External Validation of Nomograms in the Presence of Missing Data

• Activity Number: 245
• Type: Contributed
• Date/Time: Monday, August 1, 2011 : 2:00 PM to 3:50 PM
• Sponsor: ENAR
• Author(s): Sujata Patil*+
• Companies: Memorial Sloan-Kettering Cancer Center
• Keywords: nomograms ; predictive model ; data analysis ; simulation study

Abstract

Nomograms are graphical depictions of multivariate models. Clinicians use these statistical tools to provide a patient-specific overall probability of an outcome based on clinical and pathological characteristics. Nomograms are often built on single institutional datasets and internally validated by dividing the dataset into training and test groups or by bootstrapping. The concordance index (CI) is one statistic that describes the discriminative ability of the nomogram. External validation is critical before a nomogram is accepted as a useful predictive tool. Valuable opportunities to externally validate a nomogram are rare and often do not come to fruition because either external datasets do not contain a variable or there is limited information on several variables. One approach to deal with this issue is to use imputation. Here, the effect of imputation on the magnitude and variability of the CI calculated in an external test dataset is assessed. The results from simulation studies done to identify the effect of imputation on the CI under several scenarios will be reported. Suggestions on useful characteristics of an externally validating dataset will be discussed.