2.1

Selection of in vivo studies

For the clinical performance of posterior composite resin restorations, results from prospective studies on Class I/II were retrieved from the literature. The database used for the meta-analysis on Class I/II composite resin restorations and published in the Journal of Adhesive Dentistry in 2010 was up-dated with studies that were published between May 2011 and April 2015, using the same criteria as described in the publication, the same databases (PubMed and SCOPUS) and the same search words. Only those studies on composite materials were selected for whom also laboratory data had been published.

The inclusion criteria were as follows:

• 1

  Prospective clinical trial in Class II cavities or in Class I and Class II cavities.

• 2

  Minimal duration of 2 years.

• 3

  Minimal sample size at last recall: 20 restorations per material group.

• 4

  The study had to report on the following outcome variables: material fractures and anatomical shape.

For both variables, the data were graded into three categories:

1 = good or very good, corresponds to Ryge criterion “Alpha”.

2 = acceptable or repairable, corresponds to Ryge criteria “Beta” or “Charlie”.

3 = inacceptable needing replacement, corresponds to Ryge criterion “Delta”.

• 5

  These categories are based on the FDI criteria on the evaluation of posterior restoration . These criteria comprise 5 grades. In the present study grade “1” and “2” were summed up to “1” and “3” and “4” to “2”.

• 6

  The study had to report on the applied materials, conditioning technique of teeth as well as the operative technique, e.g., bevelling of enamel, preparation, isolation technique, type of curing, and layering technique.

Studies that evaluated flowable bulk fill materials (up to 4 mm layer thickness) that needed a capping layer with a high viscous composite resin materials were excluded. However, clinical studies with a normal flowable resin as first layer with a layer thickness of 1–1.5 mm were taken into consideration and so were studies with a first layer of a glass ionomer cement.

Quality assessments of the selected clinical studies were carried out by two independent reviewers, using the Cochrane Collaboration’s tool for assessing risk of bias in randomized trials . The assessment criteria contain six items: sequence generation, allocation concealment, blinding of the outcome assessors, incomplete outcome data, selective outcome reporting, and other possible sources of bias. However, only adequate sequence generation, allocation concealment and examiner blinding were considered as key domains.

During data selection and quality assessment, any disagreements between the reviewers were solved through discussion, and if needed, by consulting a third reviewer. The risk of bias was scored following recommendations as described in the Cochrane Handbook for Systematic Reviews of Interventions 5.1.0 ( ). At domain level, the judgment for each entry involved recording “yes”, indicating low risk of bias, “no” indicating high risk of bias, and “unclear” indicating either lack of information or uncertainty regarding the potential for bias. At the study level, the study was considered to be at “low” risk of bias if two or three key domains for each outcome were at “low” risk of bias. If two or more key domains were judged as “unclear”, the study as a whole was considered as “unclear”. If two or more key domains were judged as “high” risk of bias, the study as a whole was considered at “high” risk of bias.

2.2

Selection of in vitro studies

For laboratory studies on dental resin composites, only data from one study center were included to reduce the variability due to different operators, modification of tests, different test equipment, etc. The test center was the Department of Operative Dentistry and Periodontology at the Ludwig-Maximilian-University in Munich (Germany). All the data that were to be used for the in vitro/in vivo correlation had already been published in dental journals. In one publication, data on the flexural strength, flexural modulus, diametric tensile strength and compressive strength of 72 hybrid, nano-hybrid, micro-filled, packable, ormocer-based composite, flowable composites, compomers and flowable compomers were published . Another study that was published 4 years later investigated the flexural strength, flexural modulus, indentation modulus, Vickers hardness and creep of 34 nano-hybrid, micro-hybrid and flowable composite resin materials . However, data of several identical composite materials were published in both investigations. If data were present for the same material in both publications and clinical data were also present, the data of the second, more recent publication was used. In two other publications from the same research institute, composite resin materials were artificially aged with 5000 cycles thermocycling and subsequent storage in either ethanol, water, or artificial saliva . The same research center assessed the fracture toughness of 69 micro-hybrid, nano-filled, micro-filled, packable, ormocer-based, and flowable resin-based composites, compomers and flowable compomers, as well as conventional glass ionomer cements and resin-modified GIC .

For the details on the laboratory test methods, the authors refer the reader to the original publications. In what follows is a brief description of the test methods.

The flexural strength and flexural modulus were determined in a three-point bending test according to ISO/DIN 4049 and for fracture toughness (KI _C ) was determined according to ASTM test E 399-83 , using single-edge notched-beam specimens of each material. Eight specimens were fabricated for each material and for each test. The specimens were made by compressing the composite material between two glass plates with intermediate polyacetate sheets, separated by a steel mold, which had an internal dimension of 16 mm length × 2 mm height × 2 mm width. The material was cured for 40 s from both sides in a light-curing device (Dentacolor XS, Kulzer, Wehrheim). The specimens were ground with silicon carbide sand paper [grit size P 1200/4000 (Leco)] – after removal from the mold. All specimens were then stored in distilled water at 37 °C prior to testing after 24 h. For the fracture toughness test a notch (0.3 mm wide, 1 mm deep) was machined for each specimen prior to testing with a diamond saw (Proxxon, Föhren, Germany) using water coolant. The depth of the notch was measured with a light microscope; it varied between 0.90 mm and 1.10 mm. For the flexural strength and the fracture toughness test specimens were loaded until failure in a universal testing machine (MCE 2000ST, quick test Prüfpartner GmbH, Langenfeld, Germany) using a three-point bending test device. During testing, the specimens were immersed in distilled water at room temperature. The crosshead speed was 0.5 mm/min. The bending modulus was calculated from the slope of the linear part of the force-deflection diagram .

The compressive strength was determined on cylindrical specimens (8 mm in height and 4 mm in diameter) made in a Teflon mold by curing and storing the specimens similar as described previously (n = 8 per material). The specimens were placed with the flat end on the supporting plate of the universal testing machine and a compressive load was applied axially at a crosshead speed of 0.5 mm/min.

For artificial aging, the specimens were first thermocycled for 5000 cycles at 5 °C and 55 °C. Afterwards, the specimens were randomly divided into groups of 20 (specimens), stored in either distilled water, commercial artificial saliva or a 50:50 mixture of 96% ethanol and distilled water for four weeks at 37 °C. The storage solutions were exchanged daily, keeping the volume of liquid constant. The composition of the artificial saliva was as follows (ref 1 L; pH 6.9): Potassium chloride 1.2 g, sodium chloride 0.84 g, potassium phosphate 0.26 g, calcium-chloride-dihydrate 0.14 g.

2.3

Statistical analysis

2.3.1

Modeling and summarizing in vivo studies

In vivo outcomes include the probability to be free of clinical fracture and the probability not to undergo a loss of anatomical form (wear), where high probabilities represent good outcomes. We have applied a monotonic log–log transformation to convert these probabilities (which are bounded quantities between 0% and 100%) into unbounded quantities which can be modeled via a linear model. As applied in a previous study on the efficacy of Class II restorations , the dependence of an in vivo outcome on time t and material j was modeled as:

with a different slope (fixed effect) beta_j for each available material, to constrain these probabilities to be 100% at time t = 0 while decreasing monotonously with increasing values of t. Note that a linear decrease on the transformed scale implies a non-linear decrease on the original scale, where our model ensured that these probabilities do not fall below 0%. A random study effect and a random experiment effect were included in this model to take into account the fact that we had repeated measurements along time implying partly the same subjects, yielding a linear mixed model. Each measurement was weighted by the number of subjects available at a given time point, to give more weight to those measurements based on more subjects when estimating the coefficients in this model. The nlme package from the R statistical software (version 3.1.0) was used to fit these models. Each in vivo outcome parameter was considered twice, using either all clinical studies available or excluding those studies with “high” risk of bias.

In addition we included three potential confounding variables (treated as categorical factors) in the model:

• –

  tooth type ratio (percentage of the number of restored molars in relation to premolars): ≤25%, 25–50%, 50–75%, >75%.

• –

  Class I:Class II ratio (percentage of the number of Class I restorations of molars in relation to Class II restorations): 0%, 0–50%, >50%.

• –

  rubber dam (yes/no/missing information).

A slope beta_j in our model characterized thus the in vivo performance of a specific material j (the higher, the better). Since these slopes are not necessarily straightforward to interpret, they were standardized, i.e., they were divided by a robust estimate of their standard deviation after subtraction of their median, obtaining a standard deviation score (SDS). A SDS value of 0 indicates average performance while a SDS value of e.g., +1 indicates one standard deviation above average (which is good), and a SDS value of e.g., −2 indicates two standard deviations below average (which is bad).

Using the SDS allows a meaningful combination of several indicators to obtain a summary index. A clinical index to summarize the in vivo performance of a given material was thus obtained by averaging its available SDS (as standardized slopes) with respect to clinical fracture and wear.

2.3.2

Summarizing in vitro studies

Similarly, a laboratory index to summarize the in vitro performance of a given material was obtained by averaging its available SDS with respect to e.g., flexural strength [after 24 h and after aging in water/saliva/ethanol] and flexural modulus, compressive strength, and fracture toughness, where SDS was calculated by dividing the raw values of the corresponding in vitro outcome by a robust estimate of their standard deviation after subtraction of their median. Here also, a positive SDS for a given material represents a performance above average and a negative SDS a performance below average.

2.3.3

Correlation in vivo versus in vitro performance

In vivo outcomes (i.e., the slopes expressed as SDS for the different in vivo outcomes and the clinical index) were correlated with the in vitro outcomes (including the laboratory index) over all available materials via Spearman correlation. The goal was to see whether and to which extent the materials with a good in vitro performance were identical with the materials with a good in vivo performance.

Correlations for which the p-value was smaller than 0.05 were considered as statistically significant. For the main in vivo outcomes (clinical fracture, wear and the clinical index) and the main in vitro outcomes (fracture toughness, flexural strength and the laboratory index), we explored whether a linear model to predict the in vivo performance in function of the in vitro performance explained more or less accurately the variance of the in vivo response than a model with a jump at some optimal (cut-off) value of the in vitro predictor. For each combination of in vitro predictor and in vivo response, we thus selected the optimal position for a jump (i.e., maximizing R-squared) on some pre-specified grid of values for the in vitro predictor and compared the R-squared obtained in the process with the R-squared obtained from a linear model.

2.1

Selection of in vivo studies

For the clinical performance of posterior composite resin restorations, results from prospective studies on Class I/II were retrieved from the literature. The database used for the meta-analysis on Class I/II composite resin restorations and published in the Journal of Adhesive Dentistry in 2010 was up-dated with studies that were published between May 2011 and April 2015, using the same criteria as described in the publication, the same databases (PubMed and SCOPUS) and the same search words. Only those studies on composite materials were selected for whom also laboratory data had been published.

The inclusion criteria were as follows:

• 1

  Prospective clinical trial in Class II cavities or in Class I and Class II cavities.

• 2

  Minimal duration of 2 years.

• 3

  Minimal sample size at last recall: 20 restorations per material group.

• 4

  The study had to report on the following outcome variables: material fractures and anatomical shape.

For both variables, the data were graded into three categories:

1 = good or very good, corresponds to Ryge criterion “Alpha”.

2 = acceptable or repairable, corresponds to Ryge criteria “Beta” or “Charlie”.

3 = inacceptable needing replacement, corresponds to Ryge criterion “Delta”.

• 5

  These categories are based on the FDI criteria on the evaluation of posterior restoration . These criteria comprise 5 grades. In the present study grade “1” and “2” were summed up to “1” and “3” and “4” to “2”.

• 6

  The study had to report on the applied materials, conditioning technique of teeth as well as the operative technique, e.g., bevelling of enamel, preparation, isolation technique, type of curing, and layering technique.

Studies that evaluated flowable bulk fill materials (up to 4 mm layer thickness) that needed a capping layer with a high viscous composite resin materials were excluded. However, clinical studies with a normal flowable resin as first layer with a layer thickness of 1–1.5 mm were taken into consideration and so were studies with a first layer of a glass ionomer cement.

Quality assessments of the selected clinical studies were carried out by two independent reviewers, using the Cochrane Collaboration’s tool for assessing risk of bias in randomized trials . The assessment criteria contain six items: sequence generation, allocation concealment, blinding of the outcome assessors, incomplete outcome data, selective outcome reporting, and other possible sources of bias. However, only adequate sequence generation, allocation concealment and examiner blinding were considered as key domains.

During data selection and quality assessment, any disagreements between the reviewers were solved through discussion, and if needed, by consulting a third reviewer. The risk of bias was scored following recommendations as described in the Cochrane Handbook for Systematic Reviews of Interventions 5.1.0 ( ). At domain level, the judgment for each entry involved recording “yes”, indicating low risk of bias, “no” indicating high risk of bias, and “unclear” indicating either lack of information or uncertainty regarding the potential for bias. At the study level, the study was considered to be at “low” risk of bias if two or three key domains for each outcome were at “low” risk of bias. If two or more key domains were judged as “unclear”, the study as a whole was considered as “unclear”. If two or more key domains were judged as “high” risk of bias, the study as a whole was considered at “high” risk of bias.

2.2

Selection of in vitro studies

For laboratory studies on dental resin composites, only data from one study center were included to reduce the variability due to different operators, modification of tests, different test equipment, etc. The test center was the Department of Operative Dentistry and Periodontology at the Ludwig-Maximilian-University in Munich (Germany). All the data that were to be used for the in vitro/in vivo correlation had already been published in dental journals. In one publication, data on the flexural strength, flexural modulus, diametric tensile strength and compressive strength of 72 hybrid, nano-hybrid, micro-filled, packable, ormocer-based composite, flowable composites, compomers and flowable compomers were published . Another study that was published 4 years later investigated the flexural strength, flexural modulus, indentation modulus, Vickers hardness and creep of 34 nano-hybrid, micro-hybrid and flowable composite resin materials . However, data of several identical composite materials were published in both investigations. If data were present for the same material in both publications and clinical data were also present, the data of the second, more recent publication was used. In two other publications from the same research institute, composite resin materials were artificially aged with 5000 cycles thermocycling and subsequent storage in either ethanol, water, or artificial saliva . The same research center assessed the fracture toughness of 69 micro-hybrid, nano-filled, micro-filled, packable, ormocer-based, and flowable resin-based composites, compomers and flowable compomers, as well as conventional glass ionomer cements and resin-modified GIC .

For the details on the laboratory test methods, the authors refer the reader to the original publications. In what follows is a brief description of the test methods.

The flexural strength and flexural modulus were determined in a three-point bending test according to ISO/DIN 4049 and for fracture toughness (KI _C ) was determined according to ASTM test E 399-83 , using single-edge notched-beam specimens of each material. Eight specimens were fabricated for each material and for each test. The specimens were made by compressing the composite material between two glass plates with intermediate polyacetate sheets, separated by a steel mold, which had an internal dimension of 16 mm length × 2 mm height × 2 mm width. The material was cured for 40 s from both sides in a light-curing device (Dentacolor XS, Kulzer, Wehrheim). The specimens were ground with silicon carbide sand paper [grit size P 1200/4000 (Leco)] – after removal from the mold. All specimens were then stored in distilled water at 37 °C prior to testing after 24 h. For the fracture toughness test a notch (0.3 mm wide, 1 mm deep) was machined for each specimen prior to testing with a diamond saw (Proxxon, Föhren, Germany) using water coolant. The depth of the notch was measured with a light microscope; it varied between 0.90 mm and 1.10 mm. For the flexural strength and the fracture toughness test specimens were loaded until failure in a universal testing machine (MCE 2000ST, quick test Prüfpartner GmbH, Langenfeld, Germany) using a three-point bending test device. During testing, the specimens were immersed in distilled water at room temperature. The crosshead speed was 0.5 mm/min. The bending modulus was calculated from the slope of the linear part of the force-deflection diagram .

The compressive strength was determined on cylindrical specimens (8 mm in height and 4 mm in diameter) made in a Teflon mold by curing and storing the specimens similar as described previously (n = 8 per material). The specimens were placed with the flat end on the supporting plate of the universal testing machine and a compressive load was applied axially at a crosshead speed of 0.5 mm/min.

For artificial aging, the specimens were first thermocycled for 5000 cycles at 5 °C and 55 °C. Afterwards, the specimens were randomly divided into groups of 20 (specimens), stored in either distilled water, commercial artificial saliva or a 50:50 mixture of 96% ethanol and distilled water for four weeks at 37 °C. The storage solutions were exchanged daily, keeping the volume of liquid constant. The composition of the artificial saliva was as follows (ref 1 L; pH 6.9): Potassium chloride 1.2 g, sodium chloride 0.84 g, potassium phosphate 0.26 g, calcium-chloride-dihydrate 0.14 g.

2.3

Statistical analysis

2.3.1

Modeling and summarizing in vivo studies

In vivo outcomes include the probability to be free of clinical fracture and the probability not to undergo a loss of anatomical form (wear), where high probabilities represent good outcomes. We have applied a monotonic log–log transformation to convert these probabilities (which are bounded quantities between 0% and 100%) into unbounded quantities which can be modeled via a linear model. As applied in a previous study on the efficacy of Class II restorations , the dependence of an in vivo outcome on time t and material j was modeled as:

with a different slope (fixed effect) beta_j for each available material, to constrain these probabilities to be 100% at time t = 0 while decreasing monotonously with increasing values of t. Note that a linear decrease on the transformed scale implies a non-linear decrease on the original scale, where our model ensured that these probabilities do not fall below 0%. A random study effect and a random experiment effect were included in this model to take into account the fact that we had repeated measurements along time implying partly the same subjects, yielding a linear mixed model. Each measurement was weighted by the number of subjects available at a given time point, to give more weight to those measurements based on more subjects when estimating the coefficients in this model. The nlme package from the R statistical software (version 3.1.0) was used to fit these models. Each in vivo outcome parameter was considered twice, using either all clinical studies available or excluding those studies with “high” risk of bias.

In addition we included three potential confounding variables (treated as categorical factors) in the model:

• –

  tooth type ratio (percentage of the number of restored molars in relation to premolars): ≤25%, 25–50%, 50–75%, >75%.

• –

  Class I:Class II ratio (percentage of the number of Class I restorations of molars in relation to Class II restorations): 0%, 0–50%, >50%.

• –

  rubber dam (yes/no/missing information).

A slope beta_j in our model characterized thus the in vivo performance of a specific material j (the higher, the better). Since these slopes are not necessarily straightforward to interpret, they were standardized, i.e., they were divided by a robust estimate of their standard deviation after subtraction of their median, obtaining a standard deviation score (SDS). A SDS value of 0 indicates average performance while a SDS value of e.g., +1 indicates one standard deviation above average (which is good), and a SDS value of e.g., −2 indicates two standard deviations below average (which is bad).

Using the SDS allows a meaningful combination of several indicators to obtain a summary index. A clinical index to summarize the in vivo performance of a given material was thus obtained by averaging its available SDS (as standardized slopes) with respect to clinical fracture and wear.

2.3.2

Summarizing in vitro studies

Similarly, a laboratory index to summarize the in vitro performance of a given material was obtained by averaging its available SDS with respect to e.g., flexural strength [after 24 h and after aging in water/saliva/ethanol] and flexural modulus, compressive strength, and fracture toughness, where SDS was calculated by dividing the raw values of the corresponding in vitro outcome by a robust estimate of their standard deviation after subtraction of their median. Here also, a positive SDS for a given material represents a performance above average and a negative SDS a performance below average.

2.3.3

Correlation in vivo versus in vitro performance

In vivo outcomes (i.e., the slopes expressed as SDS for the different in vivo outcomes and the clinical index) were correlated with the in vitro outcomes (including the laboratory index) over all available materials via Spearman correlation. The goal was to see whether and to which extent the materials with a good in vitro performance were identical with the materials with a good in vivo performance.

Correlations for which the p-value was smaller than 0.05 were considered as statistically significant. For the main in vivo outcomes (clinical fracture, wear and the clinical index) and the main in vitro outcomes (fracture toughness, flexural strength and the laboratory index), we explored whether a linear model to predict the in vivo performance in function of the in vitro performance explained more or less accurately the variance of the in vivo response than a model with a jump at some optimal (cut-off) value of the in vitro predictor. For each combination of in vitro predictor and in vivo response, we thus selected the optimal position for a jump (i.e., maximizing R-squared) on some pre-specified grid of values for the in vitro predictor and compared the R-squared obtained in the process with the R-squared obtained from a linear model.