Predicting relevant character spans in clinical patient notes for automated scoring
Objective: Patient notes are written by physicians following a patient’s visit or any other medical procedure. Accurate and complete medical notes ensure systematic documentation of a patient’s medical history, history of present illness, diagnoses, past and current medication, allergies, treatment and overall care. So accurate documentation of patient notes is very important for the patient’s medical treatment. Medical students in order to become a licensed physician, give a licensure exam which assesses their quality of patient note making. In this exam, each patient’s note is scored by a trained physician using a rubric tool. But this is a very time consuming process for the trained physicians who are evaluating these patient notes in this licensure exam. It requires significant time along with human and financial resources. To automate the evaluation process of this licensure exam, Deep learning methods for natural language processing (NLP) can be employed. In this work, we have used deep learning techniques for Natural Language processing to automatically identify character spans in a patient note that are relevant for assessment against rubric tool.
Dataset: https://www.kaggle.com/competitions/nbme-score-clinical-patient-notes/data