Reliability of self-reported reproductive factors and childhood social class indicators in a case-control study in...

Reproductive factors are often evaluated in epidemiologic interview studies as risk factors for diseases in women. Similarly, childhood social class has been implicated in the etiology of several diseases. Nevertheless, questions related to these factors have not been thoroughly evaluated for test-retest reliability. This research measured the test-retest reliability of reproductive and childhood social class variables, and determined whether reliability differed by case-control status, age, educational level, time between interviews, and interviewer-rated quality of the interview.

Subjects were participants in a population-based case-control in-person interview study of Hodgkin's disease in northern California women. Twenty-four cases and 22 controls were reinterviewed by telephone between 1992 and 1995, with an average interval of 8 months between interviews. Reliability was assessed using kappa or intraclass correlation coefficients; mean reliability coefficients and 95% confidence intervals (CIs) were estimated using the bootstrap method.

Reliability was excellent for all variables (reliability coefficients between 0.76 and 0.96) and did not differ by case-control status (mean reliability = 0.82 for cases and 0.84 for controls), age (mean reliability = 0.85 for age < 40 and 0.82 for age > or = 40), time between interviews (mean reliability = 0.75 for 0-5 months, 0.88 for 6-11 months, and 0.87 for 1 year or more), or interviewer-rated quality of the validity of the original responses (mean reliability = 0.93 for "not too confident" and 0.83 for "confident"). However, reliability was consistently lower among less educated women (mean reliability = 0.56 for high school or less and 0.88 for more than high school), a finding consistent with results of prior studies.

These results indicate that questions about reproductive experience and childhood social class posed in in-person interviews can be answered reliably. However, inclusion of subjects at lower socioeconomic status may result in lower reliability for some interview responses.