Nutrition research produces findings that nobody wants. The ice cream paradox is one of them. It may surprise you to know that across multiple large prospective cohorts, regular ice cream consumption has been associated with a lower risk of type 2 diabetes and cardiovascular disease. Not neutral. Lower. This is a food built from sugar, cream, and calories, and it keeps outperforming the foods that were supposed to be protective. The finding has surfaced repeatedly since the early 2000s, in different datasets, under different investigators, and it has resisted the ordinary methods by which an inconvenient result gets explained away. So, what the heck is going on here?

What follows is an examination of why the finding exists, what the leading explanation gets right, and what it may be missing. It will also explain how someone interested in the underlying biology can capture plausible mechanisms while discarding the ingredients that make the original food a problem. Let’s dig in!

What the Data Actually Show

I need to get technical for a moment. If you just want the recipe, scroll down to the bottom. If you want to learn how it all works, keep reading.

Okay, the Health Professionals Follow-Up Study tracked more than forty thousand American men through the late 1980s and 1990s. Its published conclusion emphasized low-fat dairy, reporting that men consuming at least two servings of skim or low-fat milk daily had roughly a twenty-two percent lower risk of developing diabetes. However, buried in the same analysis was a finding of identical magnitude for men eating ice cream at least twice a week.

Of course, the pattern predates that study. Analyses of insulin-resistance syndrome in the early 2000s found that dairy-based desserts, a category overwhelmingly composed of ice cream, were associated with substantially reduced odds of developing the syndrome among overweight participants. In fact, a 2014 paper examining another twelve years of dietary data concluded that yogurt reduced diabetes risk while other dairy foods did not. Dariush Mozaffarian, a co-author on that paper, later stated plainly that the conclusion was not written accurately because ice cream was, in fact, associated. Kevin Klatt at Berkeley has observed that the ice cream association was more consistent across cohorts than the yogurt association that received the attention, an assessment Deirdre Tobias at Harvard shared.

Now, the most thorough treatment came in 2018, when Andres Ardisson Korat defended his doctoral dissertation on dairy and cardiometabolic outcomes at the Harvard T.H. Chan School of Public Health. Drawing on the Nurses’ Health Study, Nurses’ Health Study II, and the Health Professionals Follow-Up Study, he found, among other results, that people with diabetes consuming roughly half a cup of ice cream daily showed a meaningful reduction in cardiovascular risk. His committee, chaired by Frank Hu, sent him back repeatedly to find the error. Well, he didn’t find one.

More recent work continues to reproduce the conclusions. A 2024 systematic review and dose-response meta-analysis of prospective cohorts examining dairy and glycemic outcomes found ice cream intake linearly associated with lower prediabetes risk, with a pooled estimate of 0.85 (95% CI 0.73 to 0.99) and no measurable heterogeneity between studies. Other dairy categories in that analysis showed no statistically significant association with glycemic outcomes.

One important thing we need to understand about everything I’ve just said. The finding is not a fluke in one dataset. It replicates again and again.

The Explanation Everyone Reaches For

Reverse causation is usually the standard account, and yes, it deserves to be taken seriously. Largely because it is almost certainly doing real work here. That said…

The logic is straightforward nonetheless. A person who is already metabolically healthy, physically active, and free of any emerging glucose problem eats ice cream occasionally without concern. A person who has begun to notice weight gain, fatigue after meals, a rising fasting glucose, or a physician’s warning about prediabetes stops eating ice cream. That person then develops diabetes, because they were already developing it when they quit. The dessert did not protect anyone. It functioned as a marker of not yet being sick.

Related confounding compounds the problem. Ice cream intake is captured by food-frequency questionnaires, which depend on accurate recall of eating patterns across months or years. It’s usually not very reliable. Residual confounding by overall diet quality, socioeconomic position, physical activity, and general health consciousness survives statistical adjustment. And no randomized trial has ever demonstrated that ice cream protects against anything.

This account is probably correct for the most part. It should be stated first and stated clearly, because the alternative interpretations that follow are weaker claims. However, reverse causation is a complete explanation only if nothing in the food itself could plausibly contribute. That premise deserves examination, not assumption.

The Whey Hypothesis

Whey has genuine and well-documented metabolic effects. In fact, they run in exactly the direction the epidemiology points. This is a critical point to consider.

Whey constitutes roughly twenty percent of the protein in cow’s milk, with casein making up the remainder. When consumed before a carbohydrate-containing meal, whey substantially slows gastric emptying, stimulates secretion of glucagon-like peptide-1 and glucose-dependent insulinotropic polypeptide, raises cholecystokinin, and markedly reduces postprandial glucose excursion. This is one of the better-replicated findings in applied nutrition. But the plot thickens.

In a randomized crossover trial at Wolfson Medical Center in Israel, fifteen adults with well-controlled type 2 diabetes consumed fifty grams of whey in water thirty minutes before a deliberately provocative breakfast of white bread and sugary jelly. Postprandial glycemia actually fell by twenty-eight percent across the full three-hour window, with early insulin secretion rising roughly fifty-two percent. The investigators noted that the glucose reduction exceeded what had been observed with nateglinide and compared favorably with standard sulfonylurea therapy.

Later work established that the effect persists at considerably lower doses. In a single-blind randomized crossover study of eighteen adults with type 2 diabetes, fifteen grams of whey taken ten minutes before a mixed breakfast produced forty percent greater beta-cell function, a twenty-two percent reduction in postprandial insulin clearance, a sixteen percent reduction in glucose incremental area under the curve, and a nearly six-fold increase in GLP-1. Mechanistically, this means an efficient insulin profile was achieved without additional beta-cell stimulation, which matters quite a bit in a disease characterized by progressive beta-cell exhaustion.

Whey also delivers leucine at high density, driving muscle protein synthesis. Skeletal muscle is the primary site of insulin-mediated glucose disposal, so anything that preserves or builds it improves glycemic control through a second, slower route. Moreover, whey supplies cysteine, the rate-limiting substrate for glutathione synthesis. It contains lactoferrin, immunoglobulins, and alpha-lactalbumin, along with peptides with documented ACE-inhibitory activity relevant to blood pressure.

So the hypothesis is attractive. Ice cream contains dairy protein. Dairy protein contains whey. Whey blunts glucose excursion and improves beta-cell function. So, perhaps the paradox isn’t a paradox at all. But there is a problem.

Where the Whey Hypothesis Breaks

The problem is dose, and it’s not even close. A half-cup serving of vanilla ice cream weighs approximately sixty-six grams and contains roughly two and a third grams of total protein. So, if twenty percent of that is whey, the serving delivers slightly under half a gram. The trials demonstrating meaningful glycemic effects use fifteen grams at the low end and fifty to fifty-five grams at the high end, administered as a discrete preload ten to thirty minutes before the meal. Fine. Measured against the smallest effective dose in the literature, a bowl of ice cream falls well short by a factor of roughly thirty. Compared with the doses that produced the headline results, it falls short by more than a factor of one hundred.

I would also argue that timing compounds the mismatch. The preload effect depends on whey arriving ahead of the carbohydrate, so that gastric emptying is already slowed and incretin secretion already elevated when the glucose load appears. Ice cream delivers its trace whey simultaneously with roughly fifteen grams of sugar. The mechanism is likely not being triggered in the sequence that produces the effect.

Stated plainly: whey is a real mechanism operating at a real dose, and ice cream doesn’t contain that dose. Hence, the hypothesis as stated doesn’t survive the arithmetic. Something else has to be going on.

This matters because nutrition writing often tempts us to find a mechanism that flatters the conclusion and stop there. However, the mechanism has to survive the numbers, and this one doesn’t. What survives is something adjacent and, in many respects, more interesting.

The Stronger Version: Matrix and Fat

I’ll preface this by saying that there are a lot of people who are definitely more qualified to speak on this than I am. However, based on what I know, there seem to be about two lines of evidence that hold up better.

The first concerns the food matrix. Ice cream behaves, glycemically, less like a sugar delivery vehicle than its composition suggests. Published glycemic index values for standard ice cream cluster in the medium range rather than the high range, and the reason is structural. Fat slows gastric emptying. Protein contributes an incretin response. Much of the carbohydrate is lactose and sucrose rather than actual free glucose, both of which produce a flatter curve.

The product is consumed cold and slowly. Continuous glucose monitoring data aggregated from large commercial user cohorts point in the same direction, with premium full-fat ice cream producing glucose peaks nearly identical to standard versions and lower than soft-serve or gelato, both of which carry more sugar and less fat. Granted, that data is observational and self-selected rather than controlled, so it should be weighted as corroboration rather than evidence in its own right. However, the counterintuitive implication, consistent with the published glycemic index values, is that reducing the fat makes the glycemic response worse, because the fat was doing the buffering. And the only reason I say that is because we see clues of this in other foods.

But this is actually the same principle at work in the whey preload, operating at a dose that actually exists in the food. In other words, it’s not whey acting pharmacologically. The whole dairy matrix is moderating the absorption curve.

The second line of evidence is stronger still, because it sidesteps the recall problem entirely. Circulating pentadecanoic acid, heptadecanoic acid, and trans-palmitoleic acid function as objective biomarkers of dairy fat intake, produced by rumen bacteria and largely absent from non-fat dairy products. Well, a pooled analysis of sixteen prospective cohorts across twelve countries, comprising nearly sixty-four thousand participants and more than fifteen thousand incident diabetes cases, found inverse associations for all three. Specifically, after adjustment for adiposity and lipogenesis markers, the hazard ratio for pentadecanoic acid was 0.80, for heptadecanoic acid 0.65, and for the three combined 0.71.

If you think about it, the significance of biomarker data is that it doesn’t depend on anyone remembering what they ate. The food-frequency questionnaire critique, which is the strongest methodological objection to the ice cream finding, simply doesn’t apply. Granted, confounding by lifestyle remains possible, and mechanistic work in animal models has been mixed on whether these fatty acids are causal agents or merely faithful markers. But the observation that objectively measured dairy fat tracks with lower diabetes incidence across sixteen independent cohorts isn’t easy to dismiss, nor should it be.

As it turns out, a large Swedish cohort of more than twenty-six thousand adults, followed for a median of twenty-four years with over four and a half thousand incident cases, might prove the point. High intake of cream was associated with a hazard ratio of 0.77 for type 2 diabetes and butter with 0.82, while high intake of non-fermented milk was associated with an increased risk at 1.40 and cheese at 1.23. Fermented milk trended protective at 0.88, though its confidence interval crossed unity and the result probably shouldn’t be relied on. Either way, the cohort shows that dairy does NOT behave as a single category. The high-fat fraction and the fluid milk fraction move in opposite directions, which is a problem for any dietary guideline that treats dairy as one exposure. Of course, this whole thing is highly problematic for anyone suggesting that dairy is a devil!

What This Actually Suggests

Being objective on this is critical. Reverse causation is likely the dominant driver of the ice cream association specifically. People who quit ice cream are frequently people who have been told something concerning about their blood sugar.

However, underneath that, dairy fat appears to carry a genuine inverse association with diabetes risk that survives removing self-reported intake. Either way, the food matrix of full-fat frozen dairy blunts the glycemic response relative to what its sugar content would predict.

And separately from both, whey protein has demonstrated, replicated, dose-dependent effects on postprandial glucose, incretin secretion, and beta-cell function. That makes it a legitimate metabolic tool. The problem here is that it’s at doses ice cream doesn’t usually deliver.

Now, I should probably stress that none of this makes commercial ice cream a health food. The added sugar, the seed oils in cheaper formulations, the gums and emulsifiers, and the caloric density remain what they were. However, the interesting question is whether the plausible actives can be isolated from the liabilities and used as a tool.

They can!

Building the Version That Isolates the Mechanism

As with most things I write about, there is an angle to exploit. We just needed to understand what we are exploiting and why. So, here you go!

The formulation below prioritizes animal-sourced ingredients and eliminates added sugar, seed oils, and plant stabilizers. Heavy cream supplies the dairy fat fraction and the matrix effect. Egg yolks provide natural emulsification, which pure whey simply cannot do on its own. And finally, whey protein isolate supplies leucine-dense, rapidly absorbed protein at a dose within the range of the clinical literature, rather than thirty-fold below it.

This is the best part. The result is not a lower-calorie dessert. It’s a high-protein, very low-carbohydrate frozen food that delivers the components with mechanistic support and omits the components with none. Plus, it’s extremely tasty!

Ingredients (approximately 4 to 6 servings)

• 2 cups (480 ml) heavy whipping cream, grass-fed preferred
• 3 to 4 large egg yolks, pasture-raised preferred
• 60 to 90 g whey protein isolate, unflavored or clean vanilla or chocolate (2 to 3 scoops depending on brand density; isolate rather than concentrate for lower lactose). Bone broth protein is an acceptable substitute.
• Pinch of sea salt or Himalayan salt

Method

1. Whisk the egg yolks in a bowl until smooth.
2. Heat the cream in a saucepan over medium-low heat until it steams. Do not boil.
3. Temper the yolks. Slowly drizzle roughly half a cup of the hot cream into the yolks while whisking constantly, then return the mixture to the remaining cream.
4. Cook over medium-low heat, stirring continuously, until the mixture thickens enough to coat the back of a spoon, approximately 170 to 175°F (77 to 80°C). Remove from heat immediately.
5. Cool slightly, then whisk or blend in the whey protein isolate and salt until fully dissolved and smooth. Avoid vigorous, high-speed blending if you prefer a denser texture, since excess air incorporation lightens the final product.
6. Chill the base thoroughly, at least four hours and preferably overnight.

Freezing

Ice cream maker. Churn per manufacturer instructions to soft-serve consistency, then transfer to a container and freeze one to two hours for firmer scoops.

No-churn. Pour into a shallow freezer-safe container. Freeze thirty to forty-five minutes, then stir vigorously or re-blend briefly. Repeat 2 to 3 times over the first 2 to 3 hours to break up ice crystals. Freeze until firm.

Ninja CREAMi or equivalent. Freeze the base solid in the pint, then process on the light ice cream or ice cream setting. Re-spin with a splash of cream if the texture comes out crumbly.

Texture and Storage

You should know that high whey content raises the freezing-point-depression challenge, and whey alone freezes rock hard. Cream fat and egg yolks are what prevent that outcome, which is why neither can be reduced without consequence. Let the finished product sit at room temperature for five to ten minutes before scooping if it freezes too firm. Store in an airtight container, pressing parchment or plastic wrap directly against the surface, since air contact drives ice crystal formation.

Also, you can adjust whey upward for higher protein density or downward if the mixture turns chalky. I guess that strict zero-dairy variants are possible using rendered animal fats and egg yolks alone, though they abandon the whey component entirely. If you’re worried about that, you can use full-fat A2 milk. The flavor is mild and rich, with clean dairy character dominating. And by the way, fruit-sweetened chocolate chips work well for anyone wanting more.

What This Removes

No added sugar or high-fructose corn syrup. No vegetable oils. No artificial emulsifiers or plant gums. Carbohydrate load stays minimal, limited to residual lactose in the cream and whatever trace remains in the isolate. Protein arrives as complete animal protein rather than as an afterthought behind bulk fillers. Boom!

Closing

As I think about it, it seems to me that the ice cream paradox is most valuable as a lesson in how observational nutrition science works and how easily it is misread in both directions. I think that the people who cite it as evidence that dessert is protective have overread it. At the same time, the people who dismiss it entirely as a statistical artifact have underread it, because the dairy fat biomarker literature does not depend on the same assumptions and points in the same direction.

The finding shows that the components of a food behave differently from the food’s reputation. Fat buffers sugar absorption. Dairy protein triggers incretin secretion. The dairy fat fraction tracks inversely with diabetes across sixteen independent cohorts using objective measurement. And a food assembled from those components without the sugar is a different object entirely from the one in the freezer aisle.

Take the mechanism. Leave the formulation. That’s weaponization of the data!

If you enjoyed this article, you might also like my article titled The Dairy Devil – Consider This

---

References

Ma J, Stevens JE, Cukier K, et al. Effects of a protein preload on gastric emptying, glycemia, and gut hormones after a carbohydrate meal in diet-controlled type 2 diabetes. Diabetes Care. 2009;32(9):1600-1602. doi:10.2337/dc09-0723. https://diabetesjournals.org/care/article/32/9/1600/28666/

Jakubowicz D, Froy O, Ahren B, Boaz M, Landau Z, Bar-Dayan Y, Ganz T, Barnea M, Wainstein J. Incretin, insulinotropic and glucose-lowering effects of whey protein pre-load in type 2 diabetes: a randomised clinical trial. Diabetologia. 2014;57(9):1807-1811. doi:10.1007/s00125-014-3305-x. https://link.springer.com/article/10.1007/s00125-014-3305-x

Smith K, Taylor GS, Walker M, Brunsgaard LH, Bowden Davies KA, Stevenson EJ, West DJ. Pre-meal whey protein alters postprandial insulinemia by enhancing beta-cell function and reducing insulin clearance in T2D. J Clin Endocrinol Metab. 2023;108(8):e603-e612. doi:10.1210/clinem/dgad069. PMID: 36734166. https://academic.oup.com/jcem/article/108/8/e603/7025377

Chiang SW, Liu HW, Loh EW, Tam KW, Wang JY, Huang WL, Kuan YC. Whey protein supplementation improves postprandial glycemia in persons with type 2 diabetes mellitus: a systematic review and meta-analysis of randomized controlled trials. Nutrition Research. 2022. https://www.sciencedirect.com/science/article/abs/pii/S0271531722000392

Imamura F, Fretts A, Marklund M, et al. Fatty acid biomarkers of dairy fat consumption and incidence of type 2 diabetes: a pooled analysis of prospective cohort studies. PLoS Med. 2018;15(10):e1002670. doi:10.1371/journal.pmed.1002670. https://journals.plos.org/plosmedicine/article?id=10.1371/journal.pmed.1002670

Mozaffarian D, de Oliveira Otto MC, Lemaitre RN, et al. trans-Palmitoleic acid, other dairy fat biomarkers, and incident diabetes: the Multi-Ethnic Study of Atherosclerosis (MESA). Am J Clin Nutr. 2013;97(4):854-861. doi:10.3945/ajcn.112.045468. PMCID: PMC3607658. https://pmc.ncbi.nlm.nih.gov/articles/PMC3607658

Dairy intake in relation to prediabetes and continuous glycemic outcomes: a systematic review and dose-response meta-analysis of prospective cohort studies. Curr Dev Nutr. 2024. PMCID: PMC11570412. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11570412/

Dairy intake, plasma metabolome, and risk of type 2 diabetes in a population-based cohort. Am J Clin Nutr. 2025;121(5):1137-1148. PMCID: PMC12107495. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12107495/

Sellem L, Jackson KG, Paper L, Givens ID, Lovegrove JA. Can individual fatty acids be used as functional biomarkers of dairy fat consumption in relation to cardiometabolic health? A narrative review. Br J Nutr. 2022;128(12):2373-2386. doi:10.1017/S0007114522000289. https://www.cambridge.org/core/journals/british-journal-of-nutrition/article/CAF3688EC778B50F202C9018D8D33548