Developing a Prognostic Model Using Multiple Imputation and Bootstrap in the Presence of Missing Data

• Activity Number: 413
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section for Statistical Programmers and Analysts
• Author(s): Lie Chen*+ and Wansu Chen and Chun R. Chao and Lanfang Xu
• Companies: Kaiser Permanente and Kaiser Permanente and Kaiser Permanente and Kaiser Permanente
• Keywords: missing data ; multiple imputation ; bootstrapping ; prognostic model

Abstract

Missing data on prognostic variables is a particular challenging problem in the construction of prognostic model in small dataset. The object was to use multiple imputation (MI) and bootstrapping methods for developing a prognostic model and examine the model performance. Eighty HIV+ diffuse large B-cell lymphoma patients were included to identify prognostic tumor markers for survival. Tumor marker data were missing in the range of 0 and 22%. The combinations of MI and bootstrapping techniques were used for prognostic model development. Variables were selected based on the proportion of times appeared in the logistic model with backward elimination. The performance of prognostic models was assessed using area under the ROC curve (AUC) and integrated discrimination improvement (IDI). After combining MI and bootstrapping to adjust variation induced by sampling as well as by incomplete data, the AUC and IDI were increased compared to the AUC and IDI for the model selected based on the original data. The procedure of combining MI with bootstrapping for variable selection increased the power of an analysis, resulting in multivariable prognostic model with better prediction performanc