How Should We Read Studies Claiming That Gut Bacteria Cause Diabetes?
In this article
A Methodology Lesson Through a Case Study
In recent years, we have seen similar headlines almost every week: "Gut bacteria cause depression." "The microbiota is associated with cancer." "Gut flora triggers diabetes." Most of these studies use a method called "Mendelian randomization" — and the word "causality" associated with this method turns into very strong claims in news reports.
But do these studies really demonstrate causation? The answer is "sometimes yes, but in this field usually not" — and the reason is worth learning. In this article, we use a newly published study as a case study to show, step by step, which questions you should ask when reading a text like this. This will be useful not only for this study, but also for the hundreds you will encounter in the future .
- This study was not conducted on a patient. It is a bioinformatics analysis conducted using publicly accessible databases.
- No treatment was tested. No probiotic, no diet and no drug.
- No new test is being proposed. The identified "diagnostic markers" are not close to clinical use.
- The authors' own conclusion is also cautious: they state that the study "provides a theoretical basis" — in other words, it is exploratory research.
- Nobody should take probiotics or change their treatment on the basis of this study.
1. Mendelian randomization was used to identify bacterial groups thought to be causally associated with type 2 diabetes: 10 groups of gut bacteria .
2. A separate method (transcriptome-wide association and gene-expression analysis) identified genes thought to be causally associated with type 2 diabetes: 228 genes .
3. These 228 genes were cross-referenced with the bacterial groups to select approximately 20 genes .
4. A publicly accessible gene-expression dataset was used to apply machine learning and identify candidate diagnostic markers.
5. Single-cell data indicated that two genes were expressed at different levels in different groups.
This is not a careless study. The design is multilayered , with each layer narrowing down the previous one: first bacteria, then genes, then the intersection, then validation. It is conceptually a sound approach , and the authors present their results as a "theoretical basis," not a "treatment." The problem is not the study's intention, but the limitations of the raw data and how results like these are communicated.
Before criticizing the method, we need to understand it — because its underlying idea is very clever.
Suppose you notice that diabetes is more common in people who carry large amounts of a particular bacterium. This does not mean that the bacterium causes diabetes. There are three possibilities:
(a) The bacterium causes diabetes. (b) Diabetes (or its treatment, or the patient's diet) changes the bacterial balance — in other words, the arrow points in the opposite direction. (c) A third factor (diet, weight, antibiotic use) causes both .
Observational studies cannot distinguish these three possibilities. A randomized study can — but you cannot randomly assign people to "carry this bacterium."
The idea behind Mendelian randomization is this: Genes are randomly allocated at conception. Which gene variant someone inherits does not depend on that person's diet, weight or lifestyle — it is entirely a matter of chance. In other words, nature has performed a kind of randomization at birth. If a particular gene variant results in a person carrying more of a particular bacterium, comparing people who carry that variant with those who do not roughly resembles a randomized study. Furthermore, because this "randomization" occurs at birth , the possibility of reverse causation is also eliminated.
This is the heart of the article. The problem is not with the method itself, but with its application to gut flora. There are four separate challenges:
Your cholesterol level is largely determined by your genes. Your gut flora is not. It is predominantly determined by environmental factors such as what you eat, where you live, whether you use antibiotics and how you were born . The genetic contribution is relatively small. This means that the method's first condition — "the gene variant must strongly determine the trait under investigation" — is weakly met here. In the literature, this is called the "weak instrument" problem, and it can bias the result.
Most studies in this field use the same international database on gut flora. This source includes approximately 18 thousand people . In the world of genetic research, this number is small — for comparison, datasets on the disease side often cover hundreds of thousands of people. A small source dataset makes it harder to identify reliable genetic associations.
Genetic studies conventionally use a very strict threshold for considering an association genuine. The reason is that, when millions of locations are examined simultaneously, chance findings are inevitable.
However, when this strict threshold is applied to gut-flora data, almost no usable gene variants remain. That is why most studies in this field substantially relax the threshold. This makes it possible to obtain results — but also increases the likelihood of false-positive findings. Indeed, this study also reports using a relatively permissive threshold for gene selection.
The method generally identifies bacteria at the genus or family level — not at the species or strain level. Yet different species within the same bacterial genus may have completely opposite effects. This method also provides no information about what the bacteria do (which substances they produce). Furthermore, collecting samples from different countries using different methods adds noise to the data.
The study's second claim is that the genes identified using machine learning could be diagnostic markers . There is an instructive detail here, too.
A test's ability to discriminate is measured by a number called AUC . Its meaning is simple: the probability of correctly ranking a randomly selected patient and a healthy person.
0,5 = as good as a coin toss (in other words, of no use at all) · 0,7 = modest · 0,8 = good · 0,9 and above = very good. The markers identified in the study were described as "all above 0,7" , which was characterized as "good diagnostic efficacy" .
An AUC of 0,7 can be meaningful in many settings — particularly in diseases that are difficult to diagnose . However, type 2 diabetes is not such a disease. The tests required for diagnosis are simple, inexpensive, widely available and highly reliable: fasting blood glucose, HbA1c and the glucose tolerance test.
So there is no "diagnostic problem" that needs to be solved. This does not make the finding worthless — but it shows that the "diagnostic marker" framing has no clinical relevance . Studies like these are actually exploratory studies aimed at understanding disease mechanisms , and that is where their real value lies. The term "diagnostic marker" may even obscure the study's true contribution.
It is very important that this article does not lead to a misunderstanding. The association between gut flora and health is real and supported by robust scientific evidence — but that evidence comes from other kinds of studies.
- In cancer treatment: multiple independent studies have shown that gut flora is associated with the response to immunotherapy . Fecal microbiota transplantation studies provide much stronger evidence because they involve a direct intervention.
- In colorectal cancer: numerous studies have demonstrated an association between microbiota imbalance and the development of adenomas and cancer.
- Through diet: A fiber-rich diet has been shown to benefit gut flora, with measurable outcomes.
- And an important warning: as we have previously reported on our website, there are data suggesting that over-the-counter probiotic supplements may reduce the response to immunotherapy . So saying "the microbiota matters" does not mean "take probiotics" — sometimes it means the opposite.
The evidence above comes from studies in real people, with real samples, often involving an intervention . By contrast, the type of study discussed in this article is a statistical analysis of publicly available summary data. Both are science; but they do not carry the same evidentiary weight. One generates a hypothesis; the other tests it. The problem is reporting the former as if it were the latter.
Questions you can ask the next time you see a news story titled "gut flora causes this disease":
- Who was studied? Actual patients, or publicly available databases? A database analysis generates hypotheses, not proof.
- Was anything tested? Was an intervention (diet, probiotic, transplantation) applied, or was the study merely looking for associations?
- Where does the word "causal" come from? In Mendelian randomization, this word comes from the method's name — not from the strength of the evidence.
- Which bacterium was identified, and at what level? A finding at the "genus" level does not tell you which bacterium it refers to in practice.
- Does the result make biological sense? The identified association needs to be explainable through a known mechanism. If it cannot be explained, be cautious.
- Was it independently validated? Has it been replicated in a different population by a different team?
- And the most practical question: what does this finding change about what I should do today? The answer is often "nothing" — and that is normal. This is how science progresses.
"Scientists have proved it: Your gut bacteria cause diabetes! 10 bacterial species identified, a new blood test on the way."
The study was not conducted on patients; it is a statistical analysis of publicly available databases. No treatment was tested. The method carries the name "causality," but its basic conditions are difficult to satisfy when applied to gut flora. The identified markers have modest discriminative ability, and type 2 diabetes can already be easily diagnosed with a simple blood test . This is a hypothesis-generating study — and that is how the authors presented it.
- This article was not written to judge a single study or its authors; it is a discussion of methodology, using the study in question as an example.
- Mendelian randomization is not a flawed method and has produced highly valuable results when applied to suitable data.
- The association between gut flora and health is real; this article addresses not microbiota research itself, but how a particular type of study should be read .
- Hypothesis-generating studies are not worthless — this is how science progresses. The problem is presenting them as definitive findings.
- The assessments in this article draw on systematic reviews and methodological articles in the field.
- No treatment, supplement or test is being recommended.
Do not change anything because of this study. Do not start taking probiotics, change your diet or request a test. Research like this is conducted to guide future research , not to inform today's decisions.
If you are concerned about your diabetes risk, what to do is clear and very simple: see your doctor and have your fasting blood glucose and HbA1c checked. These tests are inexpensive, widely available and reliable.
The measures with the strongest evidence for gut health are also clear , and none requires buying supplements: a fiber-rich diet (vegetables, fruit, legumes, whole grains), fermented foods, reducing processed foods, avoiding unnecessary antibiotic use and regular physical activity. A Mediterranean-style diet is the most extensively studied model for both metabolic health and gut flora.
And a warning: Be cautious about commercial services selling "microbiome tests" or "personalized probiotics." Most of these products have captured a large market despite a weak scientific basis. This is precisely where news stories like these are most often used.
The microbiota is one of the health topics attracting the greatest public interest and the most intensive commercial exploitation. At the same time, readily accessible databases and ready-made analytical tools have made producing studies in this field very easy. The result: new publications every week titled "this bacterium causes this disease." This article's aim is not to criticize one study, but to enable readers to evaluate texts like these for themselves — because the flow will continue to accelerate.
When applied to suitable data, Mendelian randomization is a powerful and respected method for causal inference; it has produced valuable results in lipid and blood pressure research. The study in question also has a multilayered, conceptually consistent design, and its authors present their results as a "theoretical basis." The association between gut flora and health is supported by independent, stronger types of evidence : immunotherapy response, colorectal carcinogenesis and transplantation studies involving an intervention.
When Mendelian randomization is applied to gut flora, the first assumption (a strong instrument) is seriously challenged: flora is largely determined by environmental factors, the source database is relatively small, and statistical thresholds are therefore often relaxed — increasing false-positive risk. Bacteria are not identified at the species/strain level , and functional information is not provided. A systematic review states that poorly conducted studies in this field have become more numerous and sometimes produce biologically implausible results . Furthermore, the reported discriminative ability is modest , and in type 2 diabetes the diagnostic problem has already been solved , limiting the clinical relevance of looking for new markers. The study relies on a single dataset , and no independent external validation is reported.
Nothing changes for now. Type 2 diabetes is diagnosed using fasting glucose, HbA1c and, when necessary, a glucose tolerance test; these tests already perform very well. The main message for clinical practice concerns communication with patients: patients often come in with news stories like these and are directed toward commercial products such as "microbiome tests" or "personalized probiotics." Here, the doctor's role is to explain the difference between a hypothesis-generating study and an evidence-generating study. Interventions with strong evidence for gut health are diet-based and do not require supplements.
Will this method become reliable once larger and more detailed microbiome databases (at the species/strain level, with functional information) are available? Could studies in this field have a common quality-assessment tool developed — none currently exists. Are the biological functions of the identified candidate genes really related to diabetes? And a broader question: how should hypothesis-generating studies be positioned within the scientific publishing ecosystem — does their value diminish as they become easier to produce?
- Chen YY, Shi YF, Zhang XY, et al. The mediating role of genes in the influence of intestinal flora on type 2 diabetes and the screening of diagnostic markers. J Diabetes Investig. 2026. doi:10.1111/jdi.70386 (The study used as an example in this article.)
- A systematic review of the applications of Mendelian randomization assessing the causal relevance of the gut microbiome in human health and disease. medRxiv, 2025. (A comprehensive assessment of methodological problems in the field.)
- Kurilshikov A, Medina-Gomez C, Bacigalupe R, et al. Large-scale association analyses identify host factors influencing human gut microbiome composition. Nat Genet. 2021;53:156–165. (The source database used by most studies in this field.)
- Application of Mendelian randomization to appraise causality in relationships between the gut microbiome and cancer. medRxiv, 2026. (Limitations involving taxonomic resolution and measurement heterogeneity.)
- Davey Smith G, Hemani G. Mendelian randomization: genetic anchors for causal inference in epidemiological studies. Hum Mol Genet. 2014;23(R1):R89–R98. (The method's basic framework.)
- Related reading (drozdogan.com): What is gut flora, and how is it related to colorectal cancer? · How do gut bacteria affect cancer treatment? · Probiotics, prebiotics and fecal transplantation · Probiotic supplements may reduce the effectiveness of immunotherapy · Probiotics: myth or reality? · Diabetes, prediabetes and insulin resistance · Type 2 diabetes and the Mediterranean diet.
Editor's note: This content is intended solely for general information and discussion of scientific methodology; it is not medical advice, a diagnosis or a treatment recommendation. This article does not aim to judge the study used as an example or its authors; the research was conducted using methods accepted in its field, and its authors presented it as an exploratory study providing a theoretical basis . The methodological assessments reported here draw on systematic reviews and methodological publications in the field. Mendelian randomization is a valid and respected method for causal inference; the criticism in this article concerns not the method itself, but the specific limitations of applying it to gut-flora data . The association between the gut microbiota and health and disease is real and is supported by stronger types of evidence, including interventional studies; this article does not question the value of microbiota research. Hypothesis-generating studies are a natural and necessary part of scientific progress; the warning in this article concerns presenting findings like these as established evidence. The reported discriminative performance values in this study are modest , and no independent external validation is reported. Type 2 diabetes is already diagnosed with simple, inexpensive and highly reliable tests; no new test, supplement or treatment is recommended in this article. Adverse effects of over-the-counter probiotic supplements have been reported in some clinical settings — particularly in patients receiving immunotherapy; these products should not be used without consulting a doctor. Do not change any medication you take or your diet on the basis of this article. Consult the doctor who knows you for every decision concerning your health.
Was this article helpful?
Your feedback helps us improve our content.
