Is LCHF associated with CVD?
Executive summary
* A study was published in March 2024, which claimed that “After 11.8 years, 9.8% of LCHF [low carb high fat] vs 4.3% of SD [standard diet] participants experienced a cardiac event.”
* The study was first presented at a conference in March 2023. This was the peer-reviewed paper (one year on).
* The study used UK Biobank, which has data for over 500,000 participants. From these, the researchers identified just 305 people who completed one “what did you eat yesterday?” dietary questionnaire at the start of the study, who weren’t on statins and were consuming <25% carbohydrate and >45% fat (i.e., lowER carb highER fat). Just 305!
* These 305 people were matched for sex and age only with 1,220 people who self-reported eating a standard diet.
* The 305 people averaged fewer than 1,500 calories compared to the SD group intake of almost 2,000. The entire study was founded on unreliable recall and thus no claims are robust. When only those who completed two dietary questionnaires were reviewed, there were no findings.
* Having matched for sex and age only, many characteristics differed between the LCHF and SD groups. The LCHF people were more likely to have non-white ethnic origin, three times more likely to have diabetes, far more likely to be obese and to have a higher BMI. The paper did not present any model that adjusted for all baseline differences. The claim that cardiac incidents were double was based purely on incidents in each group, not incidents in each group having adjusted for all differences between the two groups.
* At the end, I summarise the rebuttal to this paper if anyone tries to use it to claim keto or LCHF diets are even associated with cardiac events (let alone cause).
Introduction
I’ve had the latest keto bashing study on my desktop since it was published at the end of March 2024. I have tens of possible articles for Monday notes at any one time and the ones that get put on the desktop are the forerunners. What held me back from doing it straight away was the thought that I would be saying the same old stuff (association, not causation; relative not absolute risk; healthy person confounder etc) but this one turned out to be more than that. Plus, if the keto bashers continue keto bashing, I need to continue the rebuttals. I receive feedback from readers to say that articles against low-carb diets are regularly used to attack their way of eating. “Hey! Look at this – you’ll die of heart disease”, kind of thing. People want a rebuttal, so here it is…
The study was called “Association of a low-carbohydrate high-fat diet with plasma lipid levels and cardiovascular risk” (Ref 1). I thought that the lead author, Iulia Iatan, looked familiar. I searched my site and came up with this post from 2023 (Ref 2). Lo and behold, that post reviewed a March 2023 conference presentation about a “keto-like diet and heart disease.” The presenter was Iulia Iatan. I said in that note that it was unusual for me to cover a conference presentation, as I normally wait for the published paper. This recent publication is that published paper.
The March 2023 conference presentation
I was able to gather quite a lot of information about the March 2023 conference presentation from the press release (Ref 3) and from one slide on Twitter (Ref 4). The conference presentation claimed that a “keto-like” diet (up to 25% carb and over 45% fat intake) might be linked to a higher risk of cardiovascular events.
The claims were based on data from 70,684 participants in the UK Biobank study, for whom there was a one-time, self-reported, “what did you eat in the past 24 hours?” questionnaire. Only 305 people from the over 70,000 were identified as following a “keto-like” diet. That’s 0.43%. This tiny group of participants were matched for sex and age only with 1,200 other participants who self-reported that they followed a ‘standard diet’ (SD). The participants were mostly women (73%), and the groups’ average age was 54 years. The average (mean) follow-up time was 11.8 years.
The researchers claimed that those following the “keto-like” diet had higher LDL-cholesterol and more cardiac events. The terms “keto-like” and “low carb high fat (LCHF)” were used interchangeably. The main claim was “9.8% of participants on an LCHF diet experienced a new cardiac event, compared with 4.3% of those on a standard diet, a doubling of risk for those on an LCHF diet.”
Let’s now look at the recent paper to see what got through peer review one year after the conference presentation.
The paper
There is no doubt that the March 2024 Iatan et al paper is the follow-up to the March 2023 conference presentation. The information and numbers are identical. The study used UK Biobank data. From tens of thousands of participants, 305 people reported consuming an LCHF diet. These were matched (by sex and age) with 1,220 people who reported consuming a standard diet (SD) (Note 5). The participants were mostly women (73%), and the groups’ average age was 54 years. The main claim was the same “After 11.8 years, 9.8% of LCHF vs 4.3% of SD participants experienced a MACE [major adverse cardiovascular event].”
The study (as reported in the presentation and paper) looked at lipid profiles (cholesterol measurements) and major adverse cardiovascular events (MACE). I will ignore lipid profiles in this note for two reasons i) we have event data, so we don’t need markers and ii) there’s enough to go through with the event data. The MACE were defined as unstable angina, myocardial infarction, ischemic stroke, peripheral arterial disease, and coronary and carotid revascularization. (Revascularization, remember, is a subjective measure. It’s essentially having a stent fitted, which is a doctor-patient decision, and one which is more likely to happen in someone with higher LDL-Cholesterol. The notion that higher LDL-C is associated with cardiac events then becomes self-fulfilling).
The term “keto-like” diet, which was used throughout the conference press release, was not used in the peer-reviewed paper. The paper focused on the term LCHF. The LCHF diet was defined as <100 g/day and/or <25% total daily energy from carbohydrates/day and >45% total daily energy from fat. Participants on a standard diet (SD) were defined as those not meeting these criteria. In this study LCHF would more appropriately mean lowER carb highER fat, as it’s not the 5-10% carb, 15% protein and 75-80% fat of the LCHF world that I know. We’ll proceed on the basis that LCHF in this note means lowER carb highER fat.
The diet in the paper was definitely not ketogenic. The paper reported “Patients on a LCHF diet had significantly higher levels of ketones… but, as expected, below levels indicative of nutritional ketosis.” Why “as expected”? Because they knew it wasn’t a keto diet?
There was a large chunk of new data in the recent paper. The paper reported on more than the 305 LCHF and 1,220 SD people. The paper came with a Supplemental file. Supplemental Figure 1 contained a flow diagram of participants. This reported that 502,546 participants were recruited into the UK Biobank study between 2006 and 2010. Of these, 194,554 participants completed one or more 24-hour dietary survey at any stage following recruitment (not necessarily at the time of recruitment, which was odd). The flow diagram then excluded 30,104 people on lipid lowering drugs (statins etc). The next number in the flow chart was 2,034 – given as the number of people following a LCHF diet. These were then matched 4:1 with 8,136 people following a standard diet.
Most of the supplemental file provided data on the 2,034 and 8,136 people. It shouldn’t have done so. These people did not complete the dietary questionnaire at baseline and should not have been included. The 70,684 number, repeated many times in the press release from the 2023 conference presentation, was nowhere to be seen in the March 2024 paper. Some number revision had occurred, therefore.
The 305 and 1,220 numbers came from a sub set of people who had completed a 24-hour dietary survey at baseline. This subset therefore had the blood tests at the same time as the dietary survey. The paper and the conference presentation focused on these 305 and 1,220 people, which was correct. There should have been no mention of 2,034 and 8,136 people since they didn’t complete a dietary questionnaire at baseline and thus all notions of follow-up periods are invalid. There is also the certainty of reverse causation confounding – people would likely change their diet after an event. Hence did the lower carb higher fat diet cause an event or follow one? (The larger numbers would have been included to make the study seem bigger than it was.)
In reality, despite UK Biobank having over half a million participants, Iatan et al could only find 305 people following a lower carb higher fat diet who had completed a baseline 24-hour dietary questionnaire and who weren’t on statins etc. I reiterate – 305 people. Why go to such trouble to attack an invented diet (<25% carb & >45% fat) followed by so few people?
Supplemental Table 7 reported on 54 out of 305 people in the LCHF group who had completed two 24-hour dietary questionnaires – one at baseline and one at any other time. These were sex and age matched with 216 SD people who had also completed two diet questionnaires. There was no significant difference in cardiac events. i.e., when two surveys were completed, presumably adding accuracy to remembered consumption, there were no findings. That alone negates the study.
The questionnaire and diet differences
The diet was assessed using the Oxford WebQ 24-hour dietary recall questionnaire (Ref 6). This recorded self-reported consumption of 206 common food and 32 beverage items in the previous 24-hours. The questionnaire was completed by participants between April 2009 and September 2010.
I first attended a low carb conference in February 2015. This event, in Cape Town, South Africa, was billed as the first global LCHF conference. The dietary questionnaires for UK Biobank were undertaken 5-6 years before this conference. Searching PubMed for the term “LCHF diet” in the title or abstract of any paper before 2011 produced three results (Ref 7). The LCHF diet was not a thing back when dietary questions were being completed. Whatever the below 25% carb/above 45% fat diet was capturing, it wasn’t LCHF as we know it.
Table 2 in the main paper summarised the differences between the 305 people eating the lower carb higher fat (LCHF) diet and the 1,220 people eating the standard diet (SD). I’ve extracted some key numbers for the averages (mean):
You can spot the first issue at a glance. When asked about what they ate in the previous 24-hours, LCHF people ‘forgot’ approximately 500 calories. Which calories would they be more likely to forget? The ones that were part of meals? The meat and eggs etc? Or the chocolate, sweets, biscuits, cakes and savoury snacks? Dietary questionnaires are always unreliable – this one cannot be used as the foundation for a study or any claims.
The carb and fat intake were calculated as a percentage of energy intake. Forgetting approximately 500 calories negates those percentage calculations. If the LCHF calories were the same as the SD calories (1,992), LCHF people would only need to consume 45 more carbohydrate grams to be over the researcher-defined 25% carb intake. That’s one American cookie.
I also wondered what type of food/diet is characterised by under 25% carb and over 45% fat. Whole milk fails that test (it’s over 30% carb) (Ref 8). Oily fish is zero carb and can be 50% fat, so that could tick the box. Eggs would tick the box too. A typical sirloin steak doesn’t. It’s zero carb, but only 41% fat. Cheese is virtually carb free and approximately 74% fat, so that would qualify. But I don’t think meat, fish, eggs and cheese are what’s being measured here. They wouldn’t provide the average 23% carb intake. I checked for junk food and a Burger King double whopper is 23% carb and 51% fat. The BK bacon double cheese burger is 20% carb and 52% fat (Ref 9). Perfect examples of a diet averaging 23% carb and 52% fat, as in the table above.
The characteristics table
Table 1 in the paper was the usual characteristics table. This confirmed that the 305 LCHF people and 1,220 SD people were matched by age and sex. As reported above, the average age was 54 years and 73% of people studied were female. Matching by only age and sex left many other characteristics different.
Table 1 helpfully reported the p-values in the final column so that we can see at a glance which characteristics were significantly different. For ethnicity, there were fewer white people and more black people in the LCHF group. There were also other ethnic differences. There were many differences in baseline medical conditions. The LCHF group had 4.9% incidence of diabetes at baseline vs 1.7% in the SD group. That’s a three fold difference. BMI was significantly different – an average of 27.7 in the LCHF group and 26.7 in the SD group. Obesity was clinically and significantly different – 26.3% of the LCHF group were obese compared to 19.8% of the SD group.
Measures of affluence (income and education) were not significantly different. Hypertension, previous history of heart disease and family history of heart disease were not significantly different. Being a current smoker and physical activity were not significantly different (smoking looked clinically different to be honest – 10.6% of LCHF people were current smokers vs 7.7% of SD people).
Adjustment
When I reviewed the March 2023 conference presentation, the following slide (from twitter) was very useful. The data at the top of the slide reported that there were 53 events among 1,220 people on the SD and 30 events among 305 people on the LCHF diet. That gave incident rates of 4.3% for the SD and 9.8% for the LCHF diet. That’s where the main claim came from. I noted at the time that the slide lacked clarity in terms of adjustment for baseline differences.
Given that the exact same data were presented in Figure 2 in the peer-reviewed paper (replicated below), I sought help to understand it. I was fortunate enough to be put in touch with Professor Adrian Soto-Mota who is a medical doctor and a PhD doctor (Oxford) and a statistics expert (Ref 10).
Adrian explained that you look at Part B of Figure 2 first. The bottom image presents Kaplan-Meier curves. These report the time to a cardiac event during the 11.8 years of follow-up for the LCHF and SD groups. It hasn’t adjusted for anything – it just asks the question – when did an event happen and was it in the LCHF or SD group. This is supposed to signal whether there’s a difference worth reviewing further. The bottom diagram suggests that there is.
The top section (A) then uses Cox regression modelling to add in another factor. The first number (2.18) is the risk ratio (HR) of a cardiac event if on a LCHF diet. The diabetes number (3.37) is the risk ratio (HR) of a cardiac event if on a LCHF diet and with diabetes. That’s not enough to remove bias from the large baseline differences in diabetes. It’s confirming in effect that the difference in diabetes at baseline would make a difference to cardiac events. Ditto, current smoking (2.44) has estimated the risk of being a current smoker and being on an LCHF diet.
Population studies typically produce models that adjust for different characteristics. Model 1 might adjust for sex and age only. Model 3 might adjust for every difference in baseline characteristics from energy intake to family history of conditions. This study has not produced models adjusted for all baseline differences. Rather it has confirmed that those differences mattered.
Rebuttal
This is what you say to anyone who waves this study at you…
“There are over 500,000 participants in the UK Biobank study. This claim is based on 305 people who tried to remember what they ate yesterday sometime between April 2009 to September 2010 when the LCHF diet was not even a thing. Those 305 people forgot approximately a quarter of their intake. This didn’t stop the researchers using the ‘oops I forgot’ intake to calculate carb and fat proportions. Based on this dodgy calculation, the researchers defined LCHF as those consuming <25% carb and >45% fat, which is not keto or LCHF, let alone how I eat. Hence this study has naff all to do with me and it has naff all to do with 99.9% of people in the UK Biobank study. Oh and the 305 people had a number of factors that increase the risk of cardiac events that weren’t adjusted for.”
If that doesn’t shut them up, then try…
“… How about you eat how you like and I eat how I like?!”
Disclaimer – I don’t follow a ketogenic diet 😉
References
Ref 1: Iatan et al. Association of a Low-Carbohydrate High-Fat Diet With Plasma Lipid Levels and Cardiovascular Risk. JACC: Advances. 2024. https://www.sciencedirect.com/science/article/pii/S2772963X24001066?
Ref 2: https://www.zoeharcombe.com/2023/07/keto-diet-heart-disease/
Ref 3: https://www.acc.org/About-ACC/Press-Releases/2023/03/05/15/07/Keto-Like-Diet-May-Be-Linked-to-Higher-Risk
Ref 4: https://x.com/safchat/status/1632464437352226817
Note 5: No information about the matching was given. The supplemental flow diagram reported that there were 52,744 people who followed a SD, who completed a questionnaire at baseline and were not on lipid lowering meds. Each LCHF person was matched with 4 SD people. Why? Why 1:4? This meant that 1,220 of the 52,744 people were chosen to match the age and sex of the 305 people. How many of the 52,744 matched age and sex? How were the 1,220 chosen from that group? Randomly? The word random/randomized was not used in the paper or supplemental file. Were people chosen because they resulted in a significant difference? Forgive me for not trusting those who are against a particular diet.
Ref 6: https://biobank.ctsu.ox.ac.uk/crystal/crystal/docs/DietWebQ.pdf
Ref 7: https://pubmed.ncbi.nlm.nih.gov/?term=lchf+diet+%5BTIAB%5D&filter=years.2007-2011&sort=date
Ref 8: https://www.zoeharcombe.com/nutrition-data/milk-3-5-fat-nutrition-data/ (See other foods on the right hand side of this page).
Ref 9: https://www.burgerking.co.uk/nutrition-explorer
Ref 10: https://x.com/adriansotomota
