Table 4.

Main Biases Domains to Be Considered When Assessing Quality of Non-Randomized Studies

Biases Domains	Description	Details/Example
Confounding Biases	Presence of factors that may influence the results such that the observed association between intervention and outcomes differs from the causal effect.	May be referred to residual confounding (not appropriate analyses of known and measured confounders) or to unmeasured confounding (factors not measured at all or not included in the analyses).
Selection Biases	Exclusion of subjects (or events/outcomes or follow-up time) that lead to systematic errors in the estimated association between intervention and outcomes.	Bias in selection of subjects can be related to both intervention and outcome. Some example are immortal-time-bias, or bias arising from exclusion of subjects with missing data.
Information Biases (measurement bias)	Presence of Misclassification of intervention status or outcomes.	Misclassification of intervention status might happen on retrospective cohort studies if availability of information on interventions are influenced by outcomes (a.k.a. recall bias). Misclassification of outcomes (detection bias), eg, when intensities of observations/measurement of outcomes differs between the intervention groups.
Reporting Biases	Selection of the reported results arising from a desire for findings to merit publication.	Biases arise when the selection of results is based on P-values, magnitude or direction of the estimated effect of intervention. Might concern selection of outcomes, selection of the analyses, or selective reporting of a subgroup of participants.