Biomimetic Remineralization Strategies on Dentin Bond Stability- Systematic Review and Network Meta-Analysis

1. Introduction

Dentin-bonding procedures pose persistent challenges in Operative Dentistry despite the currently significant successes achieved in enamel bonding [1]. A well-documented issue in the literature is the gradual deterioration of the adhesive systems' bond strength to dentin over time, primarily due to hybrid layer degradation [2]. This compromise in dentin bonding significantly limits the lifespan of adhesive restorations [3].

The ideal dentin-bonding process involves exposing the collagen network and facilitating the penetration of chelating agents or acidic functional monomers to form the crucial hybrid layer [4]. However, a portion of the exposed collagen matrix remains unfilled with resin monomers, rendering it susceptible to hydrolytic degradation over time, thus jeopardizing the longevity of dentin bonding due to nanoleakage. The incomplete water removal within hydrophilic resin monomers also creates a weak point in resin-dentin bonds [5,6]. These phenomena have led to the exploration of an innovative approach to improve dentin adhesion: the biomimetic remineralization (BR) of collagen fibrils exposed during biomineralization [7,8].

There are two primary BR strategies: incorporating mineral-promoting agents into adhesives or restorative materials and applying pre-treating solutions before adhesive systems [9,10]. For the first strategy, researchers have developed experimental adhesive systems or restorative materials containing bioactive components like calcium phosphate or other inorganic materials that supply mineral ions to remineralize the resin-dentin interface [11,12]. The second strategy involves solutions containing non-collagenous proteins or template analogs to stimulate intra/extra-fibrillar mineralization [13,14]. These remineralizing agents facilitate the formation of nanometric apatite crystals, which replace excess water, mimicking physiological remineralization [14], thus enhancing the structural integrity of dentin and extending the longevity of the dentin-composite resin bonding interface [7,15,16]. Some studies have also suggested that these agents can inhibit the degradation of exposed collagen by attracting calcium to it [17].

Therefore, it is essential to analyze the challenges posed by dentin-bonding procedures and the potential advantages of BR procedures. This systematic review uses a comprehensive network meta-analysis (NMA) to assess and compare the bond strength of human artificial caries-affected dentin (ACAD) with and without BR evaluated in in vitro studies.

2. Materials and Methods

2.1. Search Strategy

The systematic review was registered in PROSPERO and performed according to the PRISMA statement [41]. On June 2023, PubMed, ISI Web of Science, and SCOPUS were searched to identify potentially relevant studies. In addition to electronic databases, reference lists of included studies and relevant systematic reviews were also searched. Complete search strategies are available in Appendix 1.

2.2. Outcomes

The primary outcome of this systematic review was the mean difference between the bond strength of ACAD with and without BR by different adhesive application techniques —etch-and- rinse (ER) or self-etch (SE)— and ACAD protocols —chemical or biological.

2.3. Eligibility Criteria

The following inclusion criteria were established: experimental or quasi-experimental in vitro studies investigating the influence of any BR procedure on the ACAD-adhesive interface’s bond strength; having a control group (dentin without BR) for comparison; ACAD protocols in which agents were applied immediately prior to bonding; outcomes measured by shear, micro-shear, or micro-tensile bond strength (SBS, µSBS, µTBS) tests. Exclusion criteria included studies with doped materials or modified adhesive systems.

The terms “caries-affected dentin,” “demineralized dentin,” and “artificial eroded dentin” were considered as references to ACAD. ACAD consists of human dentin tissue artificially demineralized to mimic the characteristics of dentin affected by carious changes. It is created by exposing dentin tissue to acidic or demineralizing solutions to remove mineral content, leading to softening and structural alterations like those observed in natural caries-affected dentin. [46,53,54] This demineralization process is performed in a laboratory setting to replicate the conditions and properties of carious dentin.

The BR procedures considered included any technique aimed at restoring and strengthening damaged or demineralized dentin in a way that mimicked the tooth’s natural remineralization process.[3,52]

2.4. Data Extraction and Collection

Firstly, two authors (RC and JP) independently reviewed titles and abstracts to select articles for further assessment per their consensus. Disagreements were resolved by discussion until a consensus was reached. Full texts of the selected articles were retrieved, and the same two authors further evaluated and independently extracted data from them. The reference lists of included full texts were also screened and cross-referred.

In case of missing/unclear items (e.g., missing bond strength measurements, missing standard deviation values, uncertain number of samples used) or inconsistent data within or between sources (e.g., differences in data between text and figures, bond strength measurements only in figures), authors of the respective studies were contacted via e-mail. Two follow-up e-mails were sent with a one-week interval.

Search results from online databases were imported to Endnote20 (Clarivate, Philadelphia, USA), where duplicates were removed. The Rayyan app[55] was used to keep records and assist in abstract screening, full-text review, and data extraction. Data for the systematic review and NMA were extracted using a custom-made Excel worksheet.

The following items were extracted from each source: authors; year of publication; study randomization; risk of bias; means and standard deviations; number of samples; ACAD protocol (chemical or biological); BR procedure; adhesive type used (ER, SE, or universal) and adhesive application technique; method of bond strength assessment; outcome measurement time point (24h or after artificial aging method).

The authors classified and grouped the treatments by active substance into nine groups: fluorine, calcium phosphate, peptide, silica, hydroxyapatite, flavonoids, calcium, and 2-hydroxyethyl methacrylate/ethylene glycol dimethacrylate (HEMA/EDGMA).

2.5. Risk of Bias Assessment

Two authors (RC and JP) independently assessed the risk of bias in the included in vitro studies according to the QUIN tool [56]. Disagreements were resolved by discussion until a consensus was reached. Each study was graded accordingly as having high, medium, or low risk based on the final score of the tool: low risk of bias if >70%, medium risk of bias if 50–70%, and high risk of bias if <50%.

2.6. Data Synthesis and Statistical Analysis

2.6.1. Qualitative Synthesis

Qualitative evidence synthesis was performed by descriptive analysis of the studies’ characteristics, methodologic quality, and summary results, using a narrative description and summary tables providing a clear overview of the individual study characteristics, main findings, and methodological assessments.

2.6.2. Quantitative Synthesis

Quantitative syntheses were performed by random-effects NMA of the mean difference between the intervention and control groups. NMAs were conducted using the CINEMA software, based on R software packages meta an netmeta [57,58], by adhesive application technique and ACAD protocol and included all possible pair-wise comparisons based on direct and indirect evidence. In accordance with Cochrane guidelines [59], when trials had more than two arms, we combined interventions into a single group if they belonged to the same intervention category. When more than one independent treatment-comparator pair existed in each study, we treated them as if they pertained to independent studies. Following Cochrane guidelines[59] standard deviations were imputed from other included studies in cases where they were not available in the manuscript and could not be obtained upon contact with the authors.

The rating of confidence in the results was assessed following the CINEMA approach by evaluating the domains: within-study bias, reporting bias, indirectness, imprecision, heterogeneity, and incoherence. The minimal clinically important difference was established by consensus of the authors as 7 megapascals (MPa).

In addition, since it has been reported that the adhesive application technique (ER vs. SE) [8,25,26,27] and the ACAD protocol (chemical vs. biological) [7,28] might influence BR treatment’s effect, we explored effects of these two covariates in NMA effects estimates by random-effects Bayesian meta-regressions using the OpenBUGS software (Code in Appendix 1). Within a random-effects Bayesian framework, the OpenBUGS software [60] was also used to estimate each intervention's posterior median ranks and probability to be the best.

Finally, to assess the robustness of the results obtained from NMAs, as assumptions change, we conducted the following two sensitivity analyses:

• Random selection of one treatment intervention: Instead of combining interventions belonging to the same intervention category, as in the main analysis, we randomly selected only one.
• Removal of SBS test results: Instead of including all bond strength tests, as in the main analysis, we included only results from µSBS and µTBS tests.

3. Results and Discussion

3.1. Search Results

In the electronic search, 1874 records were identified after eliminating duplicates. Only 82 were selected for full-text screening. Reasons for the exclusion of screened full texts are shown in Appendix Table 1. After critical appraisal, 23 remaining articles were included in our systematic review and 22 in the NMA. A PRISMA flow diagram of the complete process is illustrated in Figure 1.

3.2. Characteristics of Included Studies

Table 1 displays the characteristics of the included studies, interventions, and outcomes. Of the 23 studies in the systematic review, 16 were experimental [8,15,18,19,20,21,22,23,24,25,26,27,28,29,30,31] and seven were quasi-experimental [7,14,32,33,34,35,36]. One study was excluded from the NMA because it lacked reporting data, which could not be obtained upon direct contact with the authors. (Appendix Table A2)

All 22 studies in the NMA performed immediate (24h) bond strength measurements. Of those studies, 13 investigated the ER technique associated with the chemical ACAD protocol [7,8,18,19,20,22,23,27,29,31,35,37], five the ER with the biological ACAD [7,15,21,28,33], 13 the SE with the chemical ACAD [8,14,20,22,23,25,26,32,34,36,37,38,39], and only one the SE with the biological ACAD [28]; the latter was insufficient to perform an NMA. In turn, 11 studies measured bond strength after artificial aging of the specimens: four used thermocycling [14,21,25,40], and seven stored them in a fluid solution for months [15,18,29,31,32,33,36].

Overall, both immediate and aged bond strength in the ACAD benefited from BR. The artificial aging method globally diminished bond strength values, and thermocycling caused the lowest bond strength.

3.3. Meta-Regressions

3.3.1. Influence of the Adhesive Technique on NMA Effect Estimates

Meta-regression results showed that the ER technique performed better than the SE in four NMA comparisons: control vs. calcium phosphate, control vs. peptide, fluorine vs. calcium phosphate, and fluorine vs. peptide. On the contrary, the SE technique performed better in the NMA comparison peptide vs. hydroxyapatite. In all other comparisons, both techniques demonstrated similar performance (Appendix Table A5).

3.3.2. Influence of the ACAD Protocol on NMA Effect Estimates

Regarding the influence of different ACAD protocols on NMA effect estimates, the chemical ACAD protocol resulted in higher bond strength values than the biological ACAD protocol in nine NMA comparisons: control vs. fluorine, control vs. calcium phosphate, control vs. peptide, control vs. HEMA, control vs. flavonoids, control vs. calcium, control vs. hydroxyapatite, fluorine vs. calcium phosphate, and fluorine vs. peptide. In all other comparisons, both protocols performed similarly (Appendix Table A6).

3.4. Network Meta-Analysis

Plots for the three performed NMAs are shown in Table 2.

Table 3 shows NMA results from the BR intervention network.

Contribution tables display Appendix Table A7, Table A8 and Table A9

3.4.1. ER Technique with Chemical ACAD Protocol

The results of this NMA suggested that no statistically significant differences exist between any BR interventions in all the network comparisons

3.4.2. ER Technique with Biological ACAD Protocol

When the ER technique and the biological ACAD protocol were used together, eight of the 10 BR intervention network’s comparisons achieved statistically significant results: the calcium phosphate intervention compared to control (MD: -21.209, 95%CI: -25.954, -16.463), flavonoids (MD: -12.771, 95%CI: -20.538, -5.003), and fluorine (MD: -17.012, 95%CI: -22.103, -11.920); the flavonoids intervention compared to control (MD:8.438, 95%CI: 2.289, 14.587); the peptide intervention compared to control (MD:18.295, 95%CI:14.418, 22.172), flavonoids (MD: 9.857, 95%CI: 2.588, 17.126), and fluorine (MD: 14.098, 95%CI: 9.684, 18.512); and the fluorine intervention compared to control (MD:4.197, 95%CI:1.080, 7.314).

3.4.3. SE Technique with Chemical ACAD Protocol

When the SE technique and the chemical ACAD protocol were used together, only two of the 36 BR intervention network’s comparisons achieved statistically significant results: the calcium phosphate (MD: -4.455, 95%CI: -8.857, -0.053) and the flavonoids (MD: -7.520, 95%CI: -14.758, -0.281) interventions compared to hydroxyapatite.

3.5. NMA Confidence Ratings

Confidence ratings for each NMA can be found in Appendix Table A10, Table A11 and Table A12.

3.5.1. ER Technique with Chemical ACAD Protocol

In this NMA, two direct comparisons (calcium vs. control and control vs. fluorine) and one indirect comparison (hydroxyapatite vs. peptide) presented very low confidence, mainly due to major imprecision, heterogeneity, or incoherence concerns. The remaining indirect and direct comparisons presented a low or moderate confidence rating.

3.5.2. ER Technique with Biological ACAD Protocol

In this NMA, all the direct and indirect comparisons presented a moderate confidence rating.

3.5.3. SE Technique with Chemical ACAD Protocol

A low confidence rating was observed for six direct comparisons (calcium phosphate vs. peptide, calcium vs. fluorine, control vs. HEMA, control vs. SiO2, fluorine vs. HEMA, and fluorine vs. peptide) and two indirect ones (calcium vs. hydroxyapatite and HEMA vs. peptide), mostly due to major concerns in heterogeneity, incoherence, and within-study bias. The remaining comparisons presented a moderate confidence rating.

3.6. Rankings

Treatment rankings and probability to rank best are displayed in Table 4.

3.6.1. ER Technique with Chemical ACAD Protocol

Among all the treatments in the NMA, hydroxyapatite achieved the highest probability of being the best treatment (46.10%), closely followed by peptide (41.55%).

3.6.2. ER Technique with Biological ACAD Protocol

In this NMA, calcium phosphate ranked first, with an 85.24% probability of being the best BR treatment.

3.6.3. SE Technique with Chemical ACAD Protocol

Compared to the other treatments in the NMA, flavonoids achieved the highest probability of being best (46.36%), followed by HEMA (17.49%).

3.7. Sensitivity Analyses

3.7.1. ER Technique with Chemical ACAD Protocol

Both sensitivity analyses showed results like those of the main analysis.

3.7.2. ER Technique with Biological ACAD Protocol

In this NMA, the sensitivity analysis where studies measuring the outcome with SBS tests were excluded was impossible because none used this test to assess the outcome. In the sensitivity analysis where we randomly selected one treatment intervention instead of combining interventions from the same category, the flavonoids vs. peptide comparison result lost statistical significance due to the loss of precision.

3.7.3. SE Technique with Chemical ACAD Protocol

When we excluded studies using SBS tests from the NMA, the flavonoids vs. hydroxyapatite comparison ceased to show differences between the two interventions due to a loss of precision. When we randomly selected one treatment intervention instead of combining interventions from the same category, eight of the 36 NMA comparison conclusions changed from not showing differences between the interventions to favoring one of them.

3.8. Discussion

This systematic review aimed to unravel the intricate interactions among different BR procedures and their influence on bond strength in human ACAD by analyzing and comparing bond strength from various in vitro studies through NMA. NMA allows the integration of data from direct and indirect comparisons, enabling a more precise estimation of treatment effects and a deeper understanding of optimal treatment options. Ultimately, this systematic review and NMA aspires to contribute to the existing knowledge on dentin-bonding procedures and offer valuable insights into the effectiveness of BR. The findings may help clinicians make informed decisions regarding dentin-bonding strategies for improved treatment outcomes [43].

This study’s systematic review and NMA have shed light on the potential benefits of BR for bond strength in human ACAD, measured both immediately and after artificial aging. Its findings indicate that BR protocols are promising in enhancing restorative materials’ bonding performance on demineralized dentin surfaces.

ACAD’s compromised nature negatively affects bond strength, and its surface is more challenging for bonding due to the incomplete infiltration of adhesives into the exposed collagen matrix [44]. Furthermore, the low pH associated with ACAD promotes the activation and activity of proteolytic enzymes, accelerating the breakdown of non-infiltrated collagen and the hybrid layer [28,38].

Our NMA findings highlighted differences between chemical and biological ACAD protocols. Chemical protocols consistently yielded higher bond strength results than biological, agreeing with previous research [45]. This difference may derive from the thicker demineralization layer associated with chemical protocols and the excessive softness of the primary dentine resulting from microbiological approaches [45].

The NMA also revealed variations in bond strength depending on the adhesive application technique. With their additional acid-etching stage, ER techniques proved more efficient in dissolving the smear layer than SE methods, which have a less acidic composition and are more sensitive [46]. Additionally, SE relies on chemical interactions with calcium ions, often found in lower concentrations in ACAD. Consequently, ER techniques yielded significantly higher bond strength values than SE, in line with the existing literature [20,22,44,47]. Moreover, when considering the ACAD surface, ER consistently demonstrated higher bond strength than SE materials [44].

This systematic review’s 23 in vitro studies showed medium heterogeneity, reflecting variations in ACAD protocols, aging methods, and bond strength tests. Thus, random-effects models were employed throughout the NMA investigation. Artificial aging methods, such as thermocycling and months of storage, generally reduce bond strength. Thermocycling promoted the most extreme breakdown of the bond interface and caused the lowest bond strength, even with associated BR, which is consistent with other studies [48]. However, different bond strength tests were used in the included investigations, which could affect the measurement results, and aspects such as specimen preparation and geometry, loading configuration, and material characteristics were not considered [3,49,50].

BR overall increased the bond strength values, even after artificial aging methods [10,51]. Nonetheless, the limited availability of studies reporting BR associated with bond strength restricts the exploration of these relationships [52]. Incorporating these BR methods into dental treatments can potentially enhance the durability and quality of the resin-dentin interface, offering promising avenues for improving clinical outcomes in restorative dentistry. In the NMA on ER with chemical ACAD, hydroxyapatite was the most effective treatment (46.10%), closely followed by peptide (41.55%), despite the low confidence in some comparisons. In the NMA on ER with biological ACAD, calcium phosphate emerged as the top-ranking BR (85.24%), significantly surpassing the control, flavonoids, and fluoride treatments. However, the NMA on SE with chemical ACAD showed low confidence in various comparisons, with flavonoids having the highest probability (46.36%) of being more effective, followed by HEMA (17.49%). These findings highlight the nuanced effectiveness of BR, influenced by different protocols and compositions. Most investigations on BR have shown its ability to remineralize ACAD in a basic manner. However, because they were carried out in vitro, their application in clinical contexts remains unexplored [52].

This study has some limitations. Most notably, in vitro studies lack the complexity of the oral environment, including oral biofluids and microbial interactions [3,44,49,50,52]. The absence of real dental caries development processes in the ACAD models is also a limitation. Future studies should address these shortcomings for a more comprehensive understanding of the clinical applicability of BR.

Another limitation is related to the sensitivity analysis for the NMA on SE with chemical ACAD. In this network, when we randomly selected one treatment intervention instead of combining interventions from the same category, eight out of the 36 NMA comparisons changed their conclusions from not showing differences between the interventions to favoring one of them. Despite this, we are confident that combining multiple arms related to the same intervention yields more reliable estimates because it does not waste useful data and evidence, as outlined and in accordance with the Cochrane recommendations. Moreover, regardless of the strategy used to cope with multiple-arm trials, six of the eight comparisons that had their conclusions changed in the sensitivity analysis were based solely on indirect evidence, which inherently carries less confidence than scenarios where direct evidence is also available.

Despite these limitations, our findings suggest that BR can enhance bond strength in ACAD, offering potential benefits for clinical practice. Dental professionals can use this knowledge to optimize treatment approaches, improve patient outcomes, and extend the longevity of adhesive bonding materials [3,44,49,50,52].

5. Conclusions

In conclusion, through a systematic review and NMAs we showed that bond strength degraded after biological or chemical ACAD protocols. As a result, surface preparation with BR procedures prior to bonding is advised to increase the bonding of ER and SE adhesives.

6. Sensitivity Analyses

Table A13. Results of Sensitivity analysis for ER with Chemical network meta-analysis.

RANDOM	Calcium
	1.106 ( -7.711, 9.922)	CaP
	-0.123 ( -8.738, 8.491)	-1.229 ( -5.371, 2.913)	Control
	-1.291 (-10.094, 7.512)	-2.397 ( -7.595, 2.802)	-1.168 ( -5.227, 2.892)	Fluorine
	5.813 ( -4.953, 16.580)	4.707 ( -2.965, 12.380)	5.937 ( -0.522, 12.395)	7.104 ( -0.524, 14.733)	HAp
	5.028 ( -4.929, 14.984)	3.922 ( -1.860, 9.704)	5.151 ( -0.520, 10.822)	6.319 ( -0.223, 12.860)	-0.786 ( -9.381, 7.809)	Peptide
WITHOUT SB	Calcium
	1.849 (-10.377, 14.074)	CaP
	-1.065 (-12.941, 10.810)	-2.914 ( -9.377, 3.549)	Control
	-1.323 (-13.357, 10.710)	-3.172 (-10.460, 4.116)	-0.258 ( -5.532, 5.016)	Fluorine
	3.777 (-10.759, 18.313)	1.928 ( -8.656, 12.513)	4.842 ( -3.540, 13.225)	5.100 ( -4.804, 15.004)	HAp
	5.947 ( -8.203, 20.096)	4.098 ( -4.949, 13.145)	7.012 ( -1.747, 15.771)	7.270 ( -2.143, 16.683)	2.170 ( -9.954, 14.293)	Peptide

¹ Note: Data in each cell are Mean Difference (MD) with 95% confidence intervals (CI) for the network comparison of row-defining treatment versus column-defining treatment. Negative values favour the intervention in the column. Legend: Calcium Phosphate (CaP), Hydroxyapatite (HAp).

Table A13. Results of Sensitivity analysis for ER with Chemical network meta-analysis.

RANDOM	Calcium
	1.106 ( -7.711, 9.922)	CaP
	-0.123 ( -8.738, 8.491)	-1.229 ( -5.371, 2.913)	Control
	-1.291 (-10.094, 7.512)	-2.397 ( -7.595, 2.802)	-1.168 ( -5.227, 2.892)	Fluorine
	5.813 ( -4.953, 16.580)	4.707 ( -2.965, 12.380)	5.937 ( -0.522, 12.395)	7.104 ( -0.524, 14.733)	HAp
	5.028 ( -4.929, 14.984)	3.922 ( -1.860, 9.704)	5.151 ( -0.520, 10.822)	6.319 ( -0.223, 12.860)	-0.786 ( -9.381, 7.809)	Peptide
WITHOUT SB	Calcium
	1.849 (-10.377, 14.074)	CaP
	-1.065 (-12.941, 10.810)	-2.914 ( -9.377, 3.549)	Control
	-1.323 (-13.357, 10.710)	-3.172 (-10.460, 4.116)	-0.258 ( -5.532, 5.016)	Fluorine
	3.777 (-10.759, 18.313)	1.928 ( -8.656, 12.513)	4.842 ( -3.540, 13.225)	5.100 ( -4.804, 15.004)	HAp
	5.947 ( -8.203, 20.096)	4.098 ( -4.949, 13.145)	7.012 ( -1.747, 15.771)	7.270 ( -2.143, 16.683)	2.170 ( -9.954, 14.293)	Peptide

¹ Note: Data in each cell are Mean Difference (MD) with 95% confidence intervals (CI) for the network comparison of row-defining treatment versus column-defining treatment. Negative values favour the intervention in the column. Legend: Calcium Phosphate (CaP), Hydroxyapatite (HAp).

Table A14. Results of Sensitivity analysis for ER with Biological network meta-analysis.

RANDOM	Calcium phosphate
	-21.320 (-26.341, -16.299)	Control
	-11.160 (-20.061, -2.258)	10.160 (2.810, 17.510)	Flavonoids
	-17.063 (-22.451, -11.675)	4.257 (0.806, 7.708)	-5.903 (-14.023, 2.217)	Fluorine
	-2.924 ( -8.500, 2.652)	18.396 (14.256, 22.535)	8.236 ( -0.200, 16.672)	14.139 (9.408, 18.870)	Peptide

¹ Note: Data in each cell are Mean Difference (MD) with 95% confidence intervals (CI) for the network comparison of row-defining treatment versus column-defining treatment. Negative values favour the intervention in the column. In blue results reaching a different conclusion from the main analysis.

Table A14. Results of Sensitivity analysis for ER with Biological network meta-analysis.

RANDOM	Calcium phosphate
	-21.320 (-26.341, -16.299)	Control
	-11.160 (-20.061, -2.258)	10.160 (2.810, 17.510)	Flavonoids
	-17.063 (-22.451, -11.675)	4.257 (0.806, 7.708)	-5.903 (-14.023, 2.217)	Fluorine
	-2.924 ( -8.500, 2.652)	18.396 (14.256, 22.535)	8.236 ( -0.200, 16.672)	14.139 (9.408, 18.870)	Peptide

¹ Note: Data in each cell are Mean Difference (MD) with 95% confidence intervals (CI) for the network comparison of row-defining treatment versus column-defining treatment. Negative values favour the intervention in the column. In blue results reaching a different conclusion from the main analysis.

Table A15. Results of Sensitivity analysis for SE with Chemical network meta-analysis.

RANDOM	Calcium
	1.938 ( -2.927, 6.803)	CaP
	1.437 ( -3.182, 6.057)	-0.500 (-3.617, 2.616)	Control
	10.407 (1.928, 18.887)	8.470 (0.706, 16.233)	8.970 (1.859, 16.081)	FLs
	0.937 ( -3.774, 5.649)	-1.000 (-4.507, 2.507)	-0.500 (-2.804, 1.804)	-9.470 (-16.945, -1.995)	Fluorine
	0.768 ( -5.753, 7.290)	-1.169 (-6.806, 4.467)	-0.669 (-5.510, 4.172)	-9.639 (-18.241, -1.036)	-0.169 (-5.004, 4.667)	HEMA
	-3.124 ( -8.605, 2.357)	-5.061 (-9.352, -0.771)	-4.561 (-7.511, -1.611)	-13.531 (-21.230,-5.832)	-4.061 (-7.804, -0.318)	-3.892 (-9.561, 1.776)	HAp
	4.256 ( -1.421, 9.932)	2.318 (-1.565, 6.201)	2.818 (-0.994, 6.632)	-6.152 (-14.220, 1.917)	3.318 (-0.931, 7.568)	3.487 (-2.601, 9.576)	7.380 (2.559, 12.200)	Peptide
	2.277 ( -5.235, 9.789)	0.340 (-6.354, 7.033)	0.840 (-5.084, 6.764)	-8.130 (-17.385, 1.125)	1.340 (-5.016, 7.696)	.509 (-6.141, 9.159)	5.401 (-1.217, 12.019)	-1.978 (-9.024, 5.066)	Silica
WITHOUT SB	Calcium
	3.937 ( -3.398, 11.273)	CaP
	0.117 ( -6.760, 6.994)	-3.820 ( -9.111, 1.471)	Control
	4.721 ( -6.357, 15.800)	0.784 ( -9.386, 10.954)	4.604 ( -4.082, 13.290)	FLs
	0.254 ( -6.695, 7.203)	-3.684 ( -9.251, 1.884)	0.137 ( -3.243, 3.516)	-4.467 (-13.787, 4.853)	Fluorine
	-2.875 (-10.771, 5.021)	-6.812 (-13.373, -0.252)	-2.992 ( -6.871, 0.887)	-7.596 (-17.108, 1.916)	-3.129 ( -8.273, 2.016)	HAp
	-2.811 (-11.987, 6.365)	-6.748 (-13.994, 0.498)	-2.928 ( -9.722, 3.866)	-7.532 (-18.559, 3.495)	-3.065 (-10.240, 4.111)	0.064 ( -7.759, 7.888)	Peptide

¹ Note: Data in each cell are Mean Difference (MD) with 95% confidence intervals (CI) for the network comparison of row-defining treatment versus column-defining treatment. Negative values favour the intervention in the column. In blue results reaching a different conclusion from the main analysis Legend: Calcium Phosphate (CaP), Hydroxyapatite (HAp), Flavonoids (FLs).

Table A15. Results of Sensitivity analysis for SE with Chemical network meta-analysis.

RANDOM	Calcium
	1.938 ( -2.927, 6.803)	CaP
	1.437 ( -3.182, 6.057)	-0.500 (-3.617, 2.616)	Control
	10.407 (1.928, 18.887)	8.470 (0.706, 16.233)	8.970 (1.859, 16.081)	FLs
	0.937 ( -3.774, 5.649)	-1.000 (-4.507, 2.507)	-0.500 (-2.804, 1.804)	-9.470 (-16.945, -1.995)	Fluorine
	0.768 ( -5.753, 7.290)	-1.169 (-6.806, 4.467)	-0.669 (-5.510, 4.172)	-9.639 (-18.241, -1.036)	-0.169 (-5.004, 4.667)	HEMA
	-3.124 ( -8.605, 2.357)	-5.061 (-9.352, -0.771)	-4.561 (-7.511, -1.611)	-13.531 (-21.230,-5.832)	-4.061 (-7.804, -0.318)	-3.892 (-9.561, 1.776)	HAp
	4.256 ( -1.421, 9.932)	2.318 (-1.565, 6.201)	2.818 (-0.994, 6.632)	-6.152 (-14.220, 1.917)	3.318 (-0.931, 7.568)	3.487 (-2.601, 9.576)	7.380 (2.559, 12.200)	Peptide
	2.277 ( -5.235, 9.789)	0.340 (-6.354, 7.033)	0.840 (-5.084, 6.764)	-8.130 (-17.385, 1.125)	1.340 (-5.016, 7.696)	.509 (-6.141, 9.159)	5.401 (-1.217, 12.019)	-1.978 (-9.024, 5.066)	Silica
WITHOUT SB	Calcium
	3.937 ( -3.398, 11.273)	CaP
	0.117 ( -6.760, 6.994)	-3.820 ( -9.111, 1.471)	Control
	4.721 ( -6.357, 15.800)	0.784 ( -9.386, 10.954)	4.604 ( -4.082, 13.290)	FLs
	0.254 ( -6.695, 7.203)	-3.684 ( -9.251, 1.884)	0.137 ( -3.243, 3.516)	-4.467 (-13.787, 4.853)	Fluorine
	-2.875 (-10.771, 5.021)	-6.812 (-13.373, -0.252)	-2.992 ( -6.871, 0.887)	-7.596 (-17.108, 1.916)	-3.129 ( -8.273, 2.016)	HAp
	-2.811 (-11.987, 6.365)	-6.748 (-13.994, 0.498)	-2.928 ( -9.722, 3.866)	-7.532 (-18.559, 3.495)	-3.065 (-10.240, 4.111)	0.064 ( -7.759, 7.888)	Peptide

¹ Note: Data in each cell are Mean Difference (MD) with 95% confidence intervals (CI) for the network comparison of row-defining treatment versus column-defining treatment. Negative values favour the intervention in the column. In blue results reaching a different conclusion from the main analysis Legend: Calcium Phosphate (CaP), Hydroxyapatite (HAp), Flavonoids (FLs).