Combining cross-sectional data on prevalence with risk estimates from a prediction model. A novel method for estimating the attributable risk

OBJECTIVES: Estimation of the attributable risk for fatal diseases by combining two different data sources.

METHODS: We derive a method to estimate the attributable risks of different risk factors by combining general mortality risks with up-to-date prevalences of the risk factors using estimates from a risk prediction model and cross-sectional data of a cohort study. Partial attributable risks have been used to illustrate the proportions of the different risk factors for the attributable risk. In addition we derive standard errors for the attributable risk based on the Taylor series expansion. Since the data of our cohort study was sampled with the same size in each 10 years age stratum which does not reflect the age-structure of the general population, the attributable risk and its standard errors are calculated using an approach that allows the weighting of the data according to population proportions of age. The formula for the standard errors has been evaluated using bootstrap-techniques.

RESULTS: We successfully implemented the method for the estimation of the attributable risk and its standard errors by integrating risk information using data of the HeartScore Germany and cross-sectional data emerging from the Gutenberg Health Study. The attributable risk can now be calculated without using the information of the overall disease rate. The bootstrap method shows, that the formula for the standard errors is useful.

CONCLUSION: Our method allows for the combination of different data sources in order to estimate attributable risks and our formula for the standard errors seems to yield a good approximation. But the validity of our method highly depends on the validity of the underlying data sources.