Artificial intelligence gets thrown around so much now that it can start to sound meaningless. Every industry is supposedly being transformed by it. A lot of that is hype.

But in fatty liver disease, AI may actually be starting to matter in more practical ways.

For years, one of the hardest things to accept in this field has been how many people with fatty liver disease are still missed until the damage is already well underway. We already know a lot about who is at risk. We know the strong links with obesity, insulin resistance, type 2 diabetes, and metabolic dysfunction. We know millions of people are affected. And we know many patients still move through the healthcare system for years without anyone taking a closer look at their liver risk.

That is why the recent growth of AI-related work in MASLD and MASH is worth paying attention to. Not because AI is some miracle solution, but because it may help with one of the biggest practical failures in this space: an overloaded healthcare system that keeps missing too many people who are right in front of it.

One place AI is starting to show up is in MASH clinical trials. A recent development in this area was the FDA qualification of an AI-assisted pathology tool designed to help make liver biopsy readings more consistent in trials. That may sound technical, but the basic idea is simple. When specialists look at biopsy slides, there can be disagreement about how much inflammation, scarring, or liver damage is really there. Those differences matter. They can affect who gets into a trial, how sick a patient is judged to be, and whether a drug appears to be helping. If AI can make those readings more consistent, it could help make trials more reliable.

“This is the first AI-powered pathology tool to receive FDA qualification through the Drug Development Tool (DDT) Biomarker Qualification Program.”

PathAI, 2026

Even more interesting, at least from a patient and public health perspective, is the possibility that AI could help with earlier diagnosis and risk stratification outside of trials.

Researchers are looking at whether AI can help identify at-risk patients using data and imaging that health systems already generate. That could mean using AI to analyze existing CT scans for signs of steatosis or fibrosis instead of letting those clues go unnoticed. It could mean automatically calculating fibrosis scores from routine lab data. It could mean flagging high-risk patients with diabetes or metabolic dysfunction in electronic health records so that liver risk is not treated like an afterthought.

“AI is also being applied to imaging, helping detect incidental fatty liver on scans performed for other reasons.”

Adam Myer, MD, via University of Cincinnati, 2026

One recent example of that showed up in a Bloomberg report about an AI model developed to help with early detection of fatty liver disease using routine CT scans. Patients are already getting scans for all kinds of reasons. If those same scans can help flag possible liver disease earlier, that could help catch at least some of the people who currently slip through the cracks.

That kind of support matters because one of the biggest problems in fatty liver disease is not simply lack of awareness. It is lack of follow-through.

People with obvious metabolic risk factors still move through primary care without anyone calculating a fibrosis score like FIB-4. Incidental fatty liver findings still get buried in radiology reports. Patients with abnormal liver enzymes still get vague advice to lose weight and come back later. Many people do not see a specialist until fibrosis is already advanced enough to become frightening. As I wrote recently, the problem is no longer just diagnosis. It is implementation.

AI may be one way to help close that gap.

There is also growing interest in whether AI-based screening approaches could make economic sense at scale. That matters because healthcare systems do not just need tools that sound promising. They need tools that can be used in the real world without overwhelming clinicians, specialists, or budgets. If AI can help identify which patients are most likely to have advanced fibrosis and need more workup, it could help create smarter referral pathways instead of simply sending every at-risk patient straight to hepatology.

“MAOSS enhanced the detection rate of high-risk cases from 16.6% to 52.4%.”

Alibaba Cloud summary of Nature Communications paper, 2026

That point matters too. I have heard hepatologists say for years that if broad screening suddenly started working perfectly and every at-risk patient got referred directly to their clinics, the system would be overwhelmed in a hurry. That is not an argument against screening. It is an argument for better triage. If AI can help health systems do a better first pass, then it may become part of a more workable solution.

There are still good reasons to be cautious. AI is already one of the most overhyped ideas in healthcare. Bad data can lead to bad recommendations. Tools that perform well in one setting may not work as well in another. Poorly designed systems could even reinforce disparities instead of reducing them. There are also real privacy concerns whenever sensitive medical data is being collected, analyzed, and shared at scale. And no algorithm is going to fix a healthcare system that is too rushed, fragmented, or inconsistent to act on the information it already has.

Still, it is not hard to imagine where this could go. Routine lab work could automatically generate liver risk alerts. CT scans done for unrelated reasons could flag steatosis or fibrosis instead of burying those findings in the fine print. Electronic medical records could identify high-risk patients with diabetes, obesity, and metabolic dysfunction before advanced liver disease develops. Clinical trials could become more consistent through better pathology tools. Over time, that could help move liver disease care away from late discovery and toward earlier detection and smarter management.

That future is not here yet. But for the first time in a while, there are signs that AI in fatty liver disease may be moving beyond buzzwords and toward something more practical.

The real test will be simple. Not whether AI sounds impressive. Not whether companies issue flashy announcements. The real test is whether it helps the healthcare system stop missing people with fatty liver disease before it is too late.

Sources: PathAI on FDA-qualified AIM-MASH AI Assist; University of Cincinnati on AI advances in liver disease; Nature Communications on multimodal AI for opportunistic steatotic liver disease screening; Bloomberg on AI and early detection of fatty liver disease; Digestive Diseases and Sciences cost-effectiveness analysis.

Back in 2019, I wrote about the challenge of diagnosing NASH and the frustrating reality that tools like FibroScan, MRE, blood-based scores, and other non-invasive diagnostics were promising but still too disconnected from everyday care. At the time, the problem felt largely technological. We needed better tools, clearer answers, and a path away from relying so heavily on liver biopsy.

That is still true, but it is no longer the whole story.

Today, the more uncomfortable reality is that we do not just have a diagnostics problem. We have an implementation problem.

We already have more tools than we used to. FibroScan is no longer some exotic research device. Risk scores like FIB-4 are simple and inexpensive. Doctors know a lot more now about who is at risk. They have also spent years showing that fatty liver disease is closely tied to obesity, insulin resistance, prediabetes, type 2 diabetes, and broader metabolic dysfunction.

“This consensus report is a call to action to screen for liver fibrosis and risk stratify people with prediabetes or type 2 diabetes.”

American Diabetes Association consensus report, 2025

That is a very different mindset from the older era, when liver disease screening still felt like a side issue in routine metabolic care.

And yet far too many people are still being missed.

For years, the conversation around fatty liver disease sounded like the main problem was scientific uncertainty. We did not know enough. We did not have enough precision. We did not have enough ways to identify the right patients without putting a needle in their liver.

Now the problem looks different. It looks less like a lack of options and more like a lack of follow-through.

Patients with obvious metabolic risk factors still move through primary care without anyone calculating a fibrosis score. People with abnormal liver enzymes are still too often told to lose weight and come back later. Fatty liver findings on imaging still get buried in reports or treated like background noise. Many patients do not reach a hepatologist until fibrosis is already advanced enough to become frightening.

That is not because the medical community has no tools. It is because the tools are still not being used consistently where they matter most.

That matters because the burden is not small. The ADA consensus report notes that liver steatosis affects approximately two out of three people with type 2 diabetes. If a risk this common is still being treated like a specialty issue, then the system is not built around where the disease actually shows up.

I have heard versions of the same concern from hepatologists for years: if widespread screening suddenly started working the way it should, and every at-risk patient got referred straight to hepatology, the system would be overwhelmed almost immediately. That is not an argument against screening. It is an argument for building smarter front-line risk stratification, clearer referral pathways, and better tools in primary care so specialists are not forced to absorb the entire burden alone.

One of the biggest shifts in recent years is that fatty liver disease is finally being treated as part of the larger metabolic health crisis, not just a niche issue for hepatologists. If MASLD and MASH are closely tied to diabetes, obesity, insulin resistance, cardiovascular risk, and metabolic dysfunction, then earlier identification should be happening in the same places where those problems are already being managed.

That disconnect is dangerous because awareness without action is mostly theater. It is good to publish guidance. It is encouraging to talk about better diagnostics and treatment progress. But none of that means much for the person who is never identified in the first place.

To be fair, there are real reasons this has been hard. Primary care is overloaded. Many clinicians are juggling too many priorities. Reimbursement and workflow do not always reward deeper liver risk assessment. There is still uncertainty about exactly who should be screened, how often, and with which tools.

But those realities cannot keep serving as an excuse for inertia.

What should happen next is not mysterious. Checking liver risk needs to become a more normal part of caring for patients with metabolic dysfunction. That does not mean every primary care office suddenly becomes a liver specialty center. It does mean clinicians need practical ways to identify who needs more workup and who does not.

“More awareness about the health risks associated with MASLD and broad adoption of screening for liver fibrosis as a new standard of care hold promise for a future without cirrhosis…”

American Diabetes Association consensus report abstract, 2025

And those paths are clearer now than they used to be. Doctors now recommend starting with a simple first-line score like FIB-4, then following borderline or elevated results with a better second-line test such as liver stiffness measurement by transient elastography. The problem is not that no path exists. The problem is that it still is not routine enough.

One possible way to ease that burden is to use AI tools more intelligently in primary care. Not as a substitute for clinical judgment, and not as hype, but as a way to flag risk patterns, calculate fibrosis scores automatically, and make it harder for obvious warning signs to get missed. Emerging work on machine learning for advanced fibrosis screening in MASLD at the primary care level points in that direction. So does the broader case for AI-augmented stratification when current workflows depend too heavily on provider awareness and follow-through. If primary care is too overloaded to catch every case manually, then part of the answer may be building better systems that do more of that first-pass triage automatically.

That would not solve everything. But it would be better than pretending that an overloaded system will somehow start identifying patients earlier on its own.

I wrote years ago about the challenge of diagnosing NASH because it was obvious even then that delayed recognition was costing people precious time. What is striking now is how much the challenge has changed. We are no longer waiting for diagnostic science to begin. We are at a stage where many of the right pieces already exist, but they are still not being used consistently enough to protect patients.

That is the real test now. Not whether we can keep talking about MASLD and MASH, but whether the healthcare system will actually build screening and follow-up into routine care before it is too late.

Sources: American Diabetes Association consensus report; PubMed abstract; Diabetes Care screening study; Karger ML implementation study; AI-augmented stratification analysis.

Federal eyes are finally on NAFLD. It’s about time.

In May 2024, I traveled to Washington, D.C. to meet with AHRQ Director Robert Valdez and talk about the need for the federal government to take NAFLD seriously. The American Liver Foundation and many patient advocates had been pushing for that kind of attention for years, because you cannot address a disease affecting nearly 100 million Americans if you are not even measuring it properly. That mission has been central to why I started writing here in the first place.

That is why AHRQ’s newly released report, Inpatient Stays Related to Nonalcoholic Fatty Liver Disease, 2016–2022, is such an important milestone. It is the first federal report of its kind focused specifically on NAFLD, and while it looks at only one part of the picture—hospitalizations—it still tells us something important. The government is finally beginning to quantify the burden of this disease.

The headline number is striking: NAFLD-related inpatient stays increased 88.1% from 2016 to 2022, rising from 370,900 to 702,500 stays. That kind of increase should be setting off alarm bells, especially for a disease that remains so poorly understood outside liver and metabolic health circles.

But the most revealing finding may be that NAFLD usually showed up as a secondary diagnosis, not the main reason a patient was admitted. In fact, hospital stays where NAFLD was the principal diagnosis stayed essentially flat, between 5,300 and 5,500 cases per year. The big increase came from patients being hospitalized for other serious conditions while NAFLD sat in the background as a coexisting disease.

That is exactly why this illness has remained so easy to miss. NAFLD often hides in plain sight. It is frequently not the headline diagnosis, but part of the larger metabolic dysfunction that contributes to a patient’s declining health over time.

AHRQ also found that 93.4% of NAFLD-related stays had at least one comorbidity, compared with 78.2% of stays without NAFLD. The most common were hypertension, obesity, and diabetes with chronic complications. Patients with secondary NAFLD were also far more likely to be hospitalized with conditions like hepatitis, hepatic failure, pancreatic disorders, obesity, and other liver disease.

NAFLD often hides in plain sight. It is frequently not the headline diagnosis, but part of the larger metabolic dysfunction that contributes to a patient’s declining health over time.

None of that is surprising to those of us who have been advocating in this space. NAFLD does not exist in isolation. It sits at the center of the broader metabolic health crisis, intersecting with obesity, diabetes, cardiovascular risk, and liver-related complications in ways the healthcare system still does a poor job of recognizing early. I have written before about the stigma surrounding liver disease and about how metabolic health conversations are often distorted by denial, confusion, and bad assumptions.

The report also found that patients with NAFLD-related stays were more likely to be between 45 and 64 years old than patients without NAFLD. That matters because it reinforces that this is not some niche condition affecting only the very old or very sick. It is hitting people in the middle of their lives, often while they are still working, raising families, and assuming that fatigue, weight gain, or abnormal labs are someone else’s problem.

For me, this report matters not just because of what it says, but because of what it represents. For years, fatty liver disease has existed in a strange policy blind spot. The prevalence is enormous, the complications are serious, and yet the public-health response has lagged badly behind. We have had growing awareness among specialists, more patient advocacy, and finally the beginning of treatment progress, but federal attention has not kept pace.

This report does not solve that problem by itself, but it does mark a turning point. Once the federal government starts measuring something, it becomes much harder to pretend it does not exist.

The American Liver Foundation recently applauded the report and said a more comprehensive AHRQ study is expected later this year. That is exactly what needs to happen next. Hospitalization data is only one slice of the story. We also need better federal data on prevalence, diagnosis, treatment, complications, and cost.

The most important takeaway here is not just that NAFLD-related hospital stays are rising sharply. It is that the federal government has finally begun to put its eyes on a disease that has remained everywhere and nowhere at the same time. That is overdue, but it is also real progress, and for patients and advocates who have been pushing for years, it is a meaningful step in the right direction.