Automated search for patient records: classification of free-text medical reports using conditional random fields

Abstract

The purpose of this research is to develop a classifier to automatically identify qualified patients for breast cancer clinical trials from free-text medical reports. Specifically, we trained a conditional random field based classifier to identify patterns in sentences from free-text reports that are consistent with trial criteria. Experimental results indicate that this method works well, outperforming several competing techniques. Our work is important in two ways. First, it could be used to increase efficiency and decrease the cost of the patient enrollment process for clinical trials. Second, it could be extended and adapted to other patient identification problems.