Tobacco Documents Online

A I-lT~9 ® 0 Amsundu.wnlEpwstlebpr . Vd.141.Na4 CopytlpM O 1GY6 by iM Jolvw Fkplbq LMMnaay adml d Hy9in. rd Pubic HWN AN ~IptN Mawd P/IuOrdfn U.sA ? Dr. Weiss Replies: Analytic Approaches for Dealing with Possible Recall :. Bias In Case-Control Stya~it ~ . ~ NOTICE : Noel 5. We/ss Recall bias is one of the principal threats to the validity of an interview- or questionnaire-based case- control study. Most strategies to ward off this threat center around the data collection process; they in- volve, for example, designing questions in such a way as to minimize the potential for inaxurate reporting of exposure histories by cases and controls. Mitchell et al. (1) recommend complementing these activities "when there 1s good reason to believe that knowledge of the hypothesis might distort subjects' reporting of exposure" by excluding subjects aware of the hypath- esis at the time of interview. Their approach will yield a less biased result than an analysis that includes all subjects to the extent lhat knowledge of the hypothe- sis: a) impairs accurate reporting; and b) is acquired to a different degree by cases than controls. If either condition (a) or (b) above is not met, the result this approach provides will be no Iess biased, but will be less precise in proportion to the prevalence of "knowledgeable" participants in the study. The ap- Ftse.Wed for pUbaeaUnn 8eptemEer B. 1994. From the aohuol of Publ{c Hetl11, errJ Qommunay Msdkdne, Deparknsnt of EpidsmWlo9y. 80•X Univeally of WashYqton, SaaNe, WA96196. Ps ertarhl may bS kaVtiN blr saPYriQh1 14r6iul! jj9$ @0b praaeh will yield a more biased result to the degree that exposed cases selectively become aware of the hypothesis following their diagnosis (2). I contend that it is the rare situation indeed in which we have information regarding the influence of "knowledge" on the accuracy of reporting of expo- sures, or on the timing of acquisition of knowledge of the hypothesis in relation to diagnosis among exposed and nonexposed cases. Mitchell et al. and I agree that the data from case-control studies themselves are not informative in these regards. Therefore, though it may be reasonable to present an analysis that excludes subjects who have knowledge of the hypothesis, it should be the unusual dreumstance in which we em- phasize the results of such an analysis. REFERENCES 1. MitcLell AA, weder MM, Shapiro S. wrespome to the commenhtjr "Should we cuusidel a subjed's knowledge of the etblopcL9poth in the analysts of nfe-oomtat stud- ies?" Am J Epldemio1~199J:14129'/-& 2. Weiss N5. Should we consider a subjea's imoavledae of the eHologic hypothesis in the andysis of nso-cmuol stadies4" Am J Bpidemiol 1994;139247-9. • I t 0 2110