Probabilistically Populated Medical Record Templates:

Improving Clinical Documentation Time Using Patient Cooperation

Tristan Naumann, Marzyeh Ghassemi, Andreea Bodnari, Rohit Joshi

Computer Science and Artificial Intelligence Laboratory MIT, Cambridge, MA, USA

Abstract

The adoption of electronic medical records was a milestone in healthcare, but they have proven tedious to maintain. Increasingly, medical records contain information that is repeated within individual and even across multiple records. Such redundancy is frustrating and contributes to poor documentation as studies have shown that clinicians often copy-and-paste old, potentially out-of-date information into patients’ records. Moreover, it requires clinicians to spend additional time with documentation that could otherwise be spent with patients who often experience unreasonably long wait times. Therefore, we propose a solution to reduce the time clinicians must spend creating and editing medical records by generating patient-specific document templates from similarities found among patients in large medical data warehouses. Specifically, this work predicts the most likely fill-in-the-blank sentences to add to patients’ records based on information the patient provides through customized surveys during hospital wait time. In this way, our solution replaces patient idle time with relevant activity; it improve clinicians’ experiences with medical records and at the same time reduces errors caused by incorrect, outdated, copy-and-pasted data.

Paper

Slides

Interactive demo

GitHub Code

Code available at the following location.