Using Feature Predictive Power as a Guide for the Clinical Decision Making Process

Given the hectic nature of patient care environments, it is important for the clinician to reach the correct patient diagnosis as soon as possible. One of the challenges that clinicians face is to identify which patient data are salient to the identification of the correct patient status. In this paper, we propose an algorithm that was developed for incorporation in a decision-support tool for the novice nurse. The algorithm guides the patient data collection process using an iterative two-step process: (a) at each stage of the clinical decision process and based on the current partial patient information, it identifies the most probable diagnoses (b) it prompts the nurse to measure the clinical indicators with the highest predictive power of the aforementioned diagnoses. Results are presented as to the effectiveness of two predictive power measures, mutual information and the divergence measure, in guiding the data collection process.