Dealing with Missing Values in a Clinical Case-Based Reasoning System

In clinical case-based reasoning systems, missing values in the case-base pose a common but serious problem that impairs the performance of the system. Imputation of missing values is popular in many knowledge-based systems. However, applications need to be adjusted to take into account that imputed values are only estimates of original values. In this paper, we introduce an imputation method that exploits the correlation between clinical data by filtering the case-base according to attributes correlated to the missing attribute. We also present a framework that shows how any imputation method can be used in a case-based reasoning system to account for the inherent uncertainty of imputation and to reflect the quality of the imputation method in the similarity calculation. The filter imputation method and the imputation framework are evaluated using a case-based reasoning decision support system developed for radiotherapy treatment planning for prostate cancer.