The effects of non-response in a prospective study of cancer: 15-year follow-up.

Out of 11,136 Japanese men identified on the island of Oahu, Hawaii in 1965 by the Honolulu Heart Program, 8006 responded to a mailed questionnaire and were examined. Some 1871 responded only to the mailed questionnaire, and 1259 did not respond at all. After 15 years of follow-up, the examined men had significantly lower risk of death from all causes and death from cancer. Minor differences were also noted between the two groups in the risk of cancer of the lung, stomach, colon, and rectum. However, the examined men had a significantly higher risk of prostate cancer. In general, the strength of these non-response effects was mainly due to risk differences in the first five years of the 15-year follow-up period. The relative risk (RR) of each of the seven endpoint events tended towards 1.0 as each of the three successive five-year follow-up intervals were considered. An exception to this was the prostate cancer incidence RR which favoured the unexamined men throughout the entire 15 years, but significantly so only in the last five-year follow-up interval. When the 8006 examined and 1871 unexamined men who responded to the mailed questionnaire were evaluated with respect to the association of cigarette smoking with lung cancer incidence, the RR for smokers was 9.77 for the examined men, and 6.73 for the unexamined men. Since these RRs are not significantly different, there should be little bias in RR estimates of cigarette smoking for lung cancer if the observation was limited to only the examined men. With regard to the association of body mass index (BMI) with colon cancer in older men, the RRs for men in the highest BMI quintile were quite comparable, at 1.37 for the examined group and 1.60 for the unexamined men. We conclude that although some non-response effects on cancer incidence exist in this cohort, they do not appear to be serious enough to have changed conclusions drawn about risk relationships.