Acute cholangitis is a potentially life-threatening bacterial illness that is often associated with gallstones. Symptoms include fever, jaundice, right upper quadrant pain, and elevated liver enzymes, similar to those of a completely different illness: alcohol-associated hepatitis. This presents a challenge to emergency department staff and other health care professionals who must diagnose and treat patients with liver enzyme abnormalities and systemic inflammatory responses.
According to new Mayo Clinic research, machine-learning algorithms can help health care employees differentiate between the two conditions. In a paper published in Mayo Clinic Proceedings, researchers demonstrate how algorithms may be useful predictive tools when used with a few variables and routinely available structured clinical information.
[The purpose of the research was to] develop machine learning algorithms (MLAs) that can help distinguish patients with acute cholangitis (AC) or alcohol-associated hepatitis (AH) using simple laboratory variables. A study was conducted of 459 adult patients admitted to Mayo Clinic, Rochester, with AH (n265) or AC (n194) from January 1, 2010, to December 31, 2019.
Eight supervised MLAs (decision tree, naive Bayes, logistic regression, k-nearest neighbor, artificial neural networks, random forest, and gradient boosting) were trained and tested for AC vs AH in patients from the MIMIC-III database. A feature selection strategy was used to select the best 5-variable combination. 143 physicians took an online quiz to distinguish AC from AH using the same 10 laboratory variables alone.
MLAs demonstrated excellent accuracy up to 0.932 and AUC up to 0.980 in external validation. Feature selection in terms of information-theoretic measures was successful, and physicians performed worse, with a mean accuracy of 0.790.
Joseph Ahn, MD, a third-year gastroenterology and hepatology fellow at Mayo Clinic in Rochester, who is the first author of the research, believes that many medical providers in the emergency department or ICU struggle to distinguish acute cholangitis and alcohol-associated hepatitis, which are two very different illnesses that may present similarly. We trained machine-learning algorithms to distinguish the two conditions using routinely available lab values that all of these patients should have.
The researchers analyzed electronic health records of 459 patients older than 18 who were admitted to Mayo Clinic in Rochester between January 1, 2010, and December 31, 2019. Ten routinely available laboratory values were collected at the time of admission. After removal of patients whose data was inadequate, 260 patients with alcohol-associated hepatitis and 194 with acute cholangitis were removed. These data were then used to develop eight machine learning algorithms.
The algorithms also outperformed physicians who participated in an online survey, which is described in the article.
According to Ahn, machine-learning algorithms have the potential to aid in clinical decision-making in instances of uncertainty. In some situations, the inability to obtain a reliable history from patients with altered mental status or limited access to imaging modalities in underserved areas may force providers to make the determination based on a limited amount of objective data.
According to the study, if machine-learning algorithms are made easily accessible through an online calculator or smartphone app, they may assist health care workers who are needed urgently presented with an acutely ill patient with abnormal liver enzymes.
According to Ahn, this would lead to improved diagnostic accuracy and a decrease in the number of additional tests or inappropriate ordering of invasive procedures. Patients may delay the correct diagnosis or face the danger of unnecessary complications.
A team from the Hanyang University computer science department in Seoul, South Korea, was also involved in the research project.
Machine Learning Tool Advances Research on Rheumatoid and Osteoarthritis, Novel Machine Learning Tool Helps Discover Genetic Risk Factors for ALS, and Optimizing Biodrug Formulations with Machine Learning.