MIT Researchers Introduce ‘MedKnowts’, A System That Combines Machine Learning And Human-Computer Interaction To Create A Better EHR

4163

Electronic health records (EHRs) have become widely used in the hopes of saving time and improving patient care quality. Physicians, however, typically spend more time navigating these systems than dealing with patients due to fragmented interfaces and time-consuming data entry procedures.

A research team from MIT and the Beth Israel Deaconess Medical Center combines machine learning and human-computer interaction to produce a better electronic health record (EHR). MedKnowts is a system that combines the activities of searching for medical records and documenting patient data into a single, interactive interface.

This “smart” EHR uses artificial intelligence to display tailored, patient-specific medical records when a physician needs them. To help doctors work more effectively, MedKnowts also provides autocomplete for clinical terminology and auto-populates fields with patient information.

The researchers had to think like doctors to build an EHR that would assist doctors. They devised a note-taking editor with a side panel that displays pertinent medical information from the patient’s past. This historical data is presented in the form of cards devoted to specific topics or concepts.

Suppose MedKnowts recognizes the clinical term “diabetes” in the text as a clinician type. In that case, the system will automatically display a “diabetes card” with drugs, lab values, and snippets from previous records pertinent to diabetes therapy. Only material related to the subject the clinician is writing about is displayed in MedKnowts.

https://arxiv.org/pdf/2109.11451.pdf

Chips, which are interactive text pieces, serve as links to related cards. Each chip is represented by a word or phrase that has been highlighted in a certain color based on the category it belongs to. The autocomplete system detects clinical phrases like drugs, lab data, and ailments when a clinician notes and converts them into chips.

Structured data about the patient’s illnesses, symptoms, and prescription usage is obtained with no further effort from the clinician using autocomplete.

The Covid-19 pandemic made the deployment more difficult. The researchers had been visiting the emergency department to acquire a sense of the process. Still, because of Covid-19, they could not stay in the hospital while the system was being implemented.

Despite the difficulties at first, MedKnowts grew in popularity among the scribes throughout the course of the month-long deployment. They gave the system an average usability value of 83.75 (out of 100).

According to the survey results, scribes considered the autocomplete tool to be particularly helpful in speeding up their work. The color-coded chips also made it easier for them to examine notes for pertinent information swiftly. Although the preliminary findings are encouraging, the researchers intend to proceed with caution as they examine the input and work on future MedKnowts iterations.

MedKnowts was created with an emergency department in mind, where doctors are most likely to visit patients for the first time. They also wish to take into account the needs of various medical users. A primary care physician with a deeper understanding of their patients would most likely have some additional qualifications.

If a doctor notices that a certain cardiology phrase is missing from MedKnowts, they can add it to a card, which will update the system for all users. The researchers hope to develop an adaptive system that clinicians may contribute to in the long run.

Paper: https://arxiv.org/abs/2109.11451

Source: https://news.mit.edu/2021/medknowts-electronic-health-record-0923