Personalised nutrition
Should you spank millions up the wall analysing the microbiome when you don't even know if measuring it has any value?
Zoe announced a few weeks ago that it’s having to reduce its staff headcount and spend in other areas.
I have heard they have been treating the staff that have been made redundant really kindly, apparently offering mental health support and allowing people to keep laptops. Credit to them and the way they are treating people who have sadly been made redundant is a lot better than how the same people would have been treated in academia.
I thought this would be an opportune time to lay out my views on this type of personalised nutrition start up, focussing specifically on the example of Zoe with the goal to be as constructive and balanced as possible.
(That doesn't mean that I don't think the science they're doing is very, very poor and just FYI I have had a banger ready to go since Nov 2023 when I thought they were finally gonna publish their trial):
Nutrition does need to be personalised!
It's worth starting with the fact that personalised nutrition is here to stay. In fact it's always been here. A clinical dietitian will base dietary recommendations on what a person is currently eating, what they like and what their goals are. This is personalisation!
You can also personalise (ie compared to giving generic dietary advice about everything) on an individual’s standard clinical risk factors. For example, if someone has high cholesterol, there are very specific things you can do with the diet to help lower it.
You might also personalise based on someone's activity levels. If you had a personalised programme worthy of the name (especially when you're aiming it at the “worried well” who probably do more physical activity than the average person), you definitely would want that program to be able to personalise based on someone's activity. And asking “how many minutes of exercise do you do that raises your heart rate” is inadequate for assessing this.
But what Zoe have done is personalise based on novel biological risk factors, which is wholly unproven even as a concept*.
Zoe went with sexy, not with clinically relevant
I think a significant problem that Zoe has had is that clinical judgment has not driven the data they have decided (and thus have based the personalisation on) to collect.
I think they've decided to collect novel, sexy data because it's novel and sexy. CGM! MICROBIOME! But in people without pre-diabetes or type 2 diabetes, glucose is a pretty/utterly irrelevant clinical marker to personalise diets on. And by now you know what I think about the microbiome.
I think if Zoe had focused their personalisation on the clinical risk factors which are the most relevant for the majority of adults their product could have been really valuable.
For example, there is some emerging evidence that the effects of saturated fat intake on LDL cholesterol levels may differ between people. So developing a personalised programme on this actually could be really valuable. If hypothetically you are not a person who has an increase in LDL cholesterol with saturated fat intake wouldn't it be wonderful if you didn't have to limit your saturated fat! HELLO LOADS OF CHEESE. Conversely, if you happen to be a person who has a large increase in LDL cholesterol following consumption of even modest amounts of saturated fat this is something you could really focus on and get the greatest bang-for-your-buck with dietary changes.
Likewise, riboflavin (vitamin B2) may only lower blood pressure in people with a specific genetic make-up.
So not only is there an existing (albeit small) evidence base for interpersonal differences in the response to saturated fat and riboflavin (as examples), the clinical impact of personalised diets to modulate cholesterol and blood pressure would actually be clinically useful!
A lack of true personalisation results in a one trick pony.
If people or a programme don't know how to personalise, all guidance will end up being just the same thing. And this is always a red flag. if someone's answer to whatever condition or goal you have is eat carnivore, that's a red flag. Likewise if someone suggests you go vegan for every known problem, from type 2 diabetes to bowel disorders to arthritis to cancer? Huge red flag. This is not how pathology, physiology or clinical dietetics works.
Anecdotally (from patients), Zoe seems to recommend “eat more plants” and “eat less refined carbs”. TO EVERYONE. Now, this is not necessarily terrible advice. In fact, for the vast majority of the British population, these two things would almost certainly improve people’s health! Potentially a lot! However:
For a person with high cholesterol, you’d also want the algorithm to recommend things like limiting saturated fat, or potentially introducing stanols or sterols into the diet.
For people with high blood pressure you could also recommend specific components of the well-evidenced DASH diet.
For people doing a lot of endurance exercise you’d want to up the protein, and focus on carbohydrate availability (quality of carb matters a lot less!).
For people over the age of 60, you’d also want to recommend ~1.2g protein per kg as opposed to 0.8g.
Etc.
This could be personalisation worth paying for.
Figure out what things you can personalise on that are actually effective
A critical line of R&D for these “big-data” personalised nutrition companies should be understanding which of the novel factors (microbiome, metabolomics, “glycaemic response to foods” etc) that they plan to personalise on actually have any effect (either in helping people change their behaviour and/or making the advice more biologically efficacious). Each of these factors which ultimately get fed into their machine learning algorithms will cost money, so they should make sure they base their personalisation on the “inputs” that have value!
For example asking people about how much weight-lifting they do or more specific questions about the duration and intensity of exercise could get really valuable information that you could personalise on, but it wouldn't cost much. Conversely, let's say you think you can personalise on some novel biological factor like metabolomics or the microbiome - this is going to cost a ton.
If I were running one of these personalised nutrition companies I would make sure these expensive algorithm inputs were actually necessary or effective.
But to date:
There is no evidence that Zoe’s algorithms are reproducible**. In other words, if you took the microbiome test on two consecutive days, would the results tell you the same thing? Or would it one day tell you that you need to eat more tomatoes, pine nuts, and Cabernet Sauvignon, but the next day cucumber and walnuts were your top recommendations and the Cab Sav was bad news?
There is no evidence that personalising based on the microbiome results in better clinical outcomes. (Zoe have a trial which should be out soon but it is not designed in a way that will answer this question).
But even if there were evidence for both these things, I still wonder about the value of basing a dietary programme on a single microbiome sample. Why? Because the microbiome is going to change in about a week or two. So if the diet recommend IS based on the microbiome, then surely you need to do another test AGAIN within two weeks?
All of this is a long-winded way of saying that I suspect Zoe are spanking tens of millions of pounds up the wall on a biological marker which is irrelevant.
Do customers or investors compare about the scientific legitimacy of a product?
A few people have said to me that I am missing the point when I am whinging about how bad Zoe’s science is. That as long as it sells, investors don’t care. I don’t know.
I think there’s always a concern when something looks sketchy. And I am no marketing genius but I can’t help but think that lots of people will look at Zoe’s ads*** - which promise you don’t have to cut out any foods or “diet” and you can still lose weight - and think it sounds too good to be true. And I don’t think they’ve helped themselves with their Marks & Spencer’s**** Gut Shot (£2 for a 150ml yoghurt) or their Smart Gut product (basically a bag of specially-chosen seeds for £42 per pack or £139 for a 4-month subscription*****).
Where do Zoe go now?
I was a bit bemused by a statement I saw that Zoe were aiming to be commissioned by the NHS. There's definitely a massive role for healthtech in helping scale up clinically effective dietary programs and making them more accessible for people who don't want to or can’t head to GP practice or a community centre. In fact, it's worth mentioning the digital version of the NHS national diabetes programme is actually just as effective as the in-person programme at helping people achieve weight loss and preventing type 2 diabetes.
If being commissioned is their goal, I think a problem they have is that they now have a programme which is very expensive, yet has no evidence that it’s clinically effective or would save the NHS any money. They report achieving about a 2-3kg weight loss in their trial (I should hope so given how intensive it is) but there are many programmes that achieve that for probably ~10% of the cost.
Let’s end on a positive
One of the hypotheses behind this type of high-fibre/whole food dietary programme which helps feed the microbiome and teaches long-lasting behaviour change is that potentially it could be better at sustaining long-term weight loss. Weight loss maintenance is the holy grail of dietary research – and indeed of obesity science.
It's still possible that as Zoe recruit more participants into their trials and follow them up long-term (>12 months) that they may find their app does something that no other dietary programme in the world has achieved to date.
Of course I am sceptical and my sense is that any diet which can help people get more whole foods etc would do the same thing. And again I reiterate I don't think that personalising based on the microbiome has any role in this. But in a way this is good news because potentially Zoe could just scrap the microbiome testing (SAVE MILLIONS HAHA) and create a pared-down Zoe programme. The issue then is what’s their USP and how can they stand out in a very crowded field.
Over to Tim et al.
Disclaimer/Conflicts of interest: I am a co-investigator on a Medical Research Council grant which is developing an app for women with gestational diabetes. I have also previously provided input (either as a consultant or as part of a research project) into Oviva’s digital programmes for people with diabetes. I have provided input as a consultant into a low-carb programme called Fixing Dad and also one for Babylon Health. And I am on the scientific advisory board of MyFitnessPal.
*If you want to read more about this I published a comment a few months ago: Big data and personalized nutrition: the key evidence gaps. It might be behind a pay wall but I'm happy to send a PDF if you let me know.
**I have no idea whether Zoe have even tested this internally. If not, they should prioritise this. If so, they should tell the world.
***If you use youtube or instagram and have not been unconscious for 2 years you will have seen a few.
****For American readers this is a bit like Trader Joes or Whole Foods.
*****Make up your own mix of whatever nuts, seeds, dried fruit or spices you like and it will do the same thing.
I would love to read your Big data and personalized nutrition: the key evidence gaps, but it is indeed behind a paywall. I would love the pdf - how can I go about it? My email is irina.mateies@gmail.com
Given the advances in genAI (especially the creation of healthcare LLMs) what about a nutrition solution that is essentially an AI deititian? An app that generates a meal plan or helps a user balance their meal plan. Then it can go on to education. Look at what Limbic AI has achieved for therapy in the NHS.
PS - your content is so refreshingly honest and needed!