4. Criteria for risk of bias assessment.

Random sequence generation (selection bias)	Was the method used to generate the allocation sequence appropriate to produce comparable groups?	This domain was graded low risk of bias if the authors described a random component in the sequence generation process (e.g. random number table, coin tossing, drawing of lots). In split‐mouth study designs, however, we graded low risk of bias where the method of allocating a tooth to an intervention was not random but quasi‐random (systematic methods that were intended to produce similar groups, e.g. sequence generated by odd or even date of birth or by some rule based on date of admission). Although quasi‐random sequence generation methods include some systematic, non‐random component, we considered that in preventive split‐mouth studies (with mainly sound tooth surfaces), the risk of selection bias is minimal. Our justification is based on the assumption that there is no right‐left asymmetry between contralateral teeth regarding caries risk, as shown by Larmas 1995 when they evaluated timing of the change from a sound erupting tooth to a filled tooth.
Allocation concealment (selection bias)	Was the method used to conceal the allocation sequence appropriate to prevent the allocation from being known in advance of, or during, enrolment?	This domain was graded low risk of bias if the authors described adequate concealment (for example, by means of central randomisation, or sequentially numbered, opaque and sealed envelopes), and graded high risk of bias if inadequate concealment was documented (for example, alternation, use of case record numbers, dates of birth or day of the week) or allocation concealment was not used. If there was insufficient or no information on allocation concealment, the judgement was unclear risk. In split‐mouth study designs, however, we considered that the study could be graded low risk of bias if the information about allocation concealment was incomplete or the sequence generation method was quasi‐random. This is because we consider that the risk of selection bias is minimal in preventive split‐mouth designs.
Blinding (performance bias)		We did not assess blinding of participants and personnel in this review because we considered that possible knowledge of personnel and participants about which of the 2 intervention groups a child belongs to was unlikely to cause bias (e.g. affect dental behaviour of a child during the trial, especially when follow‐up is long (≥ 1 year in this review)).
Blinding of outcome assessment (detection bias)	Were outcome assessors blinded to the intervention a participant had received?	Trials comparing sealant to no treatment As sealant materials are visible, blinding of the outcome assessor is possible only if a sealant has been lost. Therefore blinding of outcome assessment for caries will be at high risk of bias for all trials. Trials comparing different sealants We decided to assess the blinding of outcome assessment for caries as high risk of bias for all trials because although sealant materials may have a similar appearance when applied, the outcome assessor can discriminate between them at follow‐up (materials change differently by age, e.g. materials including glass‐ionomer lose their glossy appearance more rapidly than resins).
Incomplete outcome data (attrition bias)	How complete were the outcome data for caries outcomes? Were drop‐out rates and reasons for withdrawals reported? Were missing data imputed appropriately?	In caries prevention studies, follow‐up times can be several years, and the risk of bias for caries outcome data may differ over time. Studies with long follow‐up may have the problem of high drop‐out rates. We decided to base the judgement of this domain on caries efficacy outcomes at 24 or 36 months (commonly used follow‐up times in sealant studies). When both follow‐up times were reported, we based our judgement on 24 months. If either of these two follow‐up times was not reported, we based our judgement on the first caries efficacy outcome reported in the study (which in this review should be at least 1 year). However, we assessed the risk of bias separately and reported it in the risk of bias table for caries outcomes despite the follow‐up times, and we took the assessments into account in the overall risk of bias assessment for caries outcomes within a study. We decided to grade this domain as having low risk if the proportion of missing outcome data was < 25% at 24 or 36 months (in parallel‐group studies), and the groups were balanced in numbers and reasons for missing data; or if missing data have been imputed using appropriate methods. If no information on reasons for drop‐out across intervention groups was provided, or not reported by group, our judgement was unclear risk. Otherwise the study was judged as high risk of bias. Classifying missing data > 25% as having high risk of bias in all study designs was a pragmatic approach to this domain to make the judgement uniform and transparent. If several teeth were sealed in a child’s mouth (a child is a cluster), missing outcome data had to be stated (or counted) at child level (not at tooth level).
Selective reporting (reporting bias)	Were appropriate outcomes reported and were key outcomes missing?	For a trial to be included in this review, caries outcomes had to be reported. However, studies could report the outcome in different ways, for example, incidence of dentinal carious lesion on treated occlusal surfaces of molars or premolars (yes or no); changes in mean figures of decayed, missing and filled occlusal surfaces (DMFS). In this review, selective outcome reporting was graded as ‘low’ risk of bias if the study's prespecified caries outcomes had been reported in the prespecified way.
Other sources of bias: comparability of groups at baseline		We decided to base our judgement of comparability of groups on baseline information given to groups available at follow‐up times because if available information is only that provided at the start of the study, it is impossible to assess whether groups are balanced with each other after follow‐up time as well. The comparability of groups after follow‐up is especially problematic when small studies include children with several teeth and the drop‐out rate is high, even if drop‐outs are balanced in numbers and reasons between groups. If no information on the groups was available at follow‐up time, we decided that if the drop‐out rate (regardless of follow‐up time) was < 25% and drop‐outs were balanced in numbers and reasons by group, our judgement would be based on information given for groups at the start of the study. We decided to grade this domain as having ‘low’ risk of bias if: a) groups were balanced in demographic characteristics (such as sex, age and social class), and in baseline caries risk level; b) possible imbalance of groups at baseline and/or after follow‐up had been taken adequately into account in the analyses. If baseline characteristics in parallel‐group studies were not given to groups available at follow‐up and the drop‐out rate was > 25%, we graded the study as having ‘unclear’ risk.
Other sources of bias: co‐interventions		For a trial to be included in this review, only fluoride toothpaste and fluoridated water were accepted as co‐interventions. We decided to grade this domain as having ‘low’ risk of bias if groups were balanced in number and quality of co‐interventions, or if no co‐interventions were included in the protocol, and as having ‘high’ risk of bias if groups received different numbers or quality of co‐interventions during the trial. If no information was provided on co‐interventions, our judgement was unclear risk.