Bisgard K M, Folsom A R, Hong C P, Sellers T A
Division of Epidemiology, School of Public Health, University of Minnesota, Minneapolis 55454-1015.
Am J Epidemiol. 1994 May 15;139(10):990-1000. doi: 10.1093/oxfordjournals.aje.a116948.
Most studies are unable to follow nonrespondents prospectively to determine whether their disease rates are comparable with those of the respondents. The authors followed respondents and nonrespondents to a mailed questionnaire, sent to a random sample of Iowa women aged 55-69 years in 1986 (total sample, 98,029; 43% response), to characterize 5-year mortality rates for myocardial infarction and all causes, and attack rates for breast, endometrial, colon, lung, and all-site cancers. Compared with respondents, nonrespondents had higher myocardial infarction (1.47 vs. 0.93 per 1,000 person-years) and all-cause (12.32 vs. 7.89 per 1,000 person-years) mortality. They also had substantially higher attack rates for lung cancer (1.45 vs. 1.10 per 1,000 person-years), and slightly higher attack rates for all-site cancer (11.86 vs. 10.89 per 1,000 person-years). The associations of reported body mass index (weight/height2) with the study endpoints were generally similar among respondents and the total eligible sample, except for a more pronounced U-shaped total mortality association for the nonrespondents. Thus, although the occurrence of several diseases, especially those related to smoking, differed among respondents and nonrespondents, the association of body mass index with cancer occurrence was not appreciably affected by nonresponse bias.
