Obesity: Root Cause Analysis
How to evaluate diet hypotheses in an uncertain & complex environment
This post explains some of the ways in which I think about obesity, and why I rely more on some types of evidence than others.
Black Box Thinking: What Changed?
Black box analysis treats a system as an impenetrable mystery. It just looks at the inputs and outputs. I like this as a first step: let’s just establish the facts.
If we look at America, or even The World, something has clearly changed in the recent past:
People are obese and diabetic and sick in other ways at unprecedented rates.
The obesity rate in the U.S. has more than tripled since 1975:
Childhood obesity has quadrupled since we started recording it in the 1960-70s:
Diabetes has exploded world-wide since 1980, approximately quadrupling, growing in pretty much every region & country in the world:
If we were aliens visiting Earth in 1970 and now in 2024, we’d probably suspect something had radically changed.
The data is more sparse going further back in time, but it had actually already gotten way worse compared to 1900. Obesity is estimated to have been exceedingly rare in the 1800s, even among rich Westerners, as were diabetes (pretty much only Type 1 back then) and heart disease. The earliest data seems to be height & weight data from the Civil War, and obesity was at maybe 1% in military aged men.
Obesity, diabetes, heart disease: These are the “diseases of civilization.” Clearly, something about “civilization” is killing us. The question is just, which part?
Most people probably don’t want to give up antibiotics, digital watches, and cars. Do we all need to cosplay as stone-age hunter gatherers to be healthy? Would that even do it?
The Black Box view just tells us something is drastically wrong. It does not tell us what magical Factor X caused diabetes & obesity & all the rest.
Back Testing: What has to be true
Back testing is something commonly done with investment strategies: if I had invested my money into these stocks in 1980, how would my portfolio have performed? Would I have weathered the 2000 dot-com bust? The 2008 financial crisis? How would I have dealt with inflation in 2022?
Back testing does not prove a hypothesis (or investment strategy) correct. But for a hypothesis to be correct, it has to pass the back test. In short, a correct explanation must be able to explain the past.
In mathematical terms, this is easy to make sense of: given a set of data points in the past, there’s an infinite number of curves that we can fit to those points. They all look correct against our back testing scenarios.
But not all of those curves will correctly predict the future; in fact, most won’t. Yet any curve that correctly describes the scenario has to fit our past data points.
In terms of nutrition, this means that any hypothesis for the diabesity epidemic must explain the past reasonably well. Maybe not perfectly, because we don’t have perfect data for the past (or even the present), but roughly.
For any given Factor X causing the diabesity epidemic, the following has to be true:
It wasn’t an issue at all for millennia or millions of years pre roughly 1850-1900
It increased massively around 1900 (they were already calling an obesity epidemic back then, compared to before!)
It has increased massively since
It got even worse since 2000 - the U.S. obesity rate has DOUBLED from 2000-2024!
The 1970s seem harmless compared to us now
Almost all modern people, no matter the country, are getting way more diabese - even if they started at different base levels
Plenty of peoples living “ancestrally” (=pre modern industrial diet) were perfectly healthy with regard to these diseases of civilization, eating very different diets from all over the spectrum (from very-low carb to 95% sweet potato), so it’s probably not that everyone used to be low-carb or low-fat.
Genetics seems to play a role, to a degree: Asians seem to not get as obese, on average, but they are just as diabetic or more
Starving yourself via CICO seems to temporarily make these better, but does not seem to fix the root cause, and most people eventually relapse & all the problems come back
People can do stupid internet diets and magically lose hundreds of pounds & reverse their Type 2 diabetes effortlessly.
These stupid internet diets tend to have certain patterns in common, and that can probably help us find out what Factor X is.
Mechanisms: necessary but not sufficient
To truly understand something, you obviously need to understand the mechanism by which it works. “I guess you put gasoline in, and then it makes noises and the car goes forward” is not an answer you’d accept from your car mechanic. We know why “empty gas tank” or “seized cylinder” make the car not go, these are not “risk factors” that we magically hand-wave about.
On the other hand, the body is incredibly complex and there are millions of mechanisms and pathways.
We have now identified pathways that could make us believe pretty much anything.
That’s why I am interested in mechanisms mostly for hypothesis generation. If we find a gene that behaves a certain way in a certain context, that could lead us to perform a dietary experiment.
But I wouldn’t ever design my diet entirely based on mechanisms, because there could well be a context we don’t fully understand. Or there could be competing mechanisms, and we don’t know about the other side, or which one prevails at any given moment.
For example, there’s an idea going around in the keto world right now that, “during ketosis, polyunsaturated fatty acids are oxidized preferentially.”
What does that tell us? That PUFAs are totally fine on a ketogenic diet? I would harshly disagree with that. Maybe PUFAs are burned preferentially in the sense that they get a 5% preference, but eating Standard American Keto (bacon & nuts & soybean oil ranch dressing) quadruples your total PUFA intake.
Hence, I don’t take action just based on that mechanism: I still avoid PUFAs despite being in ketosis most of the time.
It’s Slightly Complicated
In all likelihood the answer isn’t quite as simple as “low-fat” or “low-carb” or “keto” or “eat only potatoes.”
If it was, we would’ve solved this 50 years ago, when people tried finding the root cause of diabesity in 1970. Which they did.
On the other hand, saying “it’s complicated” with a concerned face and waving your hands around isn’t particularly helpful, either.
Yes, there are an infinite number of changes in diet & lifestyle since 1970 and especially 1850. But it’s probably not a combination of ALL of them.
The best demonstration of this are the many anecdotes & small scale experiments of Stupid Internet Diets.
Anecdotes & n=small experiments are King
If all of the 1,000,000 chemicals introduced into the food supply since 1850 were at fault, then simply doing a potato diet or heavy cream diet wouldn’t lead people to easily lose a lot of fat.
I’ll be the first to admit that we don’t know what exactly about eating potatoes or heavy cream leads to effortless weight loss, but the change is definitely manageable and excludes the vast majority of random changes we’ve made in the last 200 years.
Eating potatoes does not remove any microplastics from your body. It doesn’t avoid whatever’s in the soil or water. It doesn’t replenish whatever used to be in the soil and is now missing. It doesn’t turn you into a farmer or manual laborer and it doesn’t change your genetics or epigenetics. It doesn’t remodel your (ruined?) fat cells. It doesn’t reduce air & water pollution and it doesn’t change the makeup of your kitchen & cook ware. It (presumably) doesn’t get you more sleep or reduce stress or EMF or blue light or screen time.
Unless there’s some crazy magic going on, the change is either brought about by something in the potatoes (or cream) or by cutting out something you were previously eating, and replacing it with potatoes (or cream).
Muh RCTs
We don’t have a lot of very good controlled trials, e.g. RCTs (randomized controlled trials).
These would be great to have; but they simply can’t be done for many hypotheses. There is no RCT proving that jumping out of an airplane will kill you.
There isn’t even one proving that smoking is bad for you. We couldn’t & wouldn’t do one: it would take decades, and knowingly (we think) harm a lot of people.
Similarly, we can’t do RCTs for most modern dietary hypotheses. We’d have to lock hundreds of identical twins in a lab for 80 years and see which ones get diabesity or heart disease.
Even that would likely not be enough: think how the food supply has changed in the last 80 years. Since we don’t even know which of the ubiquitous changes is responsible, it’s likely that we’d inadvertently change the twins’ diet and invalidate the results.
E.g. would we have taken care to use butter instead of seed oils, if that theory wasn’t even on our radar in *checks watch* 1944?
We could limit our RCTs to pretty small changes, but even then it’s difficult unless you lock people in a lab. Many current hypotheses work over decades, including the mainstream Lipid Hypothesis of heart disease. In fact, if you do keto and tell an LDL fan that your Coronary Artery Calcium (CAC) scan is still 0, he’ll roll his eyes and not take you seriously. After all, EVERYONE KNOWS that plenty of people have a 0 CAC and then drop like flies decades later.
Even the mainstream, accepted theories are not verified via RCTs. They’re just left over from a time when we had less strict standards of evidence, like that shed in your dad’s yard that would never be allowed to be put up today.
Not a Court of Law
Whatever the evil Factor X is that makes us all diabetic & obese, my standard of evidence is not “innocent until proven guilty beyond a reasonable doubt.”
That’s the standard used in a criminal court case. This isn’t even a civil suit (“preponderance of the evidence”).
This is us standing in the library with a dead body on the ground and a smoking gun next to it. We know someone did it. We don’t know who. But he’s still in the room with us.
The obvious standard of evidence is not “beyond a reasonable doubt,” it is to exclude as many suspects as possible and see if it gets better. If it does, we can then try to find out which one did it. This is the part of the murder mystery where we lock 3 guys in the library and then the lights go out and someone else got murdered - it couldn’t have been those 3.
Hence, cries for “more evidence” or “show me the studies” or “muh RCTs” sound silly and dishonest to me. There’s a dead body, we know someone did it.
Everyone’s a suspect until we’re safe and sound, i.e. not obese & diabetic. Save it for the judge, pal.
No Null Hypothesis
Especially hollow ring the attempts to discredit any new hypothesis (e.g. seed oils) in favor of the status quo.
The status quo has CLEARLY FAILED. If 75% overweight, 45% obesity, and 12% diabetes in this country haven’t proven that to you, then I don’t know what would convince you.
Whatever the solution is, it’s not the status quo.
We thus have no null hypothesis to fall back on. We have to eat something (cause even fasting doesn’t seem to solve anything for most people).
We have to make a decision in the face of a lot of uncertainty, and making bad decisions is very unhealthy and possibly irreversible.
Common narratives I don’t buy
There are a lot of “common knowledge” narratives to explain the diabesity epidemic, and I don’t buy any of them.
I do think it is truly still a mystery.
Food availability & people used to starve
A common one is that we have so much food available now, whereas people in the past were starving all the time cause woolly mammoths were difficult to hunt & harvests were poor.
This might’ve been true in the stone age, or maybe in 1100AD, but 1850?
It doesn’t explain 1970 either. Heck, our grandparents’ generation was pretty thin even in the late 1990s, and they sure as heck didn’t lack any food. How did obesity double since 2000?
You could maybe argue that food has gotten cheaper compared to what our grandparents ate. But if they were actually that hungry, wouldn’t they have just spent more of their money on food? Instead of riding streetcars & going to the opera & buying gramophones & all that other stuff grandparents did.
My grandparents’ fridge was stocked to the brim every time I visited. If they were starving, they hid it well.
People used to move way more
First, this isn’t necessarily true. Sure, I suppose some stone-age hunter gatherer hunting a mammoth would’ve moved a lot during the hunt. Throwing spears is a lot of work.
But the arctic explorer Stefansson, in his book The Fat of the Land, describes the eskimos he lived with as essentially sedentary for large parts of the winter. And, of course, pretty much free of modern disease.
It’s true that a lot less people work in agriculture now compared to 1900. But on the other hand, people also exercise more than ever before, and that doesn’t seem to help.
In addition, even if “moving more” was the cause: by what mechanism?
One often-suggested factor is that moving more caused people in the past to burn more calories. But that doesn’t seem to be true: as detailed by Herman Pontzer in his book Burn, African hunter-gatherers walking 10,000 steps a day to hunt game & climb trees and fetch honey don’t seem to burn any more calories than sedentary Americans.
Carolies are missing the point
And even if one of these were true, they miss the point: WHY are we eating more/moving less than in the past?
There’s clearly some regulatory system in our body (and that of every animal) making us hungry or satiated to regulate food intake and energy expenditure.
Your argument can’t be that food intake went up or energy expenditure went down.
You have to explain why this regulatory system stopped working. There are plenty of hypothesis here, but most of the common narratives don’t even attempt an explanation.
Why didn’t my grandpa spend all his income on additional cookies? Why did he buy vinyl records and rare books instead, if he was starving? Why wasn’t he feeling lethargic and starved, if his body was telling him to eat more all the time?
Causal vs. Root Cause
One final rant: in a sufficiently complex system, there’s a huge number of factors that are “causal” yet not “the cause” in a colloquial or actual sense.
I might be predisposed to think this way as a programmer, but this is just bread & butter stuff in debugging any medium-sized or larger software system.
Causality has a tree (or technically graph) structure. Let’s analyze the root-cause analysis of a somewhat famous ship sinking, killing over 190 passengers:
This is the causality tree established in a root cause analysis (also called “why-because analysis”) in this case. You might also recognize the “5 Whys” technique, usually attributed to Toyota or similar Japanese management techniques.
Any one node in this causality tree is “causal.” Here are just some examples:
A key officer (“Assistant Bosun”) was tired and fell asleep
The ballast tanks had insufficient capacity
High-capacity pumps are very expensive, and they saved on costs by using lower-capacity pumps
There was time pressure
People made assumptions
A key feature (the bow door) was not visible from the bridge, hence it being closed was incorrectly assumed
Certain safety features, which would’ve prevented the ship from flooding completely once water entered the bow door, were not implemented
Water flooded the ship
All of these, and many more, are “causal.” Had any of them not happened, the ship would not have sunk. But they did happen and the ship sunk.
These are all chained with what’s called “AND” in boolean logic:
A key officer was tired
AND pumps were expensive
AND the tanks were insufficient
AND the design was bad
AND the safety feature was missing
AND assumptions were made
AND water flooded the ship
If any of these factors had not happened, the ship wouldn’t have sunk.
But what is “the cause” here?
Did the ship sink because an officer was tired? Yes.
Did the ship sink because pumps are expensive? Yes.
Did the ship sink because a safety feature was missing? Yes.
These are all causal.
Even if we say “the ship sank because it was flooded by water” that’s not wrong, but it’s also not particularly helpful. Water flooding is the proximate cause of pretty much all ships sinking, just like “running out of money” is the proximate cause of most bankruptcies.
A proximate cause is a causal factor that’s very close (“proximate”) to the disaster happening (e.g. sinking of the ship), yet not the root cause.
The real “root cause” here is probably “design ships with better safety features” or “ensure handover protocol between crews” and “don’t overwork your crew to the point officers fall asleep.”
If we look at this disaster and come away with “don’t be tired” and “don’t let the ship get flooded” we haven’t learned anything, and we won’t prevent this from happening again. We’re being blinded by the proximate causes.
“Eating too many carolies” is the proximate cause of obesity, just like “water floods ship” is the proximate cause of sinking ships. But that doesn’t tell us, in the colloquial or actual sense, what “causes” obesity. We’re looking for the root cause.
What changed between 1850 and now? What changed between even 1970 and now?
Beware “causal,” as it’s usually used to sound impressive, but doesn’t mean very much.
Conclusion
We’re in a pretty challenging spot. To be healthy, we have to:
Make decisions in an environment of great uncertainty
With individual differences, some genetic
With incomplete historic & current data
With great urgency
Without the ability to fall back on a Null Hypothesis
About very complex biological systems with feedback loops and weird interactions
Not even knowing which factors are relevant and have to be kept track of/kept constant
With almost everyone affected, and thus no reliable control group (except for a handful of tribes world wide)
I hope this explains my thought process, and helps you evaluate various dietary hypotheses & solutions.
Merry Christmas everyone!
They actually have done an RCT of parachute use! https://www.bmj.com/content/363/bmj.k5094
They found no benefit to using a parachute. These findings might be because they tested the hypothesis by jumping from stationary planes on the ground.
incoming milk and cookies diet