Attributable fraction estimation from complex sample survey data

Abstract Purpose A review of methods for the estimation of attributable fraction (AF) statistics from case-control, cross-sectional, or cohort data collected under a complex sample design. Provide guidance on practical methods of complex sample AF estimation and inference using contemporary software tools. Methods Statistical literature on AF estimation from complex samples for the period 1980 to 2014 is reviewed. A general approach based on weighted sum estimators of the AF and application of Jackknife repeated replication and Bootstrap resampling methods for estimating the variance of AF estimates is outlined and applied to an example analysis of risk factors for alcohol dependency. Results The literature lays the theoretical foundation to address the problem of AF estimation and inference from complex samples. To date, major statistical software packages do not provide a complete program but the approach is easily implemented using the modeling software and macro/function language capabilities available in major statistical analysis packages. In an example application, weighted sum estimation and inference for the population AF showed stable and consistent results under both Jackknife repeated replication and Bootstrap methods of variance estimation. Conclusions Future work on AF estimation for complex samples should focus on simulation studies and empirical testing to investigate the properties of the resampling variance estimation methods across a range of complex study design features and populations.