We recently reviewed the provocative study by Thorson and colleagues1 that documented an association between exposure to sick or dead birds and the occurrence of mild flulike illness in our introductory course in epidemiology. When preparing for this session, we noticed 2 issues that we would like to bring to the attention of the journal's readers. First, the exposure frequency distribution provided in Table 1 is nested.1 That is, those persons exposed to “direct contact with sick or dead poultry” (n = 6702) are also included in the less severe exposure group “sick or dead poultry in household” (n = 11 755) and both of these groups are included in the exposure group “raise or keep poultry in household” (n = 38 373). We are concerned therefore, that the corresponding multivariate analysis may have been conducted on many more observations than numbers of persons in the study. This approach violates the independence assumption required. Second, the crude odds ratios (whether one uses the nested exposure variable or derives the mutually exclusive form) indicate a protective effect for the least-exposed group and null effects for the other 2 groups. It would have been useful to know more about the nature of the negative confounding that was present in these data to better understand the occurrence of flulike illness in this population.
Groome PA, Richardson H. Flulike Illness and Exposure to Sick or Dead Poultry. Arch Intern Med. 2006;166(13):1420-1421. doi:10.1001/archinte.166.13.1420-b