Alternative approaches for estimating prevalence in epidemiologic surveys with two waves of respondents

Estimates of prevalence in epidemiologic surveys are prone to bias due to selective response. Therefore, much effort is devoted to reduce the number of nonrespondents. For example, individuals who do not respond in the first round of recruitment in mail surveys are usually contacted a second (or even third or fourth) time yielding consecutive waves of responses. Yet this sequence of waves is often neglected in epidemiologic analyses in that prevalence is simply estimated as the proportion of trait-positive individuals among the total group of respondents. This paper investigates alternative estimates of prevalence that might be used in surveys with two waves of respondents. The estimates are based on different assumptions on the relation of response rates with the trait of interest. As this relation is likely to vary from survey to survey depending on the specific circumstances under which the recruitment of participants is conducted, none of the estimates is universally preferable. The performance of the different estimates is assessed in a variety of hypothetical and empirical examples, and strategies are discussed to make the best use of the different estimates in the analysis of epidemiologic studies.