The Dairy-Inflammation Evidence Is Thinner Than Both Sides Say

Somewhere in the past decade, “dairy is inflammatory” stopped being a hypothesis and became received wisdom. It turns up in podcast monologues, in elimination-diet templates from online coaches, and in the confident advice of people who have never had a breath test or an allergy assessment. It is also one of the more checkable nutrition claims going, because inflammation in this context means something measurable in blood — and researchers have spent years feeding people dairy and measuring exactly those things. This article sticks to what those studies reported, who paid for them, and what they left unanswered.

What “inflammatory” refers to in this research

Two very different things share one word. Acute inflammation is what happens around a sprained ankle: swelling, heat, immune cells arriving on purpose. Chronic low-grade inflammation is the slow background version that nutrition research cares about, and it is tracked through markers circulating in blood. The systematic review by Labonté 2013 names the ones dairy trials report on: C-reactive protein, interleukin-6 and tumour necrosis factor-alpha. A later review of the same question took in gene expression in peripheral blood mononuclear cells alongside circulating biomarkers Ulven 2019.

Those markers are the currency of this literature, and they sit a long way upstream of anything a reader would notice in a mirror. A measurable shift in one of them at the end of a feeding trial is not the same experience as feeling better, and nobody should be sold either direction of that trade as if it were.

The wellness version of the claim is vaguer: bloating, puffiness, morning stiffness, skin that looks worse than last week. Those experiences are real, but they are not the variables the trials below measured. What follows answers the narrower question — when people were fed dairy and their blood was tested, which way did the numbers move?

What happened when trials fed people dairy

The most direct evidence comes from randomised controlled feeding trials, and it has been gathered up more than once. Labonté 2013 searched PubMed in April 2012 for randomised controlled trials in humans published in English, excluding studies with no low-dairy control arm, and ended up with eight trials, all of them conducted in overweight or obese adults. It ran no meta-analysis. Of those eight, one had change in the inflammatory profile as its primary outcome, and that trial showed dairy improving pro- and anti-inflammatory biomarker concentrations compared with the low-dairy control diet. Among the seven where inflammation was a secondary or undefined outcome, three showed improvement in C-reactive protein, interleukin-6 or tumour necrosis factor-alpha and four showed no effect Labonté 2013.

The conclusion those authors drew is narrower than the use it usually gets: dairy product consumption does not exert adverse effects on biomarkers of inflammation in overweight or obese adults. They then decline to go further, writing that several methodologic factors and limitations among existing studies do not allow differentiation between a beneficial or neutral impact of dairy products on inflammation, and calling for further studies specifically designed to assess inflammation-related outcomes Labonté 2013. An absence of harm in eight trials in one population is what that review supports. It is not a health claim.

A second synthesis, registered with PROSPERO and restricting its search to trials published between 1 January 2012 and 30 April 2018 with risk of bias assessed by Cochrane methodology, covered sixteen studies reported in fifteen articles — healthy adults plus adults who were overweight or obese or had metabolic syndrome or type 2 diabetes Ulven 2019. Its abstract reports that milk and dairy products showed no pro-inflammatory effect in either group, and that the majority of studies documented a significant anti-inflammatory effect, “although not all the articles were of high quality”. Read past the abstract and the paper is more cautious about itself: its discussion and conclusion say long-term dairy supplementation showed a weak anti-inflammatory effect in both populations, and that the evidence from acute and short-term interventions is “scarce and thus inconclusive” Ulven 2019. When this review is cited for reassurance, it is almost always the abstract being cited.

The third synthesis describes itself as a systematic review of 52 clinical trials investigating inflammatory markers in relation to dairy consumption Bordoni 2017. Rather than pooling effect sizes, its authors defined an inflammatory score to quantify the interaction, and that score came out significantly positive across the entire data set, which they read as anti-inflammatory activity in humans. Stratifying by product category associated low-fat products, high-fat products and fermented products alike with anti-inflammatory activity. Two things belong alongside it: the score is the authors’ own construction rather than a pooled estimate of the kind a meta-analysis produces, and the review itself flags a large gap in knowledge on the bioavailability of the bioactive nutrients involved, recommending that future research follow the fate of those nutrients along the gastrointestinal and metabolic axes Bordoni 2017.

On how much these three overlap: the 2013 review searched to April 2012 and the 2019 review from 1 January 2012 onward Labonté 2013 Ulven 2019. We are not in a position to say how many trials any of them share with the 52 scored in 2017, so we make no claim that they represent three independent counts of the evidence.

Who funded the reassurance

The 2019 review did not appear in an ordinary issue of its journal. It appeared in a supplement, and that supplement’s own introduction records where the money came from: financial support from EU H2020 project no. 734451, entitled “Program for the promotion of milk and dairy products within the framework of good dietary practices”, plus the aid of a grant from the Interprofessional Dairy Organization (INLAC) of Spain Gil 2019. That same introduction records that the supplement is based on a conference held in Madrid on 28 February 2018, coordinated by Professor Angel Gil of the University of Granada and Professor Rosa M Ortega of Complutense University Gil 2019 — and Gil is a co-author of the review itself Ulven 2019.

The review’s own notes also state that the sponsor had no role in the design of the included studies, in collection, analysis or interpretation of data, in the writing, or in the decision to publish, and that all four authors declared no conflicts of interest. The notes add that publication costs were defrayed in part by the payment of page charges, without naming who paid them Ulven 2019. All of that can be true and the arrangement can still be worth saying out loud: a review of dairy and inflammation was published in a dairy-organisation-funded supplement built from a conference co-organised by one of its own authors and part-funded by a milk-promotion programme.

The best-known single yogurt trial has a provenance note too. Its registry entry (NCT01686204) lists the University of Wisconsin–Madison as lead sponsor and the Dairy Research Institute as a collaborator, and records masking as none — participants and investigators knew who was eating yogurt and who was eating soya pudding Pei 2017.

That is what we can put on the table. We have not obtained and read the funding or conflict-of-interest statements for the 2013 and 2017 reviews or for the two lactose papers cited below; their full texts sat behind publisher paywalls when we checked. Readers should treat their absence from this section as a gap in our verification, not as a clean bill of health.

The subgroup that pointed the other way

The 2017 stratified analysis did not return one direction for everyone. As its abstract reports, when subjects were stratified according to health status the inflammatory score was strongly indicative of anti-inflammatory activity in subjects with metabolic disorders and of pro-inflammatory activity in subjects allergic to bovine milk Bordoni 2017. That is the place in the evidence cited here where the popular claim finds support, so it deserves this article’s tightest calibration rather than its loosest.

Precisely, then: that signal is a direction produced by the authors’ own inflammatory score within a stratum of a review, not a measured effect size from a pooled analysis, and it describes people already identified as allergic to cow’s milk rather than people who suspect dairy is bothering them. Read it as a direction, not a quantity.

One practical note, because nothing else in this article should be read as encouragement here. If you think you might be allergic to milk, that belongs with a clinician who can assess it — not with a glass of milk and a notebook.

Lactose malabsorption: the measurable problem

The dairy problem with a diagnostic test attached to it is not inflammation. A National Institutes of Health consensus development conference panel — a panel statement rather than a trial, prepared from a systematic literature review contracted with the Agency for Healthcare Research and Quality, public presentations and closed deliberations, and explicitly described as an independent report of the panel rather than a policy statement of the NIH or the US government — set out the distinctions clearly Suchy 2010.

In that statement, lactose intolerance is the syndrome of diarrhoea, abdominal pain, flatulence or bloating occurring after lactose ingestion, while lactose malabsorption is a decreased ability to digest lactose caused by a deficiency in levels of the enzyme lactase. The mechanism is digestive rather than immune: the symptoms result from bacterial fermentation of undigested lactose in the colon. Sometime after weaning, the panel writes, a genetically programmed decrease in lactase — lactase nonpersistence — occurs in most children worldwide. Malabsorption can be diagnosed by ingesting a standard dose of lactose after fasting and measuring breath hydrogen, with intestinal-biopsy lactase activity and genetic testing for the common polymorphism as alternatives. And, importantly, the demonstration of lactose malabsorption does not necessarily indicate that an individual will have symptoms Suchy 2010.

On how common any of this is, the panel is blunter than most people who cite it. It states that the prevalence of lactose intolerance in the United States cannot be estimated from available data, and that lactose intolerance is a real and important clinical syndrome but its true prevalence is not known. What it does report is variation: lactose intolerance, malabsorption and lactase nonpersistence vary across racial and ethnic groups, with the lowest reported occurrence in European Americans and higher, though variable, occurrence in African Americans, Hispanic Americans, Asian Americans and Native Americans; and in children younger than six, malabsorption was low in all the studies reviewed, peaking between 10 and 16 years of age Suchy 2010.

The panel’s central health worry is not inflammation at all. It is that many individuals with real or perceived lactose intolerance avoid dairy and ingest inadequate amounts of calcium and vitamin D, which may predispose them to decreased bone accrual, osteoporosis and other adverse health outcomes. Its recommendation follows directly: most persons do not need to eliminate dairy consumption completely Suchy 2010.

How much lactose actually gets tolerated

The number that surprises people is the tolerated dose. The panel concluded that adults and adolescents with diagnosed lactose malabsorption could ingest at least 12 g of lactose in a single dose — equivalent to the lactose content of one cup of milk — with no or minor symptoms, and that larger amounts are tolerated when taken with meals and distributed throughout the day. It also notes the other end of the scale: 50 g of lactose, roughly the content of a quart of milk, usually induced symptoms in adults with malabsorption when given as a single dose without meals Suchy 2010.

The systematic review of management strategies published alongside that statement examined 36 unique randomised studies and, because of variation in enrolment criteria, outcome reporting and the composition and dosing of the agents studied, could not pool results. Its headline finding was that moderate-quality evidence indicated 12 to 15 g of lactose — approximately one cup of milk — is well tolerated by most adults, and its conclusion was that most individuals with presumed lactose intolerance or malabsorption can tolerate that amount Shaukat 2010.

These two documents are not two independent votes. The panel statement records that the conference was informed by a systematic review conducted by the Minnesota Evidence-based Practice Center — which is that management review Suchy 2010 Shaukat 2010. One evidence base, reported twice.

On the workarounds people reach for, the management review is unflattering and precise about what it did and did not assess. It found the evidence insufficient that a lactose-reduced solution or milk containing 0 to 2 g of lactose, compared with more than 12 g, reduces symptoms, and insufficient for probiotics, for incremental lactose administration aimed at colonic adaptation, and for other agents; it called for additional studies to determine the effectiveness of lactose intolerance treatment Shaukat 2010. On lactase-treated products specifically, the panel’s verdict was that they may be tolerated better than nontreated products, but that more research is needed Suchy 2010. So the practical implication is not that malabsorption is imaginary. It is that clearing out the fridge is a larger intervention than these documents call for.

Why self-diagnosis misfires

The sharpest illustration of the gap between believing you react to dairy and reacting to dairy is an old, small, well-built trial. Researchers recruited 30 adults, mean age 29.4 years, who reported severe lactose intolerance and said they consistently had symptoms after ingesting less than 240 ml of milk. Everyone was screened with end-alveolar breath hydrogen after 15 g of lactose in water, which classified 21 of them as having lactose malabsorption and nine as able to absorb lactose normally — a detail that is the point of the study, not a footnote to it. In a randomised, double-blind crossover design each participant then took 240 ml of lactose-hydrolysed 2% milk with breakfast daily for a week, and 240 ml of ordinary 2% milk sweetened with aspartame to match the taste for another week Suarez 1995.

Symptoms were minimal on both. Mean severity scores for bloating, abdominal pain, diarrhoea and flatus fell between 0.1 and 1.2 on a scale where 1 indicated trivial and 2 indicated mild symptoms, with no statistically significant differences between the two periods in the severity of those four symptoms; dietary records indicated a high degree of compliance with no additional lactose sources reported Suarez 1995.

The null result has a boundary, and honest reporting has to include it. In the lactose-malabsorption subgroup, the mean difference in episodes of flatus per day was 2.5 with a standard error of 1.1 and a 95% confidence interval of 0.2 to 4.8 — an interval that excludes zero Suarez 1995. Real milk produced measurably more flatus in the people who could not fully digest it. What did not differ was how bad the four rated symptoms felt.

The authors’ own reading was that people who identify themselves as severely lactose-intolerant may mistakenly attribute a variety of abdominal symptoms to lactose intolerance, and that when lactose intake is limited to the equivalent of 240 ml of milk or less a day, symptoms are likely to be negligible and lactose-digestive aids unnecessary Suarez 1995. It enrolled 30 people, tested one modest dose taken with food, and dates from 1995. Within those limits, what it demonstrates is that unblinded self-assessment is a poor measuring instrument.

Yogurt, and how that story has moved

If dairy carried a pro-inflammatory signature, fermented dairy would be the awkward case. In a nine-week randomised parallel-arm feeding trial, premenopausal women in two body-mass-index strata (18.5–27 and 30–40 kg/m²) were assigned to 339 g of low-fat yogurt or 324 g of soya pudding daily, 30 per group across four groups. Both yogurt groups ended with a decreased ratio of tumour necrosis factor-alpha to soluble TNF receptor II; plasma IgM antibodies to endotoxin rose regardless of obesity status; soluble CD14 was unaffected by diet while the LBP/sCD14 ratio fell in both yogurt groups; 2-arachidonoylglycerol rose in the obese yogurt group only; and peripheral blood mononuclear cell expression of NF-κB inhibitor alpha and transforming growth factor beta-1 rose in the obese yogurt group relative to its control at nine weeks Pei 2017.

Three details keep that trial in proportion. The paper states plainly that other biomarkers were unchanged by diet, so the favourable movement is confined to a handful of the markers measured. Both obese groups gained about 0.9 kg over the study, and the obese yogurt group had 3.6% lower diastolic blood pressure at week three. And the authors’ own framing is that the anti-inflammatory mechanisms of low-fat dairy consumption are largely unknown, even as their conclusion uses the verb “reduced” Pei 2017. A companion analysis from the same group and cohort examined the postprandial window Pei 2018.

Then the same laboratory ran a different design and got a different answer. In a randomised controlled crossover trial with four-week arms separated by a four-week washout, comparing 12 oz per day of low-fat dairy yogurt with soya pudding, 39 healthy adult women with a body mass index between 25 and 40 kg/m² were enrolled and 20 completed. The primary outcome — the change in the proportion of circulating T-helper-17 cells — was unaffected by the treatments, and the secondary immune-function markers were described as maintained by the yogurt treatment. The title says yogurt “maintains” those markers; the 2017 title said “reduces” Hasegawa 2024. Anyone quoting the 2017 trial as the yogurt result should quote this one in the same breath.

That pair sits next to a broader, messier question about what fermented foods actually do to the gut, and a separate question again about whether supplementing bacteria in capsule form does anything measurable for active people. The narrow point here is only that yogurt did not behave like an inflammatory food in either trial.

What this evidence does not settle

Start with the size of the base. Everything above rests on eight trials in overweight and obese adults with no meta-analysis Labonté 2013, sixteen studies reported in fifteen articles whose own authors called the long-term effect weak and the short-term evidence inconclusive Ulven 2019, 52 trials summarised through a score the reviewers devised Bordoni 2017, one nine-week open-label feeding trial with an industry collaborator Pei 2017, and one crossover trial with 20 completers that missed its primary outcome Hasegawa 2024. That is a reasonable case for the absence of harm in the populations studied. It is not a large or settled body of work.

Second, these designs capture average responses over weeks. They are poorly suited to identifying an individual who genuinely reacts badly to a specific food, and a review of group means will not find that person.

Third, the outcomes people most often describe — skin flare-ups, mucus, joint stiffness the morning after cheese — are not the outcomes these papers were built to assess. The reviews cited here examined circulating biomarkers and, in one case, gene expression in blood cells Labonté 2013 Ulven 2019. A reader looking for a verdict on dairy and acne will not find it in this evidence, and should treat anyone offering a confident one with suspicion.

Fourth, the authors themselves say what they want next, and it is worth taking at face value: trials specifically designed to assess inflammation-related outcomes Labonté 2013, studies with enhanced designs, better reporting and proper characterisation of the dairy products used Ulven 2019, and work that follows the bioactive nutrients through digestion and metabolism rather than inferring their effects Bordoni 2017.

A reasonable way to sort it out

The useful reframing is to stop asking whether dairy is inflammatory and start asking which specific, testable thing might be going on. If the symptoms are digestive — gas, cramping, urgency an hour after a latte — lactose malabsorption is the candidate this evidence can actually test for, and the NIH panel describes how it is diagnosed rather than guessed at Suchy 2010. A diagnosis usually arrives with better news than total avoidance: at least 12 g of lactose in a single dose produced no or minor symptoms in adults and adolescents with diagnosed malabsorption, and 12 to 15 g was graded as well tolerated by most adults on moderate-quality evidence Suchy 2010 Shaukat 2010. Spreading dairy across the day with meals is a smaller change than cutting it out, and the panel’s own concern about calcium and vitamin D intake is a reason to prefer the smaller change Suchy 2010.

If the symptoms look immune rather than digestive, that is a conversation with a clinician, not a swap to oat milk and not a self-administered milk test. And if the honest answer is that dairy has never bothered you and you cut it because a video mentioned inflammation, the trials cited here give you very little reason to keep going — while being smaller, shorter and more entangled with dairy funding than the confident version of either argument admits.

References