Selection bias or confounding - SACEMA选择偏倚和混杂- sacema.pptVIP

* In cohort studies, the exposure distribution within strata of the confounder may be inefficient. For example, say index was home-ownership and reference was apartment dwelling. Outcome is injured in fire. Likely confounded by age. Distribution within adult age groups is likely inefficient -- predominantly apartment dwelling at young ages, predominantly home owners at old ages, maybe balance in the middle. Select apartment dwellers and then match homeowners on age to improve the distribution of exposure within strata of age. In case-control studies, potential for skewed distributions is accentuated by sampling the base. Skewed distributions yield lower precision. Matching assures a more even distribution within strata of the confounder, so better precision. * Choice is sometimes between spending resources (time and money) to accomplish the matching or spending them to increase the study size. Example of excluding cases with no match: Brunet et al (1998 JNCI) studied the effect of smoking on the occurrence of breast cancer among women included in a registry of carriers of BRCA1 or BRCA2 mutations. 300 case subjects. A single control subject was matched to each case according to mutation in the same gene and to age within 1 year. No eligible control for 114 of the 300 cases, so these 114 cases were excluded from the analysis. Is gene mutation likely related to smoking? Why match on it, then? Is age distribution so skewed within smoking stratum that it is a good candidate for matching? These should have been controlled in the analysis without matching. For more, see: Brunet JS, Ghadirian P, Rebbeck TR, Lerman C, Garber JE, Tonin PN, Abrahamson J, Foulkes WD, Daly M, Wagner-Costalas J, Godwin A, Olopade OI, Moslehi R, Liede A, Futreal PA, Weber BL, Lenoir GM, Lynch HT, Narod SA. Effect of smoking on breast cancer in carriers of mutant BRCA1 or BRCA2 genes. J Natl Cancer Inst. 1998 May 20;90(10):761-6. Narod SA, Dube MP, Klijn J, Lubinski J, Lynch HT