Assessment

1. Is the source population (sampling frame) representative of the general population?

Examples of low risk of bias:

• Selection of target population from a representative population roster such as national population registry

Examples of intermediate risk of bias:

• Single community-based study

Examples of high risk of bias:

• Hospital-based patient records; studies where the source population cannot be defined (or enumerated), i.e. any volunteer studies using self-recruitment

2. Is the assessment of the outcome accurate both at baseline and at follow-up?

Examples of low risk of bias:

• Repeated interview or other ascertainment asking about state with validated instrument or method (with demonstrated validity).

Examples of intermediate risk of bias:

• Instrument or method with limited validity assessment and concern of accuracy of responses
• Simple assessment of the presence (or absence) of the symptom(s) without making an effort to quantify the severity/extent
• Use of different instruments at different time points with concern of accuracy of responses

Examples of high risk of bias:

• Unvalidated instrument or method with concern of accuracy of responses
• Uncertain how information was obtained
• Studies with standardized clinical interviews (including physicians’ unstructured assessment of symptoms)
• Studies, which assessed primary outcome as “physician-diagnosed condition”

Examples of low risk of bias:

• High response proportion (rate) both at baseline and follow-up with little missing data
• For instance, response proportions were more than 75% both at baseline and follow-up(s) and missing data within questionnaires less than 10%

Examples of intermediate risk of bias:

• Moderate response proportions both at baseline and follow-up with moderate level of missing data
• For instance, response proportions were 50% to 75% (at baseline and follow-up(s) and missing data with questionnaires less than 15%

Examples of high risk of bias:

• Low response proportion both at baseline and follow-up with high level of missing data
• For instance, response proportions were <50% and missing data with questionnaires more than 15%