According to research reviewed by Gram, a model combining 13 blood proteins with basic clinical measures predicted type 2 diabetes better than a common cholesterol-based ratio in about 20,000 UK Biobank participants (accuracy score 0.86 versus 0.74). The 2026 study in Nature Metabolism suggests blood protein patterns may someday offer a simpler way to spot insulin resistance early.
Insulin resistance is an early warning sign for type 2 diabetes, heart disease, and other metabolic problems, but it is hard to measure. According to research reviewed by Gram, scientists studied proteins in the blood of 161 people whose insulin sensitivity had been precisely measured. They found 488 proteins linked to how well the body responds to insulin, many coming from the liver, fat tissue, and immune cells. The team then built a model using 13 of these proteins along with basic clinical measures such as sex, HbA1c, and a triglyceride-to-HDL cholesterol ratio. The model predicted insulin sensitivity well and outperformed common markers at predicting type 2 diabetes in about 20,000 UK Biobank participants. Because the approach uses a standard blood sample, it may eventually offer a practical way to screen for metabolic risk. More testing in everyday clinical settings is still needed.
Key Statistics
A 2026 study published in Nature Metabolism found that a model combining 13 blood proteins with basic clinical measures explained about two-thirds of the differences in insulin sensitivity among 161 adults (R-squared of 0.67).
In about 20,000 UK Biobank participants, a blood protein model predicted type 2 diabetes with a concordance index of 0.86, compared with 0.74 for the triglyceride-to-HDL cholesterol ratio, according to a 2026 Nature Metabolism study.
The same 2026 Nature Metabolism research identified 488 blood proteins linked to insulin sensitivity in 161 people, with signals coming from the liver, fat tissue, and immune cells.
The Quick Take
- What they studied: Can proteins in a blood sample reveal how well a person’s body responds to insulin, and can that information predict future health problems like type 2 diabetes?
- Who participated: The main group included 161 adults with a wide range of insulin sensitivity, from very sensitive to very resistant. Additional groups included about 20,000 UK Biobank participants and an independent group of 1,700 people. The study reports 700 as its total sample size in the research details.
- Key finding: A model using 13 blood proteins plus three basic clinical measures explained about two-thirds of the differences in insulin sensitivity (R-squared of 0.67). In the large UK group, it predicted type 2 diabetes better than a common cholesterol-based ratio (accuracy score of 0.86 versus 0.74, where 1.0 is perfect).
- What it means for you: A simple blood test may someday help doctors spot metabolic risk earlier. These results come from research studies, so they do not yet mean a test is available or that you should change your care based on this report alone. Talk with a healthcare provider about your own risk.
The Research Details
The researchers used a method called proteomics, which measures thousands of proteins at once in a small amount of blood plasma. They used several laboratory workflows to make the measurements more complete and reliable.
The team started with 161 people whose insulin sensitivity had been carefully measured using a hyperinsulinaemic-euglycaemic clamp. This is considered the gold standard test, in which a steady dose of insulin is given through an IV while blood sugar is held steady, and the amount of sugar the body needs to absorb is tracked. By comparing protein levels with these precise results, the team identified 488 proteins linked to insulin sensitivity. In a subgroup of men, they also looked at how these proteins changed after an exercise program.
Next, the researchers built a prediction model using 13 proteins and three clinical measures. They tested it in a very large UK health database called UK Biobank, which follows participants over time, and in a separate group of 1,700 people. This two-step approach helps show whether the model works beyond the group it was built from.
The clamp test is accurate but expensive, time-consuming, and rarely used outside research labs. A blood-based model that can be run from a standard sample could make insulin resistance easier to measure in routine care. Testing the model in a separate group and in a very large database also helps show that the results are not just a one-time finding.
The study uses a well-accepted gold standard to measure insulin sensitivity, which strengthens its results. It was published in Nature Metabolism, a peer-reviewed scientific journal. Testing the model in an independent group and in the large UK Biobank adds credibility. However, the main detailed protein work involved a relatively small group, and most results are associations, which means they show links rather than proof that a specific protein causes a health problem. Independent replication in more diverse populations will be important.
What the Results Show
The researchers identified 488 proteins that were associated with how well the body responds to insulin. Many of these proteins come from the liver, fat tissue, and immune cells, which suggests that insulin resistance involves several organs working together rather than a single problem in one place.
From these proteins, the team selected 13, including IGFBP1, LEP, GDF15, PON3, and LDLR. When combined with sex, HbA1c (a measure of average blood sugar over about three months), and the triglyceride-to-HDL cholesterol ratio, the model explained about two-thirds of the variation in insulin sensitivity (R-squared of 0.67). In plain terms, this means the model captured a large share of the differences between people, though not all of them.
In about 20,000 UK Biobank participants, the model predicted type 2 diabetes more accurately than the triglyceride-to-HDL ratio alone. The concordance index was 0.86 for the protein model compared with 0.74 for the ratio. Unlike HbA1c, the model also appeared to predict future liver, kidney, heart and blood vessel, and breathing-related diseases, suggesting it may capture broader metabolic health.
In an independent group of 1,700 people, the model also showed stronger predictive value than common calculated insulin resistance measures, including the homeostatic model assessment of insulin resistance (HOMA-IR). This consistency across different groups supports the idea that the approach may be useful beyond the original study participants.
Some of the identified proteins shifted toward a more insulin-sensitive pattern after an exercise program in men. This suggests that these markers may respond to lifestyle changes, which could make them useful for tracking progress over time. The findings also point to the value of looking at tissue-specific signals, since different organs appear to contribute different protein patterns.
Doctors currently rely on fasting insulin, HbA1c, and lipid-based ratios to estimate metabolic risk. These tools are useful but have limits, especially for detecting early insulin resistance. This study builds on earlier work linking specific proteins such as GDF15 and LEP to metabolic health, and it extends that work by combining many proteins into a single model tested in large groups. The results suggest this approach may outperform several widely used measures.
The detailed protein analysis involved 161 people, which is a fairly small group for discovering complex patterns. Many of the findings show associations, meaning they reveal links between proteins and insulin sensitivity but do not prove cause and effect. The exercise results were limited to men, so they may not apply equally to women. Most participants were likely from specific populations, which may limit how well the results apply to other ethnic groups or age ranges. Specialized protein testing is also not yet available in most clinics, so the model would need further development and validation before routine use.
The Bottom Line
Confidence is moderate that blood protein patterns can help estimate insulin sensitivity, since the model performed well across several groups. Confidence is lower that this specific test will change clinical care soon, because it still needs validation in diverse populations and cost-effective testing methods. For now, the most reliable steps for metabolic health remain regular physical activity, a balanced diet, healthy sleep, and routine checkups that include HbA1c and lipid screening.
Adults concerned about type 2 diabetes, heart disease, or fatty liver disease may find this research encouraging, as may clinicians and researchers working on early risk detection. People already diagnosed with diabetes should continue following their doctor’s treatment plan. This research is not a reason to delay medical care, and it does not apply directly to children, pregnant people, or people with certain medical conditions without professional guidance.
Blood-based tests of this kind typically need several years of additional research, validation in different populations, and regulatory review before they become available in clinics. Meaningful lifestyle changes, such as increasing daily activity or improving diet quality, can improve insulin sensitivity within weeks to months, though full benefits for long-term disease risk often take a year or more to appear.
Frequently Asked Questions
Can a blood test show how sensitive my body is to insulin?
Current research suggests blood protein patterns may reflect insulin sensitivity. A 2026 Nature Metabolism study found a 13-protein model explained about two-thirds of insulin sensitivity differences (R-squared 0.67). The most precise current method, the clamp test, is not routine, so these tests are still being developed.
Does insulin resistance lead to type 2 diabetes?
Insulin resistance is a major driver of type 2 diabetes, and it is also linked to heart and liver disease. In a study of about 20,000 UK Biobank participants, a protein-based model predicted type 2 diabetes with a score of 0.86, compared with 0.74 for a common cholesterol ratio.
What lifestyle changes can improve insulin sensitivity?
Regular physical activity, a balanced diet with plenty of fiber, and good sleep are commonly recommended to improve insulin sensitivity. In the study, some blood proteins shifted toward a more insulin-sensitive pattern after an exercise program in men, though broader research is still needed.
Is HbA1c enough to predict future metabolic disease?
HbA1c is useful for diagnosing and monitoring diabetes, but it may miss early insulin resistance. The 2026 study found that unlike HbA1c, a blood protein model also appeared to predict liver, kidney, heart, and breathing-related diseases, suggesting broader metabolic risk insight.
Want to Apply This Research?
- Log weekly minutes of moderate-to-vigorous activity, fasting morning blood glucose readings if available, and waist circumference once per month. Track these alongside your most recent HbA1c and lipid results from your doctor, and set a goal of at least 150 minutes of activity per week.
- Add a 10-minute walk after your two largest meals each day, which research on blood sugar control has linked to improved glucose handling. Pair this with a daily check-in in the app about whether you chose vegetables, whole grains, or lean protein at meals, so you can build consistent habits.
- Review your logged activity, diet check-ins, and any lab results every three months to spot trends. Schedule HbA1c and lipid testing with your healthcare provider at least once a year, or more often if advised, and share your app data at appointments to support more personalized conversations about metabolic health.
This article summarizes a research study for general educational purposes and is not medical advice. The findings describe research tools that are not yet available for routine clinical use. Do not change medications, diet, or exercise for medical conditions without consulting a qualified healthcare provider. If you have concerns about diabetes, heart health, or insulin resistance, speak with your doctor.
This research translation is published by Gram Research, the science division of Gram, an AI-powered nutrition tracking app.