AI offers new hope for unexplained gut symptoms

They call it ‘diagnosis by exclusion’, a label given when doctors, after multiple tests, cannot find a measurable cause for a patient’s symptoms.

Irritable bowel syndrome is one of the most common conditions to fall into that category. For patients, it can be frustrating and inconclusive. For doctors, it reflects the limits of current diagnostic tools.

This matters. Approximately one in seven New Zealanders are thought to suffer from IBS; women are more than twice as likely to be diagnosed with the condition than men. Symptoms can be debilitating – pain, bloating, nausea, diarrhoea and/or constipation.

Yet blood tests, stool samples and even colonoscopies often come back normal. 

New research could make a difference. Led by the University of Auckland’s Bioengineering Institute, published in top-tier academic journal Gut Microbes, and involving a multi-disciplinary team of academics and clinicians, the study suggests there is more complexity within irritable bowel syndrome than previously understood. 

Applying AI to an existing patient dataset, researchers identified meaningful differences between groups of people diagnosed with the condition – differences which were previously invisible. The research identified at least eight distinct clusters of people.

“The crux of what we’ve found is that we have these big groups of patients we treat as having the same condition, but they don’t,” says lead researcher Dr Jarrah Dowrick.

“Imagine your car doesn’t start and the diagnosis is ‘you have a bad car’. It’s overly reductive. It could be a dead battery, bad starter motor, fuel problems, an electrical fault, or any number of other causes. In the same way, irritable bowel syndrome likely encompasses multiple different conditions.”

For patients, the consequences of that oversimplification can be frustrating and exhausting. After an IBS diagnosis, treatment often involves a long process of trial and error, including dietary changes, mental health interventions, and medications aimed at managing symptoms rather than underlying causes.

In the worst cases, when nothing seems to work, patients are told their symptoms are all in their head.

The Auckland Bioengineering Institute-led study is not the first to raise questions about the diverse nature – of irritable bowel syndrome.

A paper from the Mayo Clinic in the US, published in mid-2025 suggested the persistent lack of precise biomarkers might one day require clinicians to rethink or even ‘undiagnose’ irritable bowel syndrome.

The New Zealand research adds a level of detail that could help doctors narrow down what might work for a particular person, says the Auckland Bioengineering Institute’s Dr Tim Angeli-Gordon, a co-author of the paper.

“We’ve been able to better define irritable bowel syndrome; to narrow down the diagnosis so there will hopefully be less trial and error in treatments.”

One of the study’s key findings is a distinction between some groups of patients whose gastrointestinal symptoms are closely linked with their mental health, and some where the connection appears minimal.

“We’ve identified groups that are predominantly brain-centric and some that are predominantly gut-centric,” Angeli-Gordon says. “Traditionally, patients with IBS would often be treated the same.”

That helps explain why certain treatments work for some patients but not for others. Therapies aimed at managing stress and anxiety may benefit one group, while doing little for another.

The power of AI in research and diagnosis

The study also highlights the growing role of artificial intelligence in medical discovery. The team used data collected during a study of people undergoing a colonoscopy in Christchurch between 2016 and 2019. About 40 percent of the 315 participants had been diagnosed with irritable bowel syndrome and 40 percent had no gastrointestinal symptoms. (The remainder had other diagnoses.)

“The foundation of this paper is finding the most appropriate AI tool – in this case cluster analysis – and then applying it to answer medical questions,” Dowrick says. “As machine learning becomes more common, we will see it used more to reveal hidden patterns in complex datasets.” 

That potential, the researchers say, comes with an important caveat. AI-led medical advances depend on having large, high quality datasets for algorithms to work on. Generating those datasets requires long-term thinking, multi-disciplinary teams of clinicians and researchers, buy-in from health officials, and sustained government funding. 

Collecting and analysing the data from the Christchurch colonoscopy study was a “monumental task”, Angeli-Gordon says, involving collaborators and colleagues from around New Zealand. Equally critical was “the invaluable contribution of the participants themselves, who agreed for their data to be used in this research”.

The dataset took almost a decade to gather and analyse.

And while more than 300 participants may sound substantial, it sits at the lower end of what machine learning tools typically need, Dowrick says. AI thrives on large, diverse pools of data.

“As machine learning becomes more powerful, we will see it used to gain deeper insights into more and more medical mysteries – but only if we also improve our ability to generate high-quality data,” he says.

“I’d love to see a world where collecting the sort of information from tests and measurements that AI (now and in the future) needs to work with is part of normal clinical practice.

"If we don’t invest in data generation projects now, our machine learning investments are a waste of time.”