Issues in estimation of risk curves against driver BAC levels

There are relatively few case-control studies that have generated risk curves against BAC (blood alcohol concentration) level. Reasons for this include the difficulty in obtaining samples of controls (non-crash-involved drivers on the road) representative of the same population of drivers as the cases (crash-involved drivers). This paper discusses how the design and analysis methods used in particular studies may affect resultant risk curves and recommends a methodology for obtaining adequate samples and risk estimates. Particular issues addressed include sample design, accommodation of the sample design by the analysis method, treatment of missing values (imputation methods) and possible biases.