Populations: Are the target and study populations well defined and described? Is the referent group representative of either the unexposed population (in a cross-sectional or cohort design) or of the source population (in a case-control design)? Studies with an inappropriate referent population were given less weight.

Selection of participants: Are the methods for recruiting and enrolling study participants well described? Is there evidence of selection bias? If so, have the authors provided information on the magnitude of the bias? Whether an “effect” is observed in the exposed group is strongly influenced by the choice of the comparison or control group. Thus, the selection and composition of the comparison group is extremely important and in part determines the internal validity of the study. In some cases, there were clear selection biases (for example, selecting comparison groups for the exposed group that did not represent the counterfactual example). That introduces the possibility of selection biases that could easily create the appearance of differences, especially subtle ones, when differences do not exist.

Exposure assessment: How well do the measurements used characterize tetrachloroethylene exposure? How are exposure groups defined? If individual exposure data were available, were they used, or was assignment to exposure groups based on ecologic criteria? In most cases, exposure was estimated at the time of a study. If it is assumed that exposure has only acute, reversible effects, cross-sectional studies are more appropriate. However, if occurrence of an effect when exposure concentrations are low requires long-term exposure, it is important to consider past exposure as well. Exposure assessment ranged from biologic measurements of tetrachloroethylene exposure to environmental exposure assessments. Studies that included measurements and analyses of exposure at the individual level were given greater weight.

Assessment of neurologic outcomes: The end points that were measured in terms of relevance to the visual system and the degree to which the measures are influenced by cognitive function were considered. Studies that used less sensitive measures were given less weight, as were studies that used outcome measures that were more susceptible to observer bias or potential individual confounders (such as ability to follow instructions).

Confounding: Observational studies are always subject to confounding when the exposed and referent groups are imbalanced with respect to factors that are not a result of the exposure but that are also related to the outcome. The committee considered the potential for differences in age, education, learning disabilities, and other variables to confound associations. If the potential for confounding was present and the effects of the confounding were not addressed by the study design or analytic methods, the results of the study were considered to be less credible.

Statistical analysis: Statistical issues were considered, particularly whether the sample size was adequate and whether the approach to analysis was appropriate. Did the studies provide adequate information about the distribution