Table 1.

Questions to guide the development of criteria for each domain in epidemiology studies.

Domain and core question (target bias)	Prompting questions	Follow-up questions
Exposure measurement  Does the exposure measure reliably distinguish between levels of exposure in a time window considered most relevant for a causal effect with respect to the development of the outcome?  (information bias)	For all:  ● Does the exposure measure capture the variability in exposure among the participants, considering intensity, frequency, and duration of exposure?  ● Does the exposure measure reflect a relevant time window? If not, can the relationship between measures in this time and the relevant time window be estimated reliably?  ● Was the exposure measurement likely to be affected by a knowledge of the outcome?  ● Was the exposure measurement likely to be affected by the presence of the outcome (i.e., reverse causality)? For case-control studies of occupational exposures:  ● Is exposure based on a comprehensive job history describing tasks, setting, time period, and use of specific materials? For biomarkers of exposure, general population:  ● Is a standard assay used? What are the intra- and interassay coefficients of variation? Is the assay likely to be affected by contamination? Are values less than the limit of detection dealt with adequately?  ● What exposure time-period is reflected by the biomarker? If the half-life is short, what is the correlation between serial measurements of exposure?	Is the degree of exposure misclassification likely to vary by exposure level? If the correlation between exposure measurements is moderate, is there an adequate statistical approach to ameliorate variability in measurements? If there is a concern about the potential for bias, what is the predicted direction or distortion of the bias on the effect estimate (if there is enough information)?
Outcome ascertainment  Does the outcome measure reliably distinguish the presence or absence (or degree of severity) of the outcome?  (information bias)	For all:  ● Is outcome ascertainment likely to be affected by knowledge of exposure (e.g., consider access to health care, if based on self-reported history of diagnosis)? For case-control studies:  ● Is the comparison group without the outcome (e.g., controls in a case-control study) based on objective criteria with little or no likelihood of inclusion of people with the disease? For mortality measures:  ● How well does cause of death data reflect occurrence of the disease in an individual? How well do mortality data reflect incidence of the disease? For diagnosis of disease measures:  ● Is diagnosis based on standard clinical criteria? If based on self-report of diagnosis, what is the validity of this measure? For laboratory-based measures (e.g., hormone levels):  ● Is a standard assay used? Does the assay have an acceptable level of interassay variability? Is the sensitivity of the assay appropriate for the outcome measure in this study population?	Is there a concern that any outcome misclassification is nondifferential, differential, or both? What is the predicted direction or distortion of the bias on the effect estimate (if there is enough information)?
Participant selection Is there evidence that selection into or out of the study (or analysis sample) was jointly related to exposure and to outcome? (selection bias including attrition [primarily], confounding)	For longitudinal cohort:  ● Did participants volunteer for the cohort based on knowledge of exposure and/or preclinical disease symptoms? Was entry into the cohort or continuation in the cohort related to exposure and outcome? For occupational cohort:  ● Did entry into the cohort begin with the start of the exposure?  ● Was follow-up or outcome assessment incomplete, and if so, was follow-up related to both exposure and outcome status?  ● Could exposure produce symptoms that would result in a change in work assignment/work status (“healthy worker survivor effect”)? For case-control study:  ● Were controls representative of population and time periods from which cases were drawn?  ● Are hospital controls selected from a group whose reason for admission is independent of exposure?  ● Could recruitment strategies, eligibility criteria, or participation rates result in differential participation relating to both disease and exposure? For population-based survey:  ● Was recruitment based on advertisement to people with knowledge of exposure, outcome, and hypothesis?	Were differences in participant enrollment and follow-up evaluated to assess bias? If there is a concern about the potential for bias, what is the predicted direction or distortion of the bias on the effect estimate (if there is enough information)? Were appropriate analyses performed to address changing exposures over time in relation to symptoms? Is there a comparison of participants and nonparticipants to address whether differential selection is likely?
Confounding  Is confounding of the effect of the exposure likely?  (confounding)	Is confounding adequately addressed by considerations in…  a. … participant selection (matching or restriction)?  b. … accurate information on potential confounders, and statistical adjustment procedures?  c. … lack of association between confounder and outcome, or confounder and exposure in the study?  d. … information from other sources? Is the assessment of confounders based on a thoughtful review of published literature, potential relationships (e.g., as can be gained through directed acyclic graphing), minimizing potential overcontrol (e.g., inclusion of a variable on the pathway between exposure and outcome)?	If there is a concern about the potential for bias, what is the predicted direction or distortion of the bias on the effect estimate (if there is enough information)?
Analysis  Does the analysis strategy and presentation convey the necessary familiarity with the data and assumptions?a	● Are missing outcome, exposure, and covariate data recognized, and if necessary, accounted for in the analysis?  ● Does the analysis appropriately consider variable distributions and modeling assumptions?  ● Does the analysis appropriately consider subgroups of interest (e.g., based on variability in exposure level or duration, or susceptibility)?  ●Is an appropriate analysis used for the study design?  ● Is effect modification considered, based on considerations developed a priori?  ● Does the study include additional analyses addressing potential biases or limitations (i.e., sensitivity analyses)?	If there is a concern about the potential for bias, what is the predicted direction or distortion of the bias on the effect estimate (if there is enough information)?

^aThe evaluation of the analysis domain is focused on the appropriateness of the analysis rather than any specific form of bias.