So, is low carb bad?

There has been controversy in the public health nutrition world for decades around the causal roles of diet in disease. I don’t think there is any debate about whether diet affects health. The question is what diet affects health the most, for example is low carb bad, or good? And what advice should we give to people about this? Is it worth giving people advice about what to eat? Would this result in them improving their diet and therefore improving their health? 

What we disagree on:

School 1: Dietary fats, especially saturated fats, are a major cause of chronic disease, especially heart disease

Advice to limit energy by fat, especially saturated animal fat, will result in better health. This is the approach taken through the food pyramid and the iterations of such advice in mainstream public health nutrition over the past several decades.

This enjoys some success in that randomised trials illustrate that improvements in diet quality around this template show improvements in health.

 

School 2: Blood glucose and insulin are key causal factors in the pathway to metabolic disease

A combination of insulin resistance induced by a variety of factors including stress, poor sleep, low fitness, highly refined ultra-processed food diet, vitamin D deficiency and more, combined with a high starch and high sugar diet, mean that we must hypersecrete insulin to maintain normal glucose. Eventually we lose control of maintaining normal blood glucose even with high amounts of insulin. This is called pre-diabetes, although it comes after decades of high insulin. This progresses to type 2 diabetes where eventually the pancreas is unable to maintain insulin production.

This school of thought has support in that all five of the features of metabolic syndrome – the underlying risk and causal factors for all chronic disease – are related to high glucose and insulin. HDL is reduced with sugar and increased with higher fat. Blood pressure is primarily a problem of fluid retention at the kind driven by high glucose and insulin, central adiposity is driven by high insulin and glucose, insulin resistance/high glucose is by definition a problem of high glucose and insulin, and high triglycerides are driven up with carbs and down with extra fat.

It’s also the case that when people are randomised to treatments where carbohydrate is restricted there is a net improvement in all metabolic risk factors and diabetes markers.

 

School 3: Macronutrients don’t have much to do with anything, it’s all about food quality

We should eat foods which are low in human interference. That is to say, foods that were recently alive, running, swimming, flying or growing are the template for a healthy diet. This approach is a good one, but really untested as yet in research trials. It’s also possible that this advice is great for preventing disease and promoting health but less useful for reversing actual disease.

We should eat foods which are low in human interference. That is to say, foods that were recently alive, running, swimming, flying or growing are the template for a healthy diet. 

 

How can we really unpack the truth?

The best way to understand whether diet affects your health is to take a group of people and randomise them to two roughly similar diets. There would be a difference between the diets which any difference between the groups could be attributed to. This is the power of randomisation.

There are two different ways of doing this. The first is to have people stay in a lab/hospital/accommodation environment and feed them exactly what you want to feed them. This has the advantage of knowing exactly that the treatment was and that they got it. People can’t sneak out and get a Big Mac and a coke. The trouble here is that this has very little resemblance to the actual world we live in. It doesn’t take into account how hard it is to adhere to this regime. In scientific terms it has high internal validity (it’s a well conducted study) and low external (real world) validity.

The second approach is to counsel people into what to do and support them out on the loose in the real world. This has the problem of not everyone in each treatment arm doing what you wanted them to do. But if dietary advice has any value then it must have some effect in the real world under these conditions. Perhaps the most famous of these studies was Gardner’s A to Z diet study (see Gardner etc. al below). This was a randomised controlled trial that compared the effectiveness of four popular diets, namely Atkins, Zone, Ornish and LEARN, on weight loss and related risk factors among overweight premenopausal women. The study found that all four diets were effective in promoting weight loss, but the Atkins diet showed the most significant weight loss compared to the other three diets after 12 months.

These trials are criticised because they do not provide long term evidence, or follow people to hard endpoints.

That is to say they follow people for a year or two at most and not until they have a heart attack, get cancer or die (this is what we mean by a hard end point….).

It just costs too much to conduct these trials for decades. And who would pay anyway? There are no drugs involved….

There is a way around this, in fact a third approach, which is to conduct “natural experiments”. These aren’t really experiments at all. These are when groups of people in the population for one reason or another choose to eat in a particular way. We follow these people (a cohort) for a period and see what happens to them. In the real world people die (mortality), they get diseases which affect their quality of life (morbidity), or they stay healthy. We are “prospecting” the data to see what factors are associated with health, sickness or even death. This is why these are called prospective studies.

An obvious advantage is cost, because we don’t intervene with people we just follow them up. A second is that we can get large samples which have more statistical power to detect small differences.

The obvious disadvantage is that we cannot infer causation. This is because these studies suffer from the problem of confounding. If these confounders are known they can at least in theory be statistically allowed for. For example richer people eat better so income needs to be taken into account. There are many confounders, especially the healthy user bias which can never be adjusted for, or unknown unknowns which make these studies problematic.

A counter argument is that when studies are big enough and effects large enough then plausibility and the sheer size of the difference is enough to infer causation. A great example is smoking cigarettes is associated with a 19x increase in lung cancer. Sure, poorer people who work in unsafe conditions smoke more, but these confounders are unlikely to be the cause of an effect this big. And given we will never do a clinical trial where we ask people to smoke then it’s a fair enough piece of evidence.

But when we see another study which shows that people who eat the most meat have a 6/1000 five year risk of bowel cancer and those who eat no meat have a 4/1000 risk then the effect is real but small, and the possibility of confounding is too much to consider this effect seriously.

All of this has come to a peak recently with a study just published “Fat Intake Modifies the Association between Restricted Carbohydrate Diets and Prevalent Cardiometabolic Diseases among Adults in the United States: National Health and Nutrition Examination Survey, 1999–2018”.

We’ve heard from students and medical practitioners who are concerned that the advice about low carb is potentially dangerous.

In the actual study (see Kowalski et al, 2023 below) they used a massive representative sample to look at cardiometabolic risk predicted by carb intake stratified by fat intake, and found small effects in favour of moderate not restricted carb intake, and an even smaller effect in favour of polyunsaturated fats.

 

My view is that this study says very little because:

1. Any effects are so small that they are hard to interpret in the context of confounding factors such as the health message that “grains are healthy” means that the healthiest people are likely to eat more grains not because the grains make them more healthy but because they were healthy in the first place. This is especially true after decades of low fat dietary advice which introduces a healthy user bias to that way of eating.

2. Even if it were true, then the claim that “keto and low carb diets” are dangerous and bad for metabolic health isn’t actually studied. That is to say the definition of low carb is most often 40-50% of energy from carbs. That’s moderate carb and similar to the Standard American Diet.

3. The cardio metabolic endpoints while interesting are not hard endpoints like disease, disability or death.

 

As well, there is evidence which is contrary to this:

1. When we do clinical trials the net effect of a low carb/keto diet on cardio metabolic risk is positive (see 2022 meta analysis Lei et al).

2. There are cohort studies which show lower carb diets result in lower mortality, no difference and increased at both the low and high carb ends. So the results are mixed and the effects small (see Hu at el 2022). I’d take the totality of the cohort studies to show no clear effect using this method.

3. Carbohydrate restriction can effectively reverse diabetes, the major risk for all metabolic diseases (see Athinarayanan et al 2019).

So for most of us low carb eating is safe and effective for better metabolic and brain health.

Are there cases when you might wish to avoid this or we don’t know the safety profile? A 2020 safety review by Wananabe et al lists what we do and don’t know and contraindications for low carb and keto. They are cautious about such diets for pregnancy and breast-feeding, concurrent use of SGLT-2 inhibitors and after invasive surgery. So in these cases it’s worth consulting an expert.

 

My view

The low human interference factor (whole foods) advice would be the generic advice and we can avoid getting into the details of macro debates (fat, carbs, protein). We will need this area to grow and get more research to understand exactly how this approach works and for who.

It’s likely that most people can eat more protein and benefit from that, especially older and highly active people.

When we are trying to resolve chronic disease then we need to consider insulin resistance and glucose levels in the blood as these are causal in chronic disease and therefore disease remissions. Restricting carbohydrates is an effective and safe approach to do this. It’s not the only way to get there, but it is one we understand and can implement in the real world.