Non-respondent Bias Research Articles

Differences between respondents and non-respondents in a population-based cardiovascular disease study.

The differences in cardiovascular health status between participants and non-participants were examined in a population-based cardiovascular study. Telephone interviews with non-respondents revealed generally more cardiovascular disease but less hyperlipidemia and family history of cardiovascular disease. Non-respondents did not differ regarding known hypertension, diet or drug therapy for hyperlipidemia, or egg use. Non-respondents were more likely to be cigarette smokers. Because the amount of non-respondent bias in the study was small while the response rate was high, respondents were generally representative of the target population. However, the observed differences could have produced spuriously high estimates of risk factor prevalence, low estimates of disease prevalence, and biased relative risks if the non-response rate and/or the baseline differences had been considerably larger.

Patient interview study from the Third National Cancer Survey: overview of problems and potentials of these data.

An interview requesting information on selected epidemiologic variables from the Third National Cancer Survey (TNCS) was sought for 13,179 incident cases, a random 10% sample in eight of nine geographic areas of the United States with a combined population of 21 million. An overall response of 7,518 (57%) was obtained. Nonresponse was due to refusal by physicians (14%) or patients (7%), early patient deaths (7%), no one available for Interview (5%), and tracing problems (10%). Comparison of interview respondents with all TNCS patients for age, race, sex, marital status, method of diagnosis, vital status, country of birth, and site distribution showed only small differences and no strong nonrespondent bias. These data contain information on tobacco and alcohol consumption, occupation and Industry, socioeconomic status, height, weight, obstetric and gynecologic history, and selected chronic diseases. In the absence of normal controls, two approaches to test for associations between these variables and specific cancer sites are suggested: 1) an “intercancer comparison” of patients with cancer at one site with those having cancers at other sites in the interview study, and 2) an “external comparison” of interview patients with subjects of other studies with comparable data. The need for caution with either approach is emphasized in an effort to prevent misinterpretations of these data.