Cheese, Coffee, and GERD: What Genetic Studies Actually Show

Cheese looks protective. Coffee looks innocent. The anxiety result looks broken.

If you wanted one clean list of foods and habits that cause reflux, a new 2026 Mendelian randomization study by Zhang et al. is both useful and annoying. Useful, because Mendelian randomization can test whether an exposure is likely to play a causal role rather than merely travel with the usual suspects. Annoying, because one of the paper's splashiest findings rests on a phenotype that confuses mental health with healthcare use.

The study used genetic instruments for alcohol consumption frequency, coffee intake, cheese consumption, and "anxiety," then tested their relationship with gastroesophageal reflux disease, or GERD, in a genome-wide association meta-analysis with 78,707 GERD cases and 288,734 European-ancestry controls. The headline numbers were striking: alcohol raised GERD risk with an odds ratio of 1.52, cheese lowered it with an odds ratio of 0.36, coffee showed no significant causal association, and anxiety came in at an absurd-looking odds ratio of 22.60.

One of those findings probably matters a lot. One probably confirms what most clinicians already assume. One is a decent reason to stop blaming your espresso for everything. And one should have triggered much harder editorial scrutiny.

Why Mendelian randomization matters here

Observational reflux research is messy. People who report more symptoms also differ in body weight, smoking, diet quality, sleep, alcohol intake, stress, medication use, and how often they see doctors. That is why coffee has spent years in dietary jail despite thin causal evidence. Mendelian randomization, or MR, tries to reduce that mess by using inherited genetic variants associated with an exposure as proxies for long-term differences in that exposure.

MR is not magic. It still depends on the quality of the underlying genetic instruments and phenotype definitions. If the exposure is poorly defined, the analysis can produce a very precise estimate of the wrong thing. That caveat matters a lot for the anxiety result.

The cheese signal is the real story

The most interesting result in the 2026 study is the cheese finding, because it is not alone. An independent 2022 MR study by Sun et al. tested multiple modifiable GERD risk factors using a different GWAS cohort and found a near-identical odds ratio of about 0.36 for cheese or dairy consumption. That is roughly a 64% lower odds of GERD for genetically predicted higher cheese intake.

Replication matters more than novelty. One surprising result can be noise. Two MR studies using different datasets that land on almost the same estimate deserve attention.

Sun et al. also pushed the analysis further. In multivariable MR, the protective association persisted after accounting for BMI and smoking, which argues against the entire result being a simple artifact of healthier lifestyle patterns. Their network MR suggested body composition explained about 28% of the association. That does not prove a mechanism, but it does suggest the cheese signal may work partly through weight or adiposity pathways rather than acting only as a direct antacid.

There are plausible biological explanations. Calcium can buffer gastric acid. Dairy fat may affect lower esophageal sphincter physiology. Fermented dairy also contains metabolites that could matter. Still, plausible is not proved. The safest statement is narrower: across two independent MR analyses, genetically predicted higher cheese intake consistently tracks with lower GERD risk.

That does not tell you how many ounces of cheddar to eat, whether all cheeses behave the same way, or whether someone with active reflux symptoms will feel better after a cheese board. MR estimates lifelong genetic tendencies, not meal-by-meal symptom experiments. It also does not give an absolute risk reduction, so the finding is interesting before it is practical.

Coffee keeps getting blamed for more than it earns

The coffee result in Zhang et al. was null. The study did not find a significant causal relationship between genetically predicted coffee intake and GERD. That does not mean coffee never aggravates symptoms. It means the genetic evidence in this dataset does not support coffee as a population-level causal driver of GERD diagnosis.

That distinction helps reconcile the new paper with older clinical lore. An observational review from 2021 identified coffee and tea as common self-reported symptom triggers. That is believable. Symptoms can flare after a specific drink even if habitual coffee intake does not cause the disease itself. Triggering heartburn on Tuesday and causing GERD over years are different questions.

Older observational evidence also refuses to line up neatly. A large 2004 twin study by Nilsson et al., using a within-pair co-twin design that inherently controls for shared genetics, found no association between alcohol consumption and frequent reflux symptoms in their sample and even suggested coffee might be protective in men. That does not cancel the 2026 MR study. It does show that endpoint choice matters: symptom frequency and diagnosed GERD are not interchangeable outcomes.

Alcohol still looks like the boringly credible risk factor

Zhang et al. reported an odds ratio of 1.52 for alcohol consumption frequency and GERD. That is directionally consistent with standard medical advice, even if the older twin study complicates the picture. The most sensible read is that alcohol remains a plausible causal risk factor, but the size and consistency of that effect may vary by population and outcome definition.

The twin study looked at symptomatic reflux. Zhang et al. analyzed genetically predicted drinking frequency against a GERD GWAS diagnosis phenotype. Those are related, not identical. If two studies disagree, the first job is to compare what each one actually measured, not to average them into mush.

The anxiety finding is where the paper steps on a rake

The paper's anxiety estimate, an odds ratio of 22.60, is the sort of number that should make you put the kettle down and inspect the methods. The problem is not that the gut-brain axis is fake. It is very real. The problem is that this study's "anxiety" exposure was not a validated anxiety disorder phenotype. It was defined as having visited a general practitioner for nerves, anxiety, tension, or depression.

That is not a clean psychiatric trait. It is partly a mental health signal and partly a healthcare utilization signal. People who see doctors more often are more likely to have anxiety-like complaints recorded and more likely to receive a GERD diagnosis, regardless of whether anxiety biologically causes reflux. That is classic ascertainment bias baked into the exposure itself.

No amount of MR-Egger intercept testing or leave-one-out analysis repairs a phenotype that mixes the trait of interest with the chance of entering the medical record. If your ruler is bent, the statistics do not straighten it.

Better MR evidence points in a different direction. A 2023 bidirectional MR study by Zeng et al. found that GERD appears to causally increase anxiety disorder risk, with an odds ratio of 1.50 and a 95% confidence interval of 1.29 to 1.75 after adjustment for smoking, alcohol, and BMI in multivariable MR. That is a much smaller number, but it is also a much more believable one.

A separate 2024 MR study by Wang et al. found that mood swings causally increased GERD risk across gastrointestinal diseases, which supports the broader idea that psychological traits can affect gut outcomes. But mood swings are not the same construct as "went to a GP for nerves, anxiety, tension, or depression." One is an attempt to measure a phenotype. The other is a charting event with extra baggage attached.

What you can actually do with this evidence

The evidence supports a few practical conclusions, with the usual scientific fine print attached. Cheese is the most surprising signal and the best replicated one. Alcohol remains a reasonable target for reduction. Coffee does not look like a causal GERD driver at the disease level, even though some people may still notice acute symptoms after drinking it.

There is also broader MR evidence that GERD is not just about food. An MR study from 2025 linked sedentary leisure behavior to higher GERD risk, and an MR study from 2024 linked sarcopenia to higher GERD risk. That fits the body-composition mediation signal from Sun et al. Reflux looks increasingly like a condition shaped by diet, body composition, behavior, and symptom perception all at once.

What the evidence does not tell you is how much cheese is enough, whether one extra latte changes your lifetime risk, or whether these estimates apply outside European-ancestry populations. Zhang et al. used European-ancestry genetic data only. The published information also does not provide absolute risk estimates, so the odds ratios do not translate into a shopping list with dosage instructions.

The clearest takeaway is not that cheese is a miracle food. It is that GERD advice gets sharper when you separate symptom triggers from disease causes and stop treating messy phenotypes as clean biology.

So the current scorecard is simple. Cheese has a replicated protective signal in MR. Coffee looks innocent on disease causation, even if it annoys some esophagi in real time. Alcohol still belongs on the usual-suspects list. And any headline claiming anxiety multiplies GERD risk twenty-two-fold deserves the same response as day-old gas station sushi: back away slowly and read the methods.