Estimation of age-specific incidence rates from cross-sectional survey data.

Age-specific disease incidence rates are typically estimated from longitudinal data, where disease-free subjects are followed over time and incident cases are observed. However, longitudinal studies have substantial cost and time requirements, not to mention other challenges such as loss to follow up. Alternatively, cross-sectional data can be used to estimate age-specific incidence rates in a more timely and cost-effective manner. Such studies rely on self-report of onset age. Self-reported onset age is subject to measurement error and bias. In this paper, we use a Bayesian bivariate smoothing approach to estimate age-specific incidence rates from cross-sectional survey data. Rates are modeled as a smooth function of age and lag (difference between age and onset age), with larger values of lag effectively down weighted, as they are assumed to be less reliable. We conduct an extensive simulation study to investigate the extent to which measurement error and bias in the reported onset age affects inference using the proposed methods. We use data from a national headache survey to estimate age- and gender-specific migraine incidence rates.