#610: Rock, Paper, Salmon – Errors in Interpreting Food Substitution Models
A single substitution model can produce wildly different health conclusions depending on what food it's compared to—yet many people treat these models as if they reveal absolute truths about which foods are 'good' or 'bad.' In this deep dive, Danny Lennon and Dr. Alan Flanagan expose how food substitution analyses in epidemiology are riddled with hidden assumptions: the unit of substitution (grams vs. calories), the background diet, and the actual intake levels of the foods being compared. They warn against the 'nutritional rock-paper-scissors' fallacy—where one food beats another in a model and is declared harmful, leading to absurd conclusions like 'zero fish consumption is optimal.' The episode reveals how a study claiming organic produce reduces cancer risk when replacing conventional produce fails to account for the massive energy and nutrient differences between, say, 100g of spinach and 100g of banana. Even more strikingly, they show how artificial sweetener studies flip from showing harm to showing benefit simply by modeling whether the sweetener actually displaces sugar calories. The real takeaway? Substitution models are useful only when grounded in realistic, comparable intakes and when we stop treating them as ranking contests and start seeing them as context-dependent tools. The key insight isn't just about methodology—it's about interpretation.
A substitution model's outcome depends entirely on the comparator food—replacing red meat with white meat shows different results than replacing it with green vegetables, even if both are 'healthy'.
Using weight (grams) as the unit of substitution introduces energy differences—100g of steak has more calories than 100g of spinach, biasing results even when total energy is adjusted.
If a food is consumed in very low amounts in a population (e.g., fish twice a week), modeling its replacement with a high-consumption food (e.g., legumes daily) distorts interpretation and leads to absurd conclusions.
A study showing plant protein beats fish protein in a substitution model does not mean fish is harmful—both may independently reduce risk; the result may just mean more legumes are beneficial.
Artificial sweetener studies show opposite results depending on whether they model energy displacement (correct) or just compare intake levels (flawed), proving that analytical intent shapes outcomes.
…and 3 more takeaways available in PodZeus
The Problem with 'Compared to What?'
The episode opens by revisiting the foundational concept of 'compared to what' in nutrition science, emphasizing that the health impact of a food depends entirely on what it's being replaced with. Without this context, absolute claims about foods being 'good' or 'bad' are meaningless.
How Substitution Models Work (and Fail)
The hosts explain the mechanics of food substitution models in epidemiology, contrasting them with controlled trials. They highlight how models adjust for total energy but introduce hidden assumptions when substituting foods with different energy and nutrient profiles.
The Hidden Assumptions: Non-Specified vs. Specified Substitution
“Depending on which one of those models was used, it changes what we're comparing it to. And so given the particular population that was looked at there, in that situation where the one egg per day is compared with those equal calories from the non-egg cholesterol containing foods, you end up getting it compared to things like butter and red meat and processed meats and so on.”
The Unit of Substitution: Grams vs. Calories
“If I said 100 grams of steak versus 100 grams of lentils, they're now introducing, they're different in their total energy content, they're different in their nutrient composition as well and as a result we may expect that those differences are relevant in influencing or as we might term it, biasing the actual model estimates that one might get from that.”
The Absurdity of Real-World Application
“The interpretation of an outcome like this would certainly not be, well 100 grams of conventional fruits and vegetables should be replaced with 100 grams of organic fruits and vegetables.”
“And to go from there, to leap to say, well, in this case that defaults to us saying you must replace all fish with, in this case, legumes in order to not be doing anything harmful to your health is an absurd position to get to and doesn't really get to the point of what this particular substitution analysis allows you to conclude.”
“So if I said 100 grams of steak versus 100 grams of lentils, they're now introducing, they're different in their total energy content, they're different in their nutrient composition as well and as a result we may expect that those differences are relevant in influencing or as we might term it, biasing the actual model estimates that one might get from that.”
“So the interpretation of an outcome like this would certainly not be, well 100 grams of conventional fruits and vegetables should be replaced with 100 grams of organic fruits and vegetables.”
Host
Guest
danny lennon
person
dr. alan flanagan
person
sigma nutrition radio
media
sigma nutrition premium
product
atbc study
other
nutrinet-sante cohort
other
deirdre tobias
person
daniel ibsen
person
walter willett
person
john steven piper
person
Start discovering podcast insights today
Start with a 7-day trial and explore a growing catalog of popular podcasts. No credit card required.
No credit card required • 7-day trial • Cancel anytime