Addressing and advancing the problem of missing data.

Missing data can pose substantial risk of reaching incorrect conclusions from clinical studies. Imputation for the missing values is common, but can supply only an approximate result desired to be "close enough" to the intended true result. Prevention optimally addresses the issue. Knowledge of the effective techniques to minimize the problem, likely to vary with clinical setting, is presently inadequate. Formal evaluation of preventative methods should be encouraged and lead to publication of the assessments. Designers of clinical trials should also plan for study analysis where missing values occur. Simple imputation methods have been used and may be sufficient in some settings, but have potential to introduce bias and inaccuracy into the statistical analysis. More complex methods such as multiple imputation potentially offer reduced risk of bias. Multiple imputation also offers the potential for study designers to include some auxiliary outcome assessments that may substantially improve the quality of the imputation with limited added burden to the study. In all cases, sensitivity analyses examining the importance of the specific preferred method as compared to methods with different underlying assumptions is essential to assessing how adequately the missing data issue has been addressed.