Most Fatty Liver Cases Are Caught Only After the Liver Starts Failing

About 1 in 3 adults worldwide now carry enough fat in their livers to put them at risk. Three-quarters of the people who develop the dangerous scarring form of the disease find out only after the organ starts failing. The fix for a treatable epidemic is not a new drug. It is a 30-second blood-score calculation most clinicians never run, and a new wave of AI tools designed to run it for them.

The condition, now usually called metabolic dysfunction-associated steatotic liver disease, was renamed from non-alcoholic fatty liver disease in 2023. More than a billion adults meet the diagnosis, according to hepatologists cited in a recent Wired feature. A healthy liver holds a small amount of fat; once fat exceeds roughly 5 to 10 percent of the organ's weight, inflammation and scarring can begin, and the disease can progress to cirrhosis, liver failure, or cancer.

The damage is reversible if it is caught before scarring reaches stage 3 or stage 4 fibrosis. Weight loss through diet and exercise cuts liver fat even when total body weight drops by only 5 to 10 percent. Cutting alcohol, drinking coffee, and GLP-1 drugs such as semaglutide all help. For patients who reach moderate-to-advanced scarring, resmetirom, approved by the FDA in 2024, is the first drug shown to resolve the disease rather than manage its symptoms.

Most patients never reach any of those tools because the screening pipeline is broken. The standard first-line test, the Fib-4 index, is a 0 to 6 score built from a patient's age and two routine liver enzymes. It costs almost nothing and is already sitting in the data of nearly every adult who has had a blood panel. Pair it with the Enhanced Liver Fibrosis (ELF) test, a blood assay for two scar-tissue proteins, and detection of advanced fibrosis improves roughly four-fold over standard care. In practice, the tests get ordered in a small minority of the cases where they are clearly indicated. "There are simple tools that work, but they are not being used," said Jonathan Dranoff, a hepatologist at Yale, in the Wired feature.

That is where the AI work fits. Three models published or presented in the past 18 months show what happens when routine hospital records are scanned for risk patterns a busy clinician would miss.

Researchers at Osaka Metropolitan University trained a model to read ordinary chest x-rays, the kind taken for unrelated reasons, and flagged fatty liver with 82 percent accuracy. The relevant point is that the imaging already existed; the model pulled a liver diagnosis out of data collected for something else.

Evido Health, a Danish company now working with Roche, took a different route. Its LiverPRO model uses age and nine routine blood biomarkers and outperformed Fib-4 across a cohort of more than 470,000 middle-aged people, according to data presented at the European Association for the Study of the Liver's 2024 congress. The result has not yet cleared peer review, but the scale is what makes it interesting: the tool is built to run on lab data a primary-care doctor already has.

A third model, ALADDIN, published in the American Journal of Gastroenterology, tackled the next problem: once you flag a patient as high risk, which of the new drugs should they get? ALADDIN picked resmetirom candidates more accurately than Fib-4 or the standard risk scores.

Diagnosis itself is part of the intervention. A Danish study covered in the Wired feature found that telling patients they had liver fibrosis sharply increased their adherence to diet and exercise programs.

For now, the practical lever sits with the reader. A Fib-4 score is a few routine numbers a primary-care visit can produce on request. The question to bring to a clinician is simple: given my age and liver-enzyme results, what is my Fib-4, and do I need an ELF test or a referral? "We have the tools to find this disease early," Jeffrey Lazarus told Wired. The AI work now in journals and conference halls is, at its core, a deployment question.