‘Personalising stuff that doesn’t matter’: the trouble with the Zoe nutrition app | Nutrition

“Your body is unique, so is the food you need.” This is the central credo of personalised nutrition (PN), as professed by its leading UK advocate, the health science company Zoe. Since its launch in April 2022, 130,000 people have subscribed to the service – at one point it had a waiting list of 250,000 – which uses a pin prick blood test, stool sample and a wearable continuous glucose monitor (CGM) to suggest “smarter food choices for your body”.

Like other companies working in this space, Zoe has all the hallmarks of serious science. Its US equivalent Levels counts among its advisers many respected scientists, including Robert Lustig, famous for raising the alarm about the harms of refined carbohydrates such as sugar. Zoe is fronted by King’s College London scientist Tim Spector and claims to be “created with world-leading science”.

The problem with personalised nutrition industry is that this is still a young research field, and there is not yet good enough evidence across the field to believe that we have yet found worthwhile novel interventions that are more helpful than standard advice. Although a Food Standards Agency report stated last year that “glucose monitoring and gut microbiome analysis may prove to become more robust and actionable” it spoke for many experts when it concluded that “the benefits of PN seem somewhat marginal when compared to what is already understood about a healthy diet.”

One major issue is that the personalisation only goes so far, and leans heavily on a couple of key biomarkers. Take the use of CGMs. These allow wearers to see almost in real time their blood glucose fluctuations and especially their post-prandial peaks. Zoe’s theory has been summed up by Spector: “If you are having multiple spikes in a day, your mean glucose level will be raised. We know that raises your risk of diabetes and heart disease.” So if you can see which foods or meals produce the biggest spikes, the idea is that you can tailor your diet so as to make them flatter.

However, most scientists remain unconvinced that non-diabetic users of CGMs can glean useful health information from them. “Glucose in someone without diabetes is a minuscule part of your overall metabolic health, let alone overall health,” says Nicola Guess, an academic dietitian and researcher at Oxford University specialising in the dietary prevention and management of type 2 diabetes. “There is a lot of interpersonal variability and one person might have more and bigger peaks than another yet have the same average blood sugar level.” Doctors can accurately diagnose diabetes or pre-diabetes by a standard fasting or HbA1c blood test. In contrast, says Guess: “The data from a CGM has no such diagnostic value” – something that Zoe acknowledges.

Another problem is that personalised nutrition research bases a good deal of its findings on analysis of mountains of data collected by its users. This throws up lots of associations between diet, blood glucose levels, weight and so on. But these “cross-sectional” studies can only ever find associations, not causation. So the existence of an association between greater spikes and higher average blood glucose levels, even in healthy people, does not tell us anything about causation. The higher spikes may be a consequence of an underlying metabolic problem, not the cause of one. If that were the case, keeping the spikes down would be addressing a signal of a problem, not its cause.

Worse, in very large data sets, Guess explains, cross-sectional studies will inevitably generate false positives: associations that are statistically significant but in effect random, “like buying an iPhone on a Tuesday is associated with risk of Crohn’s”.

Given these scientific limitations, Shivani Misra, a diabetes researcher and consultant at Imperial College London, says she sees no evidence for the theory that healthy people should seek to flatten their blood glucose curves. She decries what she calls the “glucose-centricity” encouraged by CGMs, which she sees as “so unilaterally focused on one metric of metabolism” when there are “so many other inputs that we can’t capture”. “I think people are focusing on glucose as a marker simply because we have technology to measure it,” says Guess. Personalised nutrition often starts with what it can measure, not with what is most important for our health.

---

The usefulness of stool analysis is also questionable. Again, the basic premise behind the test is reasonable. Even James Kinross, a reader in colorectal surgery at Imperial College London, agrees that “the microbiome is highly individualised, and it is probably the most important determinant of our response to different disease risks or to different medicines.” However, along with many other experts, he believes that we do not yet know enough about what a healthy microbiome looks like. The best advice to nurture a healthy gut microbiome remains to eat plenty of whole foods, especially fibrous plants, and minimise your intake of broad-spectrum antibiotics.

Most importantly, there simply is no such thing as good and bad bacteria, period. A bacterium may do good in one person and bad in another. Take the example of Escherichia coli, found in most guts. This is a species with much variation. Jacques Ravel, professor of microbiology and immunology at the University of Maryland, explains: “There are some E coli that are going to give you major diarrhoea, and there are some that are essential for your wellbeing.” So a test for it lacks clinical validity, meaning that “there’s absolutely no clear way of saying how this maps on to health or ill-health”.

Moreover, Ravel has published a paper detailing several studies that question the accuracy of stool testing laboratories – some of which could not reliably identify the bacteria in the gut, with certain labs in a US study delivering different results for the same sample.

“My view is that Zoe is personalising stuff that doesn’t matter,” says Guess. “The things that kill people in the UK and globally are LDL cholesterol and blood pressure.” Data, she notes, that Zoe doesn’t measure.

A key stumbling block for personalised nutrition is that in the scientific health world, you can do cutting-edge research or you can offer well established advice, but it’s challenging to do both. Companies such as Zoe try to ride both horses at the same time. On the one hand, Zoe is a research project, in the constant process of analysing its users’ data and looking for new insights. On the other, it is already giving users advice based on its work in progress.

Sarah Berry, an associate professor at King’s College London and Zoe’s chief scientist, bites this bullet. Of Zoe’s science, she admits that “it’s fair to say it’s contentious and it’s contested”, but that’s because “anything that’s emerging is always more contentious”. Still, she justifies Zoe running “ahead of the curve” on the basis that “if we wait until we have however many RCTs [randomised control trials] and this causal link beyond doubt, I don’t think we’ll ever progress to a point of being able to give people actionable advice”.

Misra says she doesn’t buy this. “There are well designed studies that are gamechangers, that actually change outcomes for people in a compelling way and that are cost-effective and change policy. I can give you numerous examples.” One is the research into low-calorie diets, intended to push type 2 diabetes into remission. “That was a randomised controlled trial, a very high-impact study. Within three years of that finding, it’s now a national policy, and everyone can have access to the remission programme.”

Zoe also blurs another important distinction. Healthcare providers are subject to a number of onerous legal constraints. But Zoe currently operates as a wellness company, which, as Ravel says, runs “without the regulation that applies to clinical and medical operations”. Hence the disclaimer at the front of the Insights report, sent to everyone who completes Zoe’s two weeks of monitoring, which warns: “Your insights are not clinical test results … Before making any changes to your diet, please consult your physician.”

Yet the entire programme is designed to drive diet change, gamifying eating so that users seek to achieve a Zoe score of more than 75 out of 100 for their daily food intake. Its marketing is littered with health claims, with its homepage imploring people to “Eat for your body and health”, listing benefits such as “Improved gut health”, “Reach healthy weight” and “Improve overall health”.

---

When I asked Berry about this tension between very clearly offering advice yet denying any such thing, she said she would have to get back to me. Despite pressing for this, no such clarification was forthcoming. “Why aren’t the regulators more interested in this?” asks Kinross. “I do not, for the life of me, understand it.”

Yet Zoe claims that it is “scientifically proven to work”. This rests on the publication of the first peer-reviewed study of its programme this month. There were some positive but modest outcomes: an average weight loss of 2.46kg is significant but not very impressive after more than four months. However, there were no changes in various other biomarkers, including blood pressure, insulin, glucose and postprandial triglycerides.

More importantly, the study compared Zoe participants with a control group who were far from blind tested. They were simply given standard dietary advice and a helpline to call. It was entirely predictable that people who logged every meal with Zoe for 18 weeks would eat more healthily. Nor was the study group as a whole representative of the general population: 86% were women and their average body mass index was 34 when anything over 30 is considered to be obese.

Guess has already posted a blog detailing criticism of the study while Kinross says the trial looks “designed to create what I would call marketing science, which is just enough science that you can convince a layperson that this has got value”.

When challenged about the design of the experiment, Berry admitted “if we wanted to test the efficacy of purely the Zoe scores, then we would need to match the method of delivery”, so that both groups were using the same app. That would allow a study “to look at how the actual advice itself compares to the standard care advice delivered in the same way”. But although Berry says that would be good, the actual study was designed to to test “the efficacy of the Zoe programme” as a package, comparing it only to “standard care”. That seems a strange objective: if Zoe’s USP is the personalisation of the advice, why design a study that deliberately doesn’t test those elements?

In the meantime, consumers are actually paying personalised nutrition companies to have their bodies surveilled to an extent that Kinross finds “Orwellian”. Zoe requires a one-off payment of a penny short of £300 and a monthly subscription of £24.99. “People don’t understand the value of the data that they are paying to give away,” he says.

Such are the issues when personalised nutrition is a research project – and people are paying large sums to be its guinea pigs.