diagnosis may more severely misrepresent historical exposure among cases than among controls, leading to bias; this bias would be downward if behavioral changes lead to lower exposure, and upward otherwise. Of the three studies chosen by Elwood, the McBride study (which showed the weakest association) relied on personal measures; the Linet and UKCC studies did not, nor did most other studies in the pooled analyses.
Finally, meta-analysis and pooled analysis are not synonyms. Meta-analysis uses only previously published information; hence the ''meta-analyst'' is stuck with decisions made by the original authors. In contrast, pooled analysis uses raw data from previous studies, and thus can apply identical analyses to all included studies. The choices of cutpoints, reference groups, metrics, etc., in a pooled analysis may differ from the choices made in the original studies and may result in changes in the study-specific estimates. In addition to statistically more stable results, pooling allows comparison of results across different metrics and across different studies, free of artifacts introduced by analytic differences. Pooled analyses have yet to detect any relation of study results to study design or conduct, which further undermines the basis for restricting attention to a select few studies.