography. Tables 3 and 5 present full study and outcome details.
Studies reserved for sensitivity analyses
Five broadly eligible studies were reserved a priori for sensitivity analyses because their comparisons differed from agent versus autograft. Table 4 summarizes their designs, interventions, and reasons for exclusion from primary networks.
Risk of bias
Risk of bias was assessed in all 21 included studies, using RoB 2 for the 14 randomized trials and ROBINS-I for the seven nonrandomized comparative studies. Six randomized trials were rated at high risk, and seven raised some concerns, chiefly because random-sequence generation and allocation concealment were incompletely reported and blinding of surgeons and, often, outcome assessors was not feasible. Shabaan et al. [45] was the only randomized trial rated at low overall risk. Dayashankara Rao et al. [46] was rated at high risk because allocation alternated between odd and even sequences, a predictable, nonrandom method without concealment. In Thanasut et al. [50], 35% of enrolled participants were lost before analysis (20 recruited, 13 analyzed), creating an additional missing-data concern. Kurimori et al. [42] had marked baseline differences in age and cleft size, but robust block randomization with sequentially sealed envelopes suggested chance imbalance in a small sample; nevertheless, the trial was rated at high risk because the authors stated that the same investigator and surgeon collected data without blinding. Six nonrandomized studies were rated at serious risk and one at moderate risk, mainly because of confounding by indication and selection effects. Trujillo et al. [49] excluded 85 of 110 patients, primarily for incomplete records or imaging from a different scanner, and imaging intervals differed by roughly 9 months between arms. Balaji [48] reported implausibly narrow standard deviations and used unblinded 2D planimetry. Both studies were rated at serious risk. Among the five studies reserved for sensitivity analyses, two of the four randomized trials raised some concerns [35,36], mainly because allocation concealment was not described. Aldaghir et al. [34] was rated at high risk because allocation used a coin toss, participants lost to follow-up were excluded, and the reported test statistic and standard deviations were incompatible. Bajestan et al. [33] was rated at high risk because enrollment in the cleft cohort stopped early on the basis of observed outcomes and no inferential statistics were reported. The retrospective study, Seif [37], was rated at serious risk owing to confounding by indication. Risk of bias did not determine exclusion; instead, the prespecified randomized-trials-only sensitivity analysis tested the influence of higher-risk nonrandomized studies. Domain-level judgments and their summary distributions were visualized using robvis (Fig. 2A and B for the 14 randomized trials; Fig. 2C and D for the seven nonrandomized studies) [52].
Risk-of-bias assessment generated using robvis. (A, B) RoB 2 traffic-light plot and weighted summary bar plot for the 14 randomized trials. (C, D) ROBINS-I traffic-light plot and weighted summary bar plot for the seven nonrandomized studies. The CHLA row represents the pooled Francis (2013) [32], Liang (2017) [31], and Hammoudeh (2017) [30] cohort, assessed using the index report. Green indicates low risk; yellow, some concerns (RoB 2) or moderate risk (ROBINS-I); and red, high risk (RoB 2) or serious risk (ROBINS-I). RoB 2, Cochrane risk-of-bias tool for randomized trials, version 2; ROBINS-I, Risk Of Bias In Non-randomized Studies of Interventions; CHLA, Children’s Hospital Los Angeles.
Network geometry
All four networks were predominantly star-shaped, with autogenous bone graft as the central comparator and only indirect connections between agents through this reference.
• New bone formation (primary outcome): the five-node network comprised autograft, PRF, PRP, rhBMP-2, and rh- BMP-2 with zygomatic shavings. The three-arm study [48] supplied its single closed loop; every other agent connected only through autograft.
• Bone density: within-design heterogeneity (Q and df) could not be estimated because, apart from the three-arm trial incorporated using the split-control method, each direct comparison was informed by only one study. The five-node network comprised autograft, PRF, PRP, BMAC, and the tissue-engineered or stem-cell construct, with no closed loop. The PRF and PRP arms entered through the three-arm trial [24], using the split-control method.
• Graft volume: within-design heterogeneity (Q and df) could not be estimated because each direct comparison was informed by only one study. The four-node star comprised autograft, BMAC, PRP, and the tissue-engineered construct. It had no closed loop, and each agent was linked to autograft by one trial.
• Graft success: the five-node network comprised autograft, PRF, CGF, rhBMP-2, and PRP, with no closed loop. PRP connected only through study [24].
Fig. 3 shows the four networks, with agent comparisons largely indirect through autograft.
Network geometry by outcome. Nodes represent treatments, and edges represent direct comparisons. Node size reflects the number of patients or, for graft success, the contributing study’s analysis unit (patients, cleft sides, or graft exposures; see Results); edge labels indicate the number of studies. (A) New bone formation; (B) bone density; (C) graft volume; (D) graft success. Autogenous bone graft is the common reference. PRP, platelet-rich plasma; rhBMP-2, recombinant human bone morphogenetic protein-2; PRF, platelet-rich fibrin; BMAC, bone marrow aspirate concentrate; TE, tissue-engineered construct; CGF, concentrated growth factor.
Network meta-analysis findings
Primary outcome: new bone formation
Five studies contributed to this network (153 patients; seven pairwise comparisons). Every agent had a point estimate favoring greater bone formation than autograft, but none was significant: rhBMP-2 with zygomatic shavings (SMD, 1.77; 95% CI, −0.36 to 3.90), PRF (SMD, 1.21; 95% CI, −0.44 to 2.87), PRP (SMD, 0.30; 95% CI, −1.87 to 2.48), and rhBMP-2 alone (SMD, 0.08; 95% CI, −1.48 to 1.63). Fig. 4A presents the forest plot.
Forest plots of network meta-analysis estimates for each treatment versus autogenous bone graft, ordered by P-score. (A) New bone formation (SMD); (B) bone density (SMD); (C) graft volume (SMD); (D) graft success (RR). Squares indicate point estimates, and horizontal lines indicate 95% CIs. SMD, standardized mean difference; RR, risk ratio; CI, confidence interval; rhBMP-2, recombinant human bone morphogenetic protein-2; PRF, platelet-rich fibrin; PRP, platelet-rich plasma; BMAC, bone marrow aspirate concentrate; TE, tissue-engineered construct; CGF, concentrated growth factor.
Secondary outcome: bone density
Four studies contributed to the bone-density network (145 patients; five pairwise comparisons). The three-arm trial [24], which provided no direct PRF-versus-PRP comparison, was incorporated using the Cochrane split-control method, dividing its autograft arm between the two active comparisons. PRF (SMD, 3.15; 95% CI, −3.56 to 9.86), PRP (SMD, 3.15), and BMAC (SMD, 3.03) had large point estimates favoring greater density than autograft, whereas the tissue-engineered construct did not (SMD, −0.61). All CIs were wide and crossed the null, and no comparison was significant. Fig. 4B presents the forest plot.
Secondary outcome: graft volume
Three studies contributed to the graft-volume network (86 patients; three pairwise comparisons). BMAC produced significantly greater graft volume than autograft (SMD, 2.36; 95% CI, 1.51 to 3.21), whereas PRP (SMD, 0.52; 95% CI, −0.11 to 1.15) and the tissue-engineered construct (SMD, 0.12; 95% CI, −1.12 to 1.37) did not differ significantly from autograft. Network estimates also favored BMAC over PRP (SMD, 1.84; 95% CI, 0.78 to 2.89) and the tissue-engineered construct (SMD, 2.23; 95% CI, 0.73 to 3.74). Fig. 4C presents the forest plot.
Secondary outcome: graft success
The graft-success network comprised five studies and 756 analysis units, rather than 756 individual patients. It combined three units of analysis: 191 patients analyzed individually (n=75 across three arms [24], n=86 [51], n=30 [46]); 95 cleft sides with adequate postoperative imaging from 130 treated patients [43]; and 470 graft exposures from 414 patients in the CHLA cohort [30], reflecting grafting at more than one cleft site in some patients. None of the five source reports linked site- or exposure-level identifiers to shared patients, so within-patient clustering could not be adjusted for and must be considered when interpreting network estimates. Six pairwise comparisons were possible, with Yadav’s three-arm trial [24] incorporated using the Cochrane split-control method. Graft success was expressed as an RR. Compared with autograft, PRF significantly increased the probability of graft success (RR, 1.46; 95% CI, 1.07–2.01), as did CGF (RR, 1.21; 95% CI, 1.02–1.43). PRP (RR, 1.30; 95% CI, 0.77–2.21) and rhBMP-2 (RR, 1.02; 95% CI, 0.96–1.08) did not differ significantly from autograft. The network estimate also favored PRF over rhBMP-2 (RR, 1.44; 95% CI, 1.04–1.99). Node-splitting confirmed that all four agent-versus-autograft contrasts relied solely on direct evidence, whereas all six agent-versus-agent contrasts, including PRF versus rhBMP-2, relied entirely on indirect evidence (Supplementary Table 4). Fig. 4D presents the forest plot.
Treatment ranking
Primary outcome
P-scores ranked rhBMP-2 with zygomatic shavings first (0.84), followed by PRF (0.71), PRP (0.40), rhBMP-2 alone (0.30), and autograft (0.25). All comparisons were nonsignificant, and the leading estimate came from one small study; the ranking is therefore hypothesis-generating rather than confirmatory.
Secondary outcome: graft volume
BMAC ranked first for graft volume by P-score (1.00), followed by PRP (0.55), the tissue-engineered construct (0.29), and autograft (0.16). As this ranking rests on a single small trial [45], it should be regarded as hypothesis-generating.
Secondary outcome: graft success
PRF ranked first for graft success by P-score (0.87), followed by PRP (0.66), CGF (0.62), rhBMP-2 (0.23), and autograft (0.12). Unlike the volume and density outcomes, PRF and CGF exhibited statistical significance against autograft, making this ranking more than hypothesis-generating, although the PRP estimate was very imprecise and came from one trial.
Inconsistency, heterogeneity, and sensitivity
Primary outcome
Within-design heterogeneity was high (I²=86.2%; 95% CI, 60.2–95.2; Q= 13, df= 1, p<0.001), consistent with pooling different volumetric and densitometric scales as SMDs. No inconsistency was detected (design-by-treatment Q p=0.218; nodesplit p=0.69 for the only closed loop). The comparison-adjusted funnel plot and Egger test (p=0.68) showed no asymmetry (Fig. 5A), although the small number of studies limited power. Balaji’s retrospective study [48], with its narrow standard deviations, inflated both heterogeneity and the leading rank; the prespecified randomized-trials-only analysis was therefore expected to provide the most informative sensitivity check.
Comparison-adjusted funnel plots by outcome. (A) New bone formation; (B) bone density; (C) graft volume; (D) graft success. These plots are exploratory because every comparison included fewer than 10 studies; a formal Egger test is reported only for the primary outcome (p=0.68). PRF, platelet-rich fibrin; PRP, platelet-rich plasma; rhBMP-2, recombinant human bone morphogenetic protein-2; BMAC, bone marrow aspirate concentrate; TE, tissue-engineered construct; CGF, concentrated growth factor.
Secondary outcome: bone density
Yadav et al. [24] dominated these estimates because its unusually narrow standard deviations—coefficients of variation of approximately 4%–5%, compared with 12%–37% in other trials—gave it substantial weight. Excluding Yadav in a prespecified sensitivity analysis reduced the network to three studies (70 patients) and removed the PRP node, which that study alone had informed. In the reduced network, the PRF estimate reversed toward the null (SMD, −0.23; 95% CI, −1.04 to 0.57), and only BMAC remained significant (SMD, 3.03; 95% CI, 2.07–3.99), on the basis of one trial [45]. The density ranking was therefore not robust and depended heavily on one study. A comparison-adjusted funnel plot is provided for completeness (Fig. 5B), but publication bias was not formally assessed because every comparison included fewer than 10 studies.
Secondary outcome: graft volume
Because the graft-volume network was a pure star without a closed loop, inconsistency between direct and indirect evidence could not be assessed. A comparison-adjusted funnel plot is provided for completeness (Fig. 5C); publication bias was again not formally assessed because there were fewer than 10 studies. The only significant result, BMAC versus autograft, came from one small randomized trial [45]. Wide CIs for the other agents reflect very small samples, particularly the five-patient tissue-engineering trial. These estimates therefore require cautious interpretation.
Secondary outcome: graft success
Within-design heterogeneity was negligible (I²=0%; 95% CI, 0–89.6; Q= 0.1, df= 2, p=0.937), making graft success the most stable network examined. Its pure-star structure, without a closed loop, prevented assessment of inconsistency between direct and indirect evidence. A comparison-adjusted funnel plot is provided for completeness (Fig. 5D); publication bias was again not formally assessed because there were fewer than 10 studies. Yadav et al. [24] used the stricter Bergland type I definition of success and was incorporated using the split-control method. Its exclusion in a prespecified sensitivity analysis reduced the network to four studies (681 patients) and removed the PRP node, which it alone had informed. The principal finding remained robust: PRF (RR, 1.56; 95% CI, 1.01–2.40) and CGF (RR, 1.21; 95% CI, 1.02–1.43) remained significantly superior to autograft, with negligible heterogeneity (I²= 0%, Q=0, df=1, p=0.974). The only material change was that PRF was no longer significantly superior to rhBMP-2 (RR, 1.53; 95% CI, 0.99–2.37); the PRF estimate now rested on one trial [46].
Restricting the new-bone-formation network to randomized trials in the prespecified sensitivity analysis left three trials (79 patients) and removed both rhBMP-2 nodes. Point estimates were essentially unchanged for PRF (SMD, 1.23; 95% CI, −1.46 to 3.91) and PRP (SMD, 0.30; 95% CI, −3.40 to 4.00), but CIs widened and no comparison was significant. The primary analysis’s leading treatment, rhBMP-2 with zygomatic shavings, disappeared because it relied entirely on one retrospective study at serious risk of bias; PRF then ranked first (P-score, 0.73). Removal of the review’s only closed loop also prevented further assessment of inconsistency. The graft-success network was reduced to two trials (105 patients), losing the rhBMP-2 and CGF nodes, but PRF remained significantly superior to autograft with an unchanged estimate (RR, 1.46; 95% CI, 1.07 to 2.01; I²=0%). Replacing the CHLA cluster with its largest report did not change this network because the cluster’s success data already came from the index report [30]. Neither Bajestan et al. [33] nor Aldaghir et al. [34] could be reintroduced into any network: Bajestan et al. [33] reported horizontal ridge width gain in millimeters, which was not compatible with the analyzed outcome metrics, and both arms in Aldaghir et al. [34] received autogenous bone, so the contrast concerned bone source rather than agent versus autograft. Studies [35-37] each reported different outcomes and could not be pooled as a single substitute-plus-concentrate subgroup; all three are summarized narratively.
Three additional sensitivity analyses were performed post hoc. Restricting graft success to radiographic definitions left three studies (200 patients) and removed the CGF node; PRF remained significantly superior to autograft (RR, 1.46; 95% CI, 1.07–2.01) and rhBMP-2 (RR, 1.44; 95% CI, 1.02–2.04). Excluding Chen et al. [38] from the graft-volume network left every estimate unchanged. Excluding Eid et al. [39] from the bone-density network left each contrast informed by one trial, forcing the between-study variance to zero and narrowing all intervals to within-study precision alone; the PRF estimate rose to SMD 6.61 (95% CI, 4.96–8.27). This apparent improvement in precision is a structural artifact of removing all replication, not evidence of a stronger effect (Supplementary Table 5).
Certainty of evidence
Evidence certainty was moderate for one contrast, low for six, and very low for the other nine (Table 2). PRF versus autograft for graft success was the only estimate not downgraded for imprecision, heterogeneity, or indirectness; it was downgraded once for within-study risk of bias, and its estimate was unchanged across all sensitivity analyses, including restriction to randomized evidence alone. All four new-bone-formation contrasts and three of the four bone-density contrasts had very low certainty because of wide CIs, high statistical heterogeneity, and reliance on single small studies.
DISCUSSION
PRF (RR, 1.46; 95% CI, 1.07–2.01) and CGF (RR, 1.21; 95% CI, 1.02–1.43) significantly improved graft success over autograft. PRF ranked highest (P-score, 0.87) and exceeded rhBMP-2 (RR, 1.44; 95% CI, 1.04–1.99). No agent significantly improved new bone formation or bone density. Only BMAC increased graft volume, based on one small trial (SMD, 2.36; 95% CI, 1.51–3.21; P-score, 1.00). These findings support platelet concentrates as adjuncts, with autogenous bone remaining the volumetric reference standard (Fig. 6).
Heatmap of treatment rankings (P-scores) across four outcomes. Higher P-scores, shown in green, indicate higher rank; color represents rank, not statistical significance. Asterisks identify treatments significantly superior to autogenous bone graft, the common reference. PRF, platelet-rich fibrin; NR, treatment not connected in that outcome network; CGF, concentrated growth factor; PRP, platelet-rich plasma; BMAC, bone marrow aspirate concentrate; TE, tissue-engineered; MSC, mesenchymal stem cell; rhBMP-2, recombinant human bone morphogenetic protein-2.
The graft-success findings differed in robustness. PRF remained significant with moderate certainty in every sensitivity analysis: RR 1.46 in both the full network (five studies, 756 analysis units) and the randomized-trials-only network (two trials, 105 patients), increasing to 1.56 after excluding Yadav et al. [24]. The low-certainty CGF estimate came from one retrospective study defining success as absence of sequestrum and oronasal fistula; the node disappeared with restriction to randomized or radiographic evidence. In that study, Lai et al. [51], CGF was placed on the nasal and labial surfaces, and bone resorption did not differ between groups.
PRF and CGF contain a 3D fibrin scaffold that releases bonehealing growth factors—including platelet-derived growth factor, transforming growth factor beta, and vascular endothelial growth factor—as fibrin degrades over 1–2 weeks (Fig. 7) [53,54]. These factors recruit mesenchymal stem cells and osteoblasts and promote osteogenic differentiation through Smad and Wnt signaling; vascular endothelial growth factor also couples angiogenesis to osteogenesis and supports early graft vascularization [53,55]. CGF forms a denser matrix, promotes osteogenic differentiation, and directs macrophages toward a proregenerative phenotype [55,56]. These mechanisms may favor grafting reliability over absolute bone volume [57].
Biological mechanisms of regenerative agents in secondary alveolar cleft grafting. Autogenous bone provides the osteogenic, osteoinductive, and osteoconductive reference standard. rhBMP-2 drives osteoblast differentiation through Smad signaling; platelet concentrates (PRF, CGF, and PRP) release bone-healing growth factors; and BMAC supplies osteogenic cells. These pathways converge on osteoblast activity, angiogenesis, new bone formation, and graft consolidation. rhBMP-2, recombinant human bone morphogenetic protein-2; PRF, platelet-rich fibrin; CGF, concentrated growth factor; PRP, platelet-rich plasma; BMAC, bone marrow aspirate concentrate; DBM, demineralized bone matrix; βTCP, beta-tricalcium phosphate; HA/TCP, hydroxyapatite/tricalcium phosphate; BMP, bone morphogenetic protein; FDA, Food and Drug Administration; PDGF, platelet-derived growth factor; TGF-β, transforming growth factor beta; VEGF, vascular endothelial growth factor; BM-MSC, bone marrow-derived mesenchymal stem cell.
rhBMP-2 estimates were close to those for autograft, with a low graft-success ranking (P-score, 0.23; nonsignificant RR, 1.02); PRF outperformed it (RR, 1.44; 95% CI, 1.04–1.99). Pediatric meta-analyses likewise found no clear advantage over autograft, and a pairwise synthesis found greater bone formation with autogenous bone at 6 months [12,58]. Postoperative swelling and edema are the main pediatric complications, and the Food and Drug Administration has warned of life-threatening airway edema [12,58]. Complications increase with dose, favoring low doses of about 0.5–0.7 mg per level [59,60]. Despite higher material costs, demineralized bone matrix with rhBMP-2 was more cost-effective than iliac crest grafting for alveolar clefts (US$4,836 vs. US$6,892), owing to shorter operations and same-day discharge [61]. BMAC’s volume gain rests on one small trial and cannot yet support routine recommendation.
Divergent graft-success and volumetric findings partly reflect inconsistent success definitions. A scoping review identified 20 nonradiographic, 2D, and 3D scoring scales; none captures every surgical goal [62]. Bergland and Chelsea scales cannot measure depth and may classify an apical defect with normal interdental height as successful [62]. Volumetric CBCT avoids distortion and superimposition and has high measurement reliability [63], but lacks standardized landmarks, protocols, and timing [64,65]. Graft resorption can approach half the implanted volume within 6 months [64]. Measurement variation may contribute to heterogeneity; standardized CBCT workflows with automated segmentation are emerging to reduce observer variability [64,66].
Earlier pairwise reviews separately evaluated platelet concentrates with autogenous bone [67], stem-cell or tissue-engineered constructs versus scaffold or autograft [68], and bovine xenograft versus iliac bone [69]; broader appraisals retained autogenous bone as the volumetric gold standard [70]. Network meta-analysis enables simultaneous treatment ranking despite scarce head-to-head trials, as demonstrated for bone substitutes in sinus augmentation [71].
To our knowledge, this is the first network meta-analysis to rank the full range of regenerative agents against autogenous bone for alveolar cleft reconstruction, with prespecified sensitivity analyses and two risk-of-bias tools covering all studies [72]. However, few studies informed each comparison, and samples were small. Predominantly star-shaped networks limited head-to-head evidence, leaving agent comparisons largely indirect and transitivity untestable [73,74]. With one or two trials per contrast, heterogeneity increases realized error and can produce inconsistency despite exchangeability [75]. Continuous outcomes had high heterogeneity because different volumetric and densitometric scales were combined as SMDs; one trial with implausibly narrow standard deviations dominated density estimates, and among the trials contributing to the continuous-outcome analyses, most raised some concerns or had a higher risk of bias, with two rated at high risk. Rankings must therefore be considered alongside evidence certainty [76]. PRF’s bone-density SMD shifted from −0.23 after excluding Yadav et al. [24] to 3.15 in the primary analysis, to 6.61 after excluding Eid et al. [39], without changes to underlying trial data. P-scores are relative rankings within a network, not effect sizes. Effect modifiers differed: Eid et al. [39] used mandibular symphysis rather than iliac crest; Lai et al. [51] used a corticocancellous block with titanium fixation, whereas other autograft arms used particulate bone; and analysis units included patients, cleft sites, clefts, and graft exposures without clustering adjustment (Supplementary Table 2). Outcome definitions, imaging, and follow-up varied; several estimates depended on single trials or three-arm trials analyzed using split controls.
Platelet concentrates improve grafting reliability and may shorten healing but lack the structural scaffold to maintain volume in large defects, supporting their use to augment rather than replace autogenous bone [77,78]. Adequately powered randomized trials should compare agents with one another and autograft using standardized CBCT volumetric endpoints, graft-success definitions, and platelet-concentrate protocols [64,77]. Consistent reporting of harms, donor-site morbidity, long-term outcomes, and pediatric safety is needed, particularly given the scarcity and geographic concentration of registered alveolar bone-grafting trials [78].
Seif [37] reported autograft-arm values identical to those in Shawky and Seifeldin [47], despite a different sample size and follow-up interval; the reports also shared an author and institution. Possible cohort overlap could not be excluded, so Seif [37] was retained only in the substitute-plus-concentrate sensitivity analysis and did not contribute to any primary network.
CONCLUSION
Autogenous bone graft remains the reference standard for alveolar cleft reconstruction; no regenerative agent produced significantly more new bone. PRF significantly improved graft success with moderate-certainty evidence, and this finding persisted across all sensitivity analyses, including restriction to radiographically defined outcomes and randomized evidence alone. CGF also improved graft success, but the low-certainty estimate came from one retrospective study with a clinical definition of success and did not persist when analysis was restricted to radiographic outcomes. Given the limited and largely indirect evidence, these findings remain hypothesis-generating and require confirmation in adequately powered randomized trials with standardized methods.
Notes
Conflict of interest
No potential conflict of interest relevant to this article was reported.
Funding
None.
Author contributions
Conceptualization: Muhammad Irpan Hendrawan, Andi Tajrin. Data curation: Muhammad Irpan Hendrawan, Andi Arfandi Arifuddin, Hadira Hadira. Formal analysis: Muhammad Irpan Hendrawan, Husni Mubarak. Investigation: Muhammad Irpan Hendrawan, Andi Arfandi Arifuddin, Hadira Hadira, Nurwahida Nurwahida. Methodology: Muhammad Irpan Hendrawan, Husni Mubarak, Irham Taufiqurrahman, Muhammad Irfan Rasul, Andi Tajrin. Visualization: Nurwahida Nurwahida, Nining Dwi Suti Ismawati. Supervision: Andi Tajrin. Validation: Irham Taufiqurrahman. Writing−original draft: Muhammad Irpan Hendrawan. Writing−review & editing: Husni Mubarak, Muhammad Irfan Rasul, Irham Taufiqurrahman, Nining Dwi Suti Ismawati, Andi Tajrin. All authors read and approved the final manuscript.
Supplementary materials
Supplemental data can be found at: https://doi.org/10.7181/acfs.2026.0106.
Assignment of each study to each outcome network
Distribution of potential effect modifiers across contributing studies
Certainty of evidence by Confidence in Network Meta-Analysis domain
League table and node-splitting for the graft-success network
Summary of all sensitivity analyses
Abbreviations
2D
two-dimensional
3D
three-dimensional
ABG
autogenous bone graft
AIC
anterior iliac crest
AICBG
anterior iliac crest bone graft
A-PRF
advanced platelet-rich fibrin
BD6
bone density at 6 months
BMAC
bone marrow aspirate concentrate
BM-MSCs
bone marrow-derived mesenchymal stem cells
BV6
bone volume at 6 months
CBCT
cone-beam computed tomography
CGF
concentrated growth factor
CHLA
Children’s Hospital Los Angeles
CI
confidence interval
CL/P
cleft lip and/or palate
CT
computed tomography
DBM
demineralized bone matrix
DFDB
demineralized freeze-dried bone
HA
hydroxyapatite
HU
Hounsfield unit
ICBG
iliac crest bone graft
IOPA
intraoral periapical radiography
IQR
interquartile range
NS
not significant
OPG
orthopantomography
PCBM
particulate cancellous bone and marrow
PICOS
population, intervention, comparison, outcomes, and study design
PRF
platelet-rich fibrin
PRISMA
Preferred Reporting Items for Systematic Reviews and Meta-Analyses
PRP
platelet-rich plasma
RCT
randomized controlled trial
rhBMP-2
recombinant human bone morphogenetic protein-2
RoB 2
version 2 of the Cochrane risk-of-bias tool
ROBINS-I
Risk Of Bias In Non-randomized Studies of Interventions
RR
risk ratio
SD
standard deviation
SMD
standardized mean difference
SVF
stromal vascular fraction
TCP
tricalcium phosphate
TE
tissue- engineered construct
βTCP
beta-tricalcium phosphate
