Table 3.

Criteria for diabetes and insulin resistance.

Level	Exposure measurement	Outcome ascertainment	Participant selection	Confounding	Analysis
Good	Will be developed for the specific exposure Diabetes  ● Exposure must be measured before diabetes onset (e.g., prospective cohort). Insulin resistance  ● Can be measured concurrent with outcome. Criteria that apply to all exposures ● Valid exposure assessment methods used which represent the etiologically relevant time period of interest.  ● Exposure misclassification is expected to be minimal.	Diabetes  ● For undiagnosed cases, American Diabetes Association (ADA) definition with or without repeated measures, and:  ο Indicates fasting was required.  ο Includes information on the laboratory test requirements for hemoglobin A1c and for conducting the oral glucose tolerance tests.  ο Provides some information on assays, quality control procedures, reliability or validity measures.  ● For diagnosed cases, self-reported physician diagnosis or medical treatment for diabetes can be used. Participants without a diagnosis should be tested to avoid false negatives. Insulin resistance  ● Homeostatic model assessment (HOMA) of insulin resistance (HOMA-IR) and β-cell function (HOMA-β) in the absence of diabetes from fasting glucose and insulin concentrations measured in plasma. AND/OR  ● Fasting insulin and fasting glucose measures – indicates fasting was required. Fasting time accounted for in analysis if compliance was an issue. AND  ● Provides some information on assays, quality control procedures, reliability or validity measures.	Diabetes  ● Prospective cohort or other design that allows for identification of incident disease.  ● Must exclude individuals with diabetes at baseline. Prediabetes can be included but should be addressed in analysis via stratification or sensitivity analysis.  ● Cohort followed for sufficient period to allow for development of disease. Insulin resistance  ● Cross-sectional studies are appropriate.  ● Must exclude individuals with known diabetes. Note for both: for studies of children/adolescents, lack of exclusion of diabetes cases may be acceptable due to low prevalence. Criteria that apply to all outcomes  ● Minimal concern for selection bias based on description of recruitment process.  ● Exclusion and inclusion criteria specified and would not induce bias.  ● Participation rate is reported at all steps of study (e.g., initial enrollment, follow-up, selection into analysis sample). If rate is not high, there is appropriate rationale for why it is unlikely to be related to exposure (e.g., comparison between participants and nonparticipants or other available information indicates differential selection is not likely).	Diabetes and Insulin resistance  ● Risk factors that should be considered as possible confounders include age, gender, BMI/adiposity, lifestyle factors (physical activity, diet, smoking, alcohol), SES, co-exposures Criteria that apply to all outcomes Contains all of the following:  ● Conveys strategy for identifying key confounders. This may include: a priori biological considerations, published literature, or statistical analyses; with recognition that not all “risk factors” are confounders.  ● Inclusion of potential confounders in statistical models not based solely on statistical significance criteria (e.g., p < 0.05 from stepwise regression).  ● Does not include variables in the models that have been shown to be influential colliders or intermediates on the causal pathway.  ● Key confounders are evaluated and considered to be unlikely sources of substantial bias. This will often include:  ο Presenting the distribution of potential confounders by levels of the exposure and/or the outcomes of interest (with amount of missing data noted); or  ο Consideration that potential confounders were rare among the study population, or were expected to be poorly correlated with exposure of interest; or  ο Presenting a progression of model results with adjustment for different potential confounders, including consideration of the function forms of potential confounders, if warranted.	Criteria that apply to all outcomes  ● Use of an optimal characterization of the outcome variable.  ● Quantitative results presented (e.g., effect estimates and confidence limits).  ● Descriptive information about outcome and exposure provided (where applicable):  ο Amount of missing data noted and addressed appropriately.  ο For exposure, includes LOD (and percentage less than LOD) and discussion of cut-points and transformations.  ● Includes analyses that address robustness of findings (e.g., examination of exposure-response, relevant sensitivity analyses). Effect modification examined based only on a priori rationale with sufficient numbers.  ● No deficiencies in analysis evidence. Discussion of some details may be absent (e.g., examination of outliers).
Adequate	Will be developed for the specific exposure Diabetes  ● Exposure must be measured before diabetes onset. Insulin resistance  ● Can be measured concurrent with outcome. Criteria that apply to all exposures  ● Valid exposure assessment methods used which represent the etiologically relevant time period of interest.  ● Exposure misclassification may exist but is not expected to greatly change the effect estimate.	Diabetes  ● ADA definition without repeated measures, and no other information provided. or  ● The use of medical records with details on criteria used to define diabetes, in the absence of laboratory tests conducted specifically for the study. Insulin resistance  ● Fasting insulin and fasting glucose measures – no details of fasting or laboratory assays provided.	Same as Good, but: Criteria that apply to all outcomes  ● Enough of a description of the recruitment process to be comfortable that there is no serious risk of bias.  ● Inclusion and exclusion criteria specified and would not induce bias.  ● Participation rate is incompletely reported but available information indicates participation is unlikely to be related to exposure.	Criteria that apply to all outcomes Similar to Good, but may not have included all key confounders, or less detail may be available on the evaluation of confounders (e.g., sub-bullets in Good). It is possible that residual confounding could explain part of the observed effect,but concern is minimal.	Criteria that apply to all outcomes Same as Good, except:  ● Descriptive information about exposure provided but may be incomplete; might not have discussed missing data, or cut-points, or shape of distribution. or  ● Some important analyses that address the robustness of findings are not performed.
Deficient	Will be developed for the specific exposure Diabetes  ● Exposure must be measured before diabetes onset unless the exposure is persistent in the body. Insulin resistance  ● Can be measured concurrent with outcome. Criteria that apply to all exposures  ● Valid exposure assessment methods used which represent the etiologically relevant time period of interest. There may be concerns about reverse causality, but there is no direct evidence that it is influencing the effect estimate.  ● Exposed groups are expected to contain a notable proportion of unexposed or minimally exposed individuals, the method did not capture important temporal or spatial variation, or there is other evidence of exposure misclassification.	Diabetes  ● Treatment for diabetes or use of medical records without details on criteria used to define diabetes. or  ● Self-reported physician diagnosis is the sole criteria for case ascertainment; participants without a diagnosis are not tested and presumed to be non-cases. Insulin resistance – none defined.	Criteria that apply to all outcomes  ● Little information on recruitment process, selection strategy, sampling framework, and/or participation. OR  ● Aspects of the recruitment process, selection strategy, sampling framework, or participation raise the potential for bias (e.g., healthy worker effect, survivor bias).	Criteria that apply to all outcomes  ● Does not include variables in the models that have been shown to be influential colliders or intermediates on the causal pathway. And any of the following:  ● The potential for bias to explain some of the results is high based on an inability to rule out residual confounding, such as a lack of demonstration that key confounders of the exposure-outcome relationship were considered.  ● Descriptive information on key confounders (e.g., their relationship relative to the outcomes and exposure levels) are not presented.  ● Strategy of evaluating confounding is unclear or is not recommended (e.g., based on statistical significance criteria or stepwise regression only).	Criteria that apply to all outcomes  ● Descriptive information about exposure levels not provided (where applicable). or  ● Effect estimate presented without standard error or confidence interval. or  ● Non-optimal analysis methods used (e.g., correlation instead of linear regression)
Critically deficient	Will be developed for the specific exposure Diabetes  ● Exposure measure does not reflect exposure before onset of diabetes or is known to be affected by disease status. Criteria that apply to all exposures  ● Exposure measurement does not characterize the etiologically relevant period for the outcome or is not valid.  ● There is direct evidence that reverse causality could account for the observed association.  ● Exposure measurement was not independent of outcome status.	Diabetes  ● Studies using self-reported diabetes with no clear information on the questions used to ascertain diabetes status.  ● Use of glucosuria to identify diabetes cases.  ● Use of diabetes mortality. Insulin resistance  ● Reporting HOMA-β in the absence of HOMA-IR Criteria that apply to all outcomes  ● Invalid/insensitive marker of outcome.  ● Outcome ascertainment was not independent from exposure status.	Criteria that apply to all outcomes  ● Aspects of the processes for recruitment, selection strategy, sampling framework, or participation, or specific available data result in concern that bias resulted in a large impact on effect estimates.	Criteria that apply to all outcomes  ● Includes variables in the models that have been shown to be influential colliers or intermediates in the causal pathway, indicating that substantial bias is likely from this adjustment. or  ● Confounding is likely present and not accounted for, indicating that the results were most likely due to bias (i.e., confounders associated with the outcome and exposure in the study could explain the majority of the reported results).	Criteria that apply to all outcomes  ● Results presented as statistically “significant”/“not significant” or just p-values (i.e., without including effect estimates). or  ● Effect modification examined without clear a priori rationale and without providing main effects. or  ● Analysis methods were not appropriate for the design or data.