| Clinical trials allow researchers to draw conclusions about the effectiveness of a treatment. However, the statistical analysis used to draw these conclusions will inevitably be complicated by the problem of attrition. Attrition, or patient dropout, is actually “so common that one should be highly suspicious of any clinical trial report that claims to have no missing data” (Murray 1998). When faced with missing data, many resort to ad hoc methods such as case-deletion or mean imputation. This can lead to biased results especially if the amount of missing data is high. Multiple imputation, on the other hand, provides the researcher with an approximate solution that can be generalized to a number of different data sets and statistical problems. Multiple imputation is known to be statistically valid when n is large. However, questions still remain to the validity of multiple imputation for small samples.; The purpose of this research is to investigate the performance of different multiple imputation methods when the sample sizes are small. Specifically, the problem of patient dropout in longitudinal clinical trials where the parameter of interest is the mean change from baseline to endpoint is explored. The relative effectiveness of five common multiple imputation methods and the Last Observation Carried Forward method is investigated through a simulation study designed to emulate data typically found in longitudinal clinical trials. Guidelines are presented as to how the imputation methods perform under the different data conditions. I have also developed a nonparametric multiple imputation procedure that incorporates even more information than those currently used in the literature and does so for less computational expense. |
