AI Helps Identify Liver Cirrhosis Using Electronic Health Records

Cirrhosis, which is the end-stage of chronic liver disease and ranked as the 9th leading cause of death in 2021 by the Centers for Disease Control and Prevention, can result from various forms of liver damage and disease. Identifying patients who are likely to progress to cirrhosis has been difficult. However, early diagnosis could improve disease management. Artificial intelligence (AI) can be used to collect and analyze vast amounts of data, often from the electronic health record (EHR) containing the patient’s health history. While computers can easily interpret data entered into forms, it has been challenging to extract information from narrative text, such as clinician notes or discharge summaries. Previous attempts to extract information relied on keyword searches, which required input from a clinician familiar with the disease and multiple rounds of trial and error.

Researchers at the Medical University of South Carolina (MUSC, Charleston, SC, USA) have created a new AI method to automate the detection of liver cirrhosis by utilizing extensive data from EHRs. The AI model, called a convolutional neural network (CNN), mimics the neurons in the brain and was trained on EHRs of patients previously diagnosed with cirrhosis. By analyzing information embedded in narrative text and utilizing multiple layers of artificial neurons, the neural network can extract features and patterns to help identify cirrhosis.

After training on patient records manually reviewed to confirm cirrhosis diagnosis, the researchers applied their deep learning-based AI model that does not require prompts to a new set of health records. The model demonstrated exceptional success in identifying cirrhosis patients based solely on narrative text in clinician notes. Specifically, the trained CNN model achieved a precision rate of 97% when identifying cirrhosis patients using clinical text found in patient discharge summaries alone. While AI and machine learning have the potential to revolutionize the medical field, the researchers believe that these models are not meant to replace clinical judgment but rather to support and enhance it. Therefore, clinicians remain responsible for solving the case, with AI serving as a powerful tool to assist them, according to the researchers.

“The nice thing about using deep learning models is that the model learns from the examples you give it, without training it to look for certain words,” said Jihad Obeid, M.D., a professor in Biomedical Informatics at MUSC. “I think it's exciting that it was successful at identifying cirrhosis using just the text in the discharge summaries, as is the idea of taking it to the next level to see if we can apply it for earlier identification.”

Related Links:
MUSC