The effect of nonresponse on estimation of relative risk in a case-control study

Although it is understood that low response rates can bias a study's results and that follow-up can increase response rates, the effect of follow-up on the odds ratio estimates in a case-control study is not obvious.

We used the data from a case-control study of renal cell carcinoma conducted in Ontario. Information on risk factors was collected with a mailed questionnaire; the number of telephone or mail follow-ups attempted was recorded. Sex, age group, residence, and follow-up method were known for all cases and controls.

Cases, women, subjects under age 60, subjects living outside of Toronto, and subjects with telephone follow-up were all more likely to be respondents. This pattern of response did not bias the odds ratio estimates. Over all categories of follow-up intensity, the odds ratio estimates for the risk factors varied little. For example, the odds ratio estimate for ever smoking cigarettes ranged from 1.94 to 2.01 for males and from 2.27 to 1.80 for females.

The data indicate that the substantive conclusions of the study would not have changed if response rates had been lower. There is a suggestion, however, that the odds ratios for smoking by men may be overestimated.