Details of the risk of bias assessments for each of the included studies are presented in the 'Risk of Bias' tables in the Characteristics of included studies table, and in Figure 2.
Figure 2. Risk of bias summary: review authors' judgements about each risk of bias item for each included study.
Random sequence generation (selection bias) Allocation concealment (selection bias) Selective cluster recruitment Blinding of participants and personnel (performance bias): All outcomes Blinding of objective outcome assessment (detection bias) Blinding of subjective outcome assessment (detection bias) Incomplete outcome data (attrition bias): All outcomes Selective reporting (reporting bias) Other bias
Abdul Rashid 2013 + ? - + + ? +
Barnabas 2016 + + - ? ? ? ?
Belzer 2015 ? ? - + - - ? +
Bull 2016 ? ? - - - - -
-Castano 2012 + + - - - +
-Constant 2014 + + - + - + +
-Cook 2015 ? ? + ? ? - -
-da Costa 2012 + + - + - - ? +
Delamere 2006 ? ? ? ? ? ? ?
de Tolly 2012 ? ? - - - -
-Downing 2013 + ? + + + ? +
Garofalo 2016 + + - ? - + + +
Gerdts 2015 ? ? - ? - ? ?
Gold 2011 - ? - - - ? ?
Hou 2010 + + - + - - ?
-Huang 2013 ? + - ? - - ? +
Ingersoll 2015 + + - ? ? ? +
Jeffries 2016 ? ? - ? ? - ?
Joseph Davey 2016 + + - ? - ? +
Lee 2016 ? ? - ? + ? ?
Leiby 2016 ? ? ? + ? - ?
-Lester 2010 + + - + - ? + +
Lim 2012 + ? - ? - - + +
Figure 2. (Continued)
Four studies reported adequate random sequence generation and allocation concealment, and so were at low risk of bias in both of these domains (Castano 2012; Garofalo 2016; McCarthy 2016; Ybarra 2017). Three studies were at low risk of bias for the random sequence generation domain, but did not provide sufficient information on their allocation procedures and therefore were at unclear risk of bias for the allocation concealment domain (Lim 2012; Rokicki 2017; Suffoletto 2013). The remaining studies were at either unclear or high risk of bias for both the random sequence generation and the allocation concealment domains (Bull 2016; Belzer 2015; Delamere 2006; Gold 2011; Jeffries 2016;
Reed 2014).
Adults
Ten studies reported adequate random sequence generation and allocation concealment, and so were at low risk of bias in both of these domains (Barnabas 2016; Constant 2014; da Costa 2012; Hou 2010; Ingersoll 2015; Joseph Davey 2016; Lester 2010; Mbuagbaw 2012; Odeny 2012; Smith 2015). A further five studies were at low risk of bias for the random sequence generation domain, but at unclear risk of bias for allocation concealment (Abdul Rashid 2013; Downing 2013; Norton 2014; Pop-Eleches 2011; Young 2015).
Three studies were at low risk of bias for allocation concealment, but provided insufficient information on their random sequence generation and so were categorised as having unclear risk of bias on this domain (Huang 2013; Mugo 2016; Shet 2014). The rest of the trials were at unclear risk of bias for both the random sequence generation and the allocation concealment domains (Cook 2015; de Tolly 2012; Gerdts 2015; Lee 2016; Leiby 2016; Nsagha 2016; Ruan 2017; Russell 2012; Rutland 2012).
Blinding Adolescents
Given the nature of the interventions, the majority of trials were unable to blind participants and were considered at high risk bias on this domain, and two studies did not provide sufficient information and therefore were was at unclear risk of bias on the blinding of participants and personnel domain (Delamere 2006;
Ybarra 2017).
Two studies reported adequate blinding of objective outcome assessment and were at low risk of bias on this domain (Belzer 2015;
McCarthy 2016). Six studies did not provide sufficient information on blinding of objective outcome assessment and so were at unclear risk of bias (Delamere 2006; Garofalo 2016; Jeffries 2016;
Lim 2012; Reed 2014; Suffoletto 2013). A further five studies did not measure objective outcome measures and so were not applicable for this domain (Bull 2016; Castano 2012; Gold 2011; Rokicki 2017;
Ybarra 2017).
One study was at low risk of bias for the blinding of subjective outcomes domain (Ybarra 2017), nine studies were at high risk of bias for this domain (Belzer 2015; Bull 2016; Castano 2012; Garofalo 2016; Gold 2011; Lim 2012; McCarthy 2016; Rokicki 2017; Suffoletto 2013). The remaining studies did not record subjective outcomes (Delamere 2006; Jeffries 2016; Reed 2014).
Adults
Two studies were at low risk of bias for blinding of participants and personnel (Cook 2015; Downing 2013). As was the case for trials conducted among adolescent populations, the majority (N = 20) of trials were unable to blind participants and were considered at high risk bias on this domain. The remaining six studies did not provide sufficient information and therefore were at unclear risk of bias on
the blinding of participants and personnel domain (Delamere 2006;
Leiby 2016; Norton 2014; Pop-Eleches 2011; Russell 2012; Young 2015).
Ten studies reported adequate blinding of objective outcome assessment and were at low risk of bias on this domain (Abdul Rashid 2013; Cook 2015; Constant 2014; da Costa 2012; Downing 2013; Hou 2010; Leiby 2016; Lester 2010; Mugo 2016; Odeny 2012).
Eleven trials were at unclear risk of bias for the blinding of objective outcome assessment domain (Barnabas 2016; Cook 2015; Gerdts 2015; Huang 2013; Ingersoll 2015; Joseph Davey 2016; Norton 2014;
Pop-Eleches 2011; Russell 2012; Rutland 2012; Young 2015). Eight studies did not report objective outcome measures and so were not applicable for this domain (Barnabas 2016; de Tolly 2012; Lee 2016;
Mbuagbaw 2012; Nsagha 2016; Ruan 2017; Shet 2014; Smith 2015).
One study was at low risk of bias for the blinding of subjective outcomes domain (Mbuagbaw 2012). Six studies were at unclear risk of bias for blinding of subjective outcomes (Barnabas 2016; Lee 2016; Leiby 2016; Russell 2012; Rutland 2012; Smith 2015). Eight studies were at high risk of bias on this domain (Constant 2014;
da Costa 2012; de Tolly 2012; Hou 2010; Huang 2013; Lester 2010;
Nsagha 2016; Ruan 2017). The remaining studies did not record subjective outcomes (Abdul Rashid 2013; Barnabas 2016; Downing 2013; Gerdts 2015; Ingersoll 2015; Joseph Davey 2016; Mugo 2016;
Norton 2014; Odeny 2012; Pop-Eleches 2011; Shet 2014; Young remaining eight studies were at unclear (Delamere 2006; Jeffries 2016) or high risk of bias (Belzer 2015; Bull 2016; Castano 2012; Gold 2011; Lim 2012; Suffoletto 2013) on this domain.
Adults
Ten studies among adult populations were at low risk of bias for the incomplete outcome data domain, reporting low levels of loss to follow-up (Abdul Rashid 2013; Constant 2014; Downing 2013; Lee 2016; Mugo 2016; Nsagha 2016; Odeny 2012; Ruan 2017; Shet 2014;
Young 2015). Seven studies did not provide sufficient information, and so were at unclear risk of bias on this domain (Barnabas 2016; Ingersoll 2015; Lester 2010; Mbuagbaw 2012; Pop-Eleches 2011; Russell 2012; Rutland 2012). The remaining 10 studies were considered to be at high risk of bias for the incomplete outcome data domain due to the high levels of attrition reported (Cook 2015;
da Costa 2012; de Tolly 2012; Gerdts 2015; Hou 2010; Huang 2013;
Joseph Davey 2016; Leiby 2016; Norton 2014; Smith 2015).
Selective reporting Adolescents
Four studies among adolescent populations were at low risk of bias for the selective outcome reporting domain as their protocols and/or trial registry entries could be identified and all expected outcomes were reported (Castano 2012; Garofalo 2016; Lim 2012;
McCarthy 2016). Seven studies were at unclear risk of bias as their protocols could not be identified (Belzer 2015; Delamere 2006;
Gold 2011; Reed 2014; Rokicki 2017; Suffoletto 2013; Ybarra 2017).
Two studies were at high risk of bias on the selective outcome reporting domain due to inconsistencies between the pre-specified and actual outcome reporting (Bull 2016; Jeffries 2016).
Adults
Six studies were at low risk of bias for selective outcome reporting (Constant 2014; Lester 2010; Mbuagbaw 2012; Mugo 2016; Pop-Eleches 2011; Smith 2015). Seventeen studies were at unclear risk of bias (Abdul Rashid 2013; Barnabas 2016; da Costa 2012;
Downing 2013; Gerdts 2015; Hou 2010; Huang 2013; Ingersoll 2015;
Joseph Davey 2016; Lee 2016; Leiby 2016; Norton 2014; Nsagha 2016; Ruan 2017; Russell 2012; Rutland 2012; Young 2015). The remaining four studies were at high risk of bias on this domain due to inconsistencies between the pre-specified and actual outcome reporting (Cook 2015; de Tolly 2012; Odeny 2012; Shet 2014).
Selective cluster recruitment Adolescents
Two trials among adolescent populations were cluster RCTs, both of which were categorised as being at high risk of bias for selective cluster recruitment (Bull 2016; Rokicki 2017).
Adults
None of trials carried out among adult populations were cluster RCTs.
Other potential sources of bias Adolescents
Four studies among adolescent populations were at high risk of other bias as a result of issues such as conducting only per protocol analyses, available case analyses, or substantial baseline imbalances related to the outcomes under study (Bull 2016;
Castano 2012; Gold 2011; Reed 2014). Six trials were at low risk of other bias (Belzer 2015; Garofalo 2016; Lim 2012; McCarthy 2016;
Rokicki 2017; Ybarra 2017), and three were at unclear risk of other bias (Delamere 2006; Jeffries 2016; Suffoletto 2013).
Adults
Five studies among adults were considered to be at high risk of other bias as a result of conducting only per protocol analyses, available case analyses, substantial contamination, or substantial baseline imbalances related to the outcomes under study (Constant 2014; Cook 2015; de Tolly 2012; Hou 2010; Leiby 2016). Fourteen trials were at low risk of other bias (Abdul Rashid 2013; da Costa 2012; Downing 2013; Huang 2013; Ingersoll 2015; Joseph Davey 2016; Lester 2010; Mbuagbaw 2012; Mugo 2016; Nsagha 2016;
Odeny 2012; Ruan 2017;Smith 2015; Young 2015), and six were at unclear risk of other bias (Barnabas 2016; Gerdts 2015; Lee 2016;
Norton 2014; Pop-Eleches 2011; Russell 2012; Rutland 2012; Shet 2014).