Criteria for considering studies for this review
Types of studies
Randomised controlled trials (RCTs), including split-mouth studies.
Types of participants
Adults (16 years or over) with one or more defective resin composite restoration(s) in a molar or premolar tooth/teeth treated by like-for-like replacement (i.e. replacement with resin composite) or like-for-like repair (i.e. repair with resin composite) or both. Participants in whom a tooth undergoes further restoration or an extraction for reasons not connected with the repair/replacement restoration (e.g. extraction due to periodontal disease) were not included.
A composite restoration may be considered defective as a result of secondary caries, chipping or fracture of the restoration, chipping or fracture of the tooth or alternatively the presence of a marginal defect/s, i.e. gap/s between the restoration and the tooth surface.
Types of interventions
Studies with the following interventions and controls were included.
In the context of this systematic review, the terms 'repair' and 'replacement' are defined as follows. A repair to a resin composite restoration is defined as the removal of only the defective part of the restoration or adjacent tooth tissue or both followed by the placement of a new partial restoration. A replacement resin composite restoration is defined as the removal of an entire restoration including any bases, liners and secondary caries and tooth tissue where appropriate, followed by the placement of a new restoration.
This systematic review was not concerned with studies involving the refurbishment of resin composite restorations. In order to ensure that there is clarity regarding the inclusion and exclusion criteria, refurbishment of a resin composite restoration is defined as reshaping or refinishing or removal of overhangs in an existing restoration which does not require the placement of additional restorative material.
In order to be included in this review, studies must have used clearly defined criteria for assessing whether restorations were defective. Studies were expected to use the same criteria at baseline and follow-up stages. Although criteria for standardising the diagnosis of defective restorations are not well defined or universally accepted, the US Public Health Service (USPHS) criteria and modified Ryge criteria provide models for this (Ryge 1981).
Types of outcome measures
The main outcome of interest was success or failure of the replacement or repair restoration and associated tooth as assessed by clinical examination. The primary outcome measures were therefore the clinical acceptability or unacceptability of each restoration, defined by the USPHS criteria, Ryge criteria or modifications of these scales, and assessed by clinical examination. It was anticipated that this would be recorded as success or failure of the restoration and/or that further repair or replacement of the restoration was necessary.
Success or failure of restoration, as defined by the USPHS criteria. If this was not reported using the USPHS criteria, information from other measures such as the Ryge criteria or other modifications of these scales and assessment by clinical examination would be used.
Further restoration (repair, restoration, placement of crown inlay; root filling) required (studies should have determined success or failure according to the same criteria used in the decision to replace or repair the restoration).
Presence of clinical symptoms (pain, swelling, diagnosis of pulpitis, abscess formation).
Extraction of tooth due to decay.
Pain or discomfort associated with the procedure (perioperative or postoperative (i.e. within 48 hours)).
Patient satisfaction as measured by patient satisfaction or aesthetic scales.
Timing of outcome assessment
Outcome data from all periods of follow-up were to be included, but where the period of follow-up differed between studies, this was to be categorised as medium term (less than five years) or long term (five years and above). Time-to-event (survival data) was to be collected and analysed where available.
Search methods for identification of studies
For the identification of studies included or considered for this review, detailed search strategies were developed for each database searched. These were based on the search strategy developed for MEDLINE via OVID (Appendix 1) but revised appropriately for each database to take account of differences in controlled vocabulary and syntax rules.
The following databases were searched.
The Cochrane Oral Health Group's Trials Register (to 24 July 2013) (Appendix 2);
The Cochrane Central Register of Controlled Trials (CENTRAL) (The Cochrane Library 2013, Issue 6) (Appendix 3);
MEDLINE (1946 to 24 July 2013) (Appendix 1);
EMBASE (1980 to 24 July 2013) (Appendix 4);
BIOSIS via Web of Knowledge (1969 to 24 July 2013) (Appendix 5);
Web of Science (1945 to 24 July 2013) (Appendix 6);
OpenGrey (to 24 July 2013) (Appendix 7).
No restrictions were placed on the language or date of publication when searching the electronic databases.
Searching other resources
Researchers, experts and organisations known to be involved in this field were contacted in order to trace unpublished or ongoing studies.
Only handsearching done as part of the Cochrane Worldwide Handsearching Programme and uploaded to CENTRAL was included (see the Cochrane Masterlist for details of journal issues searched to date).
Reference lists of all eligible trials and review articles, and in turn their reference lists, were checked for studies not already identified.
Data collection and analysis
Selection of studies
At least two review authors independently assessed the abstracts of studies resulting from the searches. Full text copies of all relevant and potentially relevant studies, those appearing to meet the inclusion criteria, or for which there were insufficient data in the title and abstract to make a clear decision, were obtained. The full text papers were assessed independently by these two review authors and any disagreement on the eligibility of potentially included studies were resolved through discussion and consensus. If consensus could not be reached by the two review authors, a third review author was consulted. After assessment by the review authors, studies that did not match the inclusion criteria were excluded and the reasons for their exclusion were noted. If any of the studies of interest had multiple publications, these were identified and grouped under the same study.
No eligible studies meeting the inclusion criteria were found in the original review (published in 2010) and in the subsequent update (in 2014). If any studies were eligible for inclusion, the standard methods as outlined in the Cochrane Handbook for Systematic Reviews of Interventions would have been undertaken for data analysis and discussion of results (Higgins 2011).
Data extraction and management
Data would have been extracted independently by the two review authors using specially designed data extraction forms. For each trial included, the following would have been recorded and presented in study tables: the date that the study was conducted, the country, year of publication and its duration; details of study design, types of intervention, treatments, controls, outcomes; sample size, number recruited, details of withdrawals by study group, age and characteristics of subjects; outcomes, assessment methods and time intervals; study setting and source of funding. If necessary, authors were to be contacted for further information or clarification of their publications or both.
Assessment of risk of bias in included studies
If any relevant studies had been identified, two review authors would have independently assessed the risk of bias using the Cochrane risk of bias assessment tool according to criteria described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). Any discrepancies of ratings between review authors were to be resolved through discussions.
The following domains would have been assessed:
blinding (of participants, personnel and outcome assessors);
incomplete outcome data;
selective outcome reporting;
The review authors would have reported these assessments for included studies in a Risk of bias table in the included studies section in Review Manager (RevMan) (RevMan 2012).
Measures of treatment effect
For dichotomous outcomes, the estimate of effect of an intervention would have been expressed as risk ratios. Survival or time-to-event data would have been analysed as hazard ratios if data were available.
For any possible continuous outcomes, mean differences and standard deviations were to be used to summarise the data for each group. If different scales were used to measure the same outcome and these were considered to be similar enough for pooling, standardised mean differences would have been used to summarise the pooled data.
All measures of treatment effect were to be reported together with their respective confidence intervals.
Unit of analysis issues
The units of randomisation and analysis in the included trials would ideally all have been at the level of the individual. When split-mouth studies were included and each individual trial participant had one tooth randomised to intervention and another randomised to control, analyses would have taken into account the paired nature of the data. In trials where the unit of randomisation was the tooth and the number of teeth included in the trial was not more than twice the number of participants, the data were to be treated as if the unit of randomisation was the individual. It was recognised that the resulting 95% confidence intervals produced would appear narrower (i.e. the estimate would seem to be more precise) than they should be, and would therefore have been interpreted accordingly.
Dealing with missing data
If studies met the inclusion criteria of the review, attempts would have been made to retrieve missing data from authors of trials. Methods for estimating missing standard deviations in section 7.7.3 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011) and techniques described by Follmann (Follmann 1992) would also be used to estimate the standard error of the difference for split-mouth studies, where the appropriate data were not presented and could not be obtained.
Proportions of participants for whom outcome data were not provided would have been recorded in the summary of study characteristics table and also the risk of bias table. If missing data were significant, the risk would have been assessed by undertaking available case and intention-to-treat analyses and comparing these using sensitivity analysis.
Assessment of heterogeneity
Apart from statistical heterogeneity, clinical heterogeneity in terms of patient population, intervention, comparison and how outcomes were measured and reported would have been considered before any decisions to pool the data were made and in the description of results.
If more than one study was found and included in the meta-analysis, forest plots would have been visually inspected for the presence of heterogeneity. Formal statistical tests using Cochran's Q statistic (Chi2 test with K-1 degrees of freedom, where K is the number of studies) and the I2 statistic would have been used. Statistical heterogeneity would be considered as present if P value of Chi2 was 0.1 or I2 value was 50% or higher (Higgins 2011).
Assessment of reporting biases
Trials registers were searched and any completed studies which had not been published would have been identified. Outcome reporting bias would have been assessed as part of the risk of bias assessment as described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011).
If sufficient studies were found and included, publication and other reporting biases would have been tested for using funnel plots and appropriate statistical tests.
Meta-analysis would only have been undertaken using comparable studies in which the same outcome measures were reported. Heterogeneity would have firstly been assessed by examining the types of participant, interventions and outcomes in each study. For dichotomous outcomes, the estimate of an intervention would have been summarised as risk ratios with 95% confidence intervals. Any continuous outcomes would have been recorded as mean differences with 95% confidence intervals. Random-effects models would have been used for all analyses involving more than three trials otherwise fixed-effect models were to be used.
If split-mouth studies were included, this would have been combined with data from parallel group trials using the method outlined by Elbourne (Elbourne 2002), using the generic inverse variance method in RevMan.
Subgroup analysis and investigation of heterogeneity
Subgroup analyses would have been undertaken to take account of the use of different inclusion criteria, participants, interventions, techniques, materials, or outcome measures if appropriate.
If sufficient studies were identified, sensitivity analysis would have been undertaken to assess the effects of randomisation, allocation concealment, missing data and blinding on the overall estimates of effect.