Random sequence generation

1. Where robust methods of sequence allocation were employed, we will record the risk of bias 'low' (Schultz 2002).

2. Where nonrandom or nonsystematic approaches were employed, we will record the risk of bias as 'high'.

3. Where insufficient detail is provided to make a judgement, we will record the risk of bias as 'unclear'.

Allocation concealment

1. Where robust methods of concealment were employed, and participants and investigators could not determine assignment prior to allocation, we will record the risk of bias as 'low'.

2. Where the possibility for allocation disclosure and consequent selection bias was present, we will record the risk of bias as 'high'.

3. Where insufficient detail is provided to make a judgement, we will record the risk of bias as 'unclear'.

Blinding of participants and personnel

1. Where blinding of participants and study personnel was maintained, or where no blinding or incomplete blinding occurred but the review authors judge that the ocutome was not likely to have been influenced by the lack of blinding, we will record the risk of bias as 'low'.

2. Where no or incomplete blinding occurred and could have affected outcomes, or where blinding occurred but there is a likelihood that it could have been broken and the outcome influenced as a result, we will record the risk of bias as 'high'.

3. Where insufficient detail is provided to make a judgement, we will record the risk of bias as 'unclear'.

Blinding of outcome assessment

1. Where blinding was robustly applied, there was partial blinding of participants or key personnel, or no blinding took place but the review authors judge that the lack of blinding is unlikely to affect the measures employed or reported outcomes of the study, we will record the risk of bias as 'low'.

2. Where incomplete or inefficient blinding occurred, and the measures or outcomes are likely to be affected as a result, we will record the risk of bias as 'high'.

3. Where insufficient detail is provided to make a judgement, we will record the risk of bias as 'unclear'.

Incomplete outcome data

1. Where there are no missing data, the reasons for missing data are unlikely to be related to the true outcome, or the effect of missing data is not enough to have a clinically relevant impact, we will record the risk of bias as 'low'.

2. Where the reason for missing data is likely to be related to outcomes, or is sufficient to produce a clinically relevant bias, we will record the risk of bias as 'high'.

3. Where insufficient detail is provided to make a judgement, we will record the risk of bias as 'unclear'.

Selective Outcome Reporting

1. Where outcomes have been reported in accordance with the protocol, or all the expected outcomes have been presented, we will record the risk of bias as 'low'.

2. Where there is some variance in reporting outcomes from that specified in the protocol, reporting is incomplete, or the study fails to include results for a key outcome, we will record the risk of bias as 'high'.

3. Where insufficient detail is provided to make a judgement, we will record the risk of bias as 'unclear'.