Deep learning can be used to guide treatment, predict albumin-bilirubin (ALBI) scores in the next 10 years, select the correct dose of ursodeoxycholic acid (UDCA) and prevent excessive use of different medications in elderly adults with primary biliary cholangitis (PBC), according to a recently published study in the medical journal Hepatology.
The ALBI score has recently been developed to evaluate liver function and assess prognosis in patients with liver cancer.
The study was conducted in Japan, a country with a rapidly aging population and subsequent high rates of elderly-onset PBC. Older adults, who often take six or more different medications, may struggle to take all their pills regularly, the researchers stated.
“Disease types are divided into hepatic failure type, portal hypertensive type and asymptomatic to slowly progressive type. Even in the slowly progressive type, the pattern of progression may not be linear due to treatment modification,” the researchers wrote.
Read more about PBC prognosis
The researchers evaluated the ability of artificial intelligence (AI) to predict future ALBI scores, reduce the number of different medications per patient, improve low medication adherence, and estimate the safety of UDCA dose reduction.
The study enrolled 140 PBC patients with at least 10 years of previous medical data. It incorporated liver biopsy results, disease stage classification, liver biomarkers including aspartate transaminase (AST), alanine transaminase (ALT), albumin, total bilirubin, the presence of enlarged veins in the esophagus, itchy skin, UDCA dosage at diagnosis and ALBI score after 10 years into a Deep Neural Network as supervised data. Data from 10 patients with disease onset at age 65 years or older and with 10 years of follow-up were validated as test data.
According to the analysis, 85% of patients were female, with a median age of 58 years. About 18% of them were classified as asymptomatic. Their median ALBI score was -2.95 at diagnosis, and the median was -2.85 at 10 years.
Machine learning results had an overall accuracy rate of over 90%. In patients with asymptomatic PBC detected by medical checkups or promising long-term prognosis, and in routine clinical practice, deep learning can be used to predict ALBI scores in about 80% of patients whose scores changed over 10 years. One-year data can be used as a decision-making tool for therapeutic interventions, the researchers stated.
