The accuracy of self-reported health behaviors and risk factors relating to cancer and cardiovascular disease in the general population: a critical review.

Abstract

OBJECTIVE: To critically review the literature concerning the accuracy of self-reported health behaviors and risk factors relating to cancer and cardiovascular disease among the general population. METHOD: A literature search was conducted on three major health research databases: MEDLINE, HealthPLAN, and PsychLit. The bibliographies of located articles were also checked for additional relevant references. Studies meeting the following five inclusion criteria were included in the review: They were investigating the accuracy of self-report among the general population, as opposed to among clinical populations. They employed an adequate and appropriate gold standard. At least 70% of respondents consented to validation, where validation imposed minimal demands on the respondent; and 60% consent to validation was considered acceptable where validation imposed a greater burden. They had a sample size capable of estimating sensitivity and specificity rates with 95% confidence intervals of width +/-10%. The time lag between collection of the self-report and validation data for physical measures did not exceed one month. RESULTS: Twenty-four of 66 identified studies met all the inclusion criteria described above. In the vast majority, self-report data consistently underestimated the proportion of individuals considered "at-risk." Similarly, community prevalences of risk factors were considerably higher according to gold standard data sources than they were according to self-report data. CONCLUSIONS: This review casts serious doubts on the wisdom of relying exclusively on self-reported health information. It suggests that caution should be exercised both when trying to identify at-risk individuals and when estimating the prevalence of risk factors among the general population. The review also suggests a number of ways in which the accuracy of individuals' self-reported health information can be maximized.