According to Gram Research analysis, machine learning models can predict dementia risk with 80% accuracy by analyzing routine blood tests, particularly thyroid hormones, vitamin B12, and cholesterol levels. A 2026 case-control study of 50,598 people found that an AI system called XGBoost identified dementia risk by recognizing patterns in medical records collected over years. This breakthrough suggests doctors could use existing blood tests to spot who might develop dementia early, when treatments may be most effective.
Researchers developed an artificial intelligence system that can predict who might develop dementia by analyzing routine blood tests and medical records. The study looked at nearly 50,000 people’s health information from 2017 to 2024 and found that a machine learning model called XGBoost could identify dementia risk with 80% accuracy. The system flagged three key warning signs: abnormal thyroid hormone levels, low vitamin B12, and low HDL cholesterol. This breakthrough means doctors could potentially catch dementia risk early using tests they already do, without expensive new screening procedures.
Key Statistics
A 2026 case-control study of 50,598 people found that an XGBoost machine learning model achieved 80% accuracy (AUC of 0.802) in predicting dementia risk using routine blood tests and medical records from a hospital system.
According to Gram Research analysis of the study, the three strongest predictors of dementia risk were thyroid-stimulating hormone (TSH) levels, vitamin B12 levels, and HDL cholesterol levels, all measured through routine clinical care.
The machine learning model’s performance remained consistent across men and women, and results held up in multiple sensitivity analyses, suggesting the dementia risk prediction system is reliable across different populations and analytical approaches.
The Quick Take
- What they studied: Can a computer program predict who will develop dementia by looking at routine blood tests and medical history?
- Who participated: Nearly 50,600 people from a large hospital system (5,622 who developed dementia and 44,976 who didn’t) tracked from 2017 to 2024
- Key finding: An AI model correctly identified dementia risk 80% of the time by analyzing thyroid hormones, vitamin B12 levels, and cholesterol numbers
- What it means for you: Your regular blood work might contain hidden clues about dementia risk. If you’re over 50, ask your doctor about these three markers during routine checkups. However, this is a screening tool, not a diagnosis—abnormal results don’t mean you’ll definitely develop dementia.
The Research Details
Researchers used a case-control study design, which is like comparing two groups of people: those who developed dementia and those who didn’t. They looked back at 7 years of medical records and blood test results from a large hospital system in Massachusetts. They fed all this information into a machine learning program—essentially teaching a computer to recognize patterns that appear before someone develops dementia.
The computer program they used is called XGBoost, which works by building many decision trees (simple yes/no questions) and combining them to make predictions. Think of it like a doctor who has reviewed thousands of patient files and learned which combinations of test results tend to appear before dementia develops.
They tested whether the model worked equally well for men and women, and whether the results held up when they changed their analysis methods slightly. This helps ensure the findings are reliable and not just lucky coincidences.
This approach is important because it uses data doctors already collect during routine visits. There’s no need for expensive new tests or brain scans. If this system works in real hospitals, it could help identify people at risk before symptoms appear—when treatments might be most effective. Early detection is crucial because some dementia treatments work better when started early.
The study is reasonably strong because it included a large number of people (over 50,000), used real medical data from actual patient care, and tested the model in multiple ways to ensure reliability. The researchers compared their results across different groups and used sensitivity analyses (checking if results change with different assumptions). However, the study only included people from one hospital system in Massachusetts, so results might differ in other regions or populations. The study was published in 2026, making it very recent research.
What the Results Show
The XGBoost machine learning model achieved an AUC (a measure of accuracy) of 0.802, which means it correctly identified dementia risk about 80% of the time. This is significantly better than random guessing and suggests the model could be useful in clinical practice.
The three strongest warning signs the computer learned to recognize were: thyroid-stimulating hormone (TSH) levels, vitamin B12 levels, and HDL cholesterol levels. When these markers were abnormal in a person’s medical history, the risk of developing dementia increased. The model worked equally well for both men and women, suggesting it’s fair and reliable across sexes.
When researchers tested the model using different methods and assumptions, the results stayed consistent. This means the findings aren’t fragile or dependent on one specific way of analyzing the data. The model’s performance remained strong even when researchers made changes to their analysis approach.
Beyond the three main markers, the model also considered other health conditions and patterns in medical records. The fact that the model performed consistently across different subgroups and sensitivity analyses suggests it captures real biological patterns rather than statistical flukes. The researchers found that using longitudinal data (information collected over time) was more powerful than just looking at a single point in time.
Previous research has suggested links between thyroid problems, vitamin B12 deficiency, and dementia risk, but this study is among the first to show that machine learning can combine these signals with other medical data to make practical predictions. The 80% accuracy rate is competitive with other dementia risk prediction models, but this one has the advantage of using only routine tests that most people already get.
The study only included people from one hospital system in Massachusetts, so the results might not apply to other regions or populations with different demographics. The study looked backward at medical records (retrospective), which is less powerful than following people forward over time (prospective). The model was developed and tested on the same dataset, which could make it seem more accurate than it actually is in real-world use. The researchers didn’t test whether using these predictions actually helps doctors prevent dementia—that would require a separate study. Finally, the model identifies risk but doesn’t explain why these specific markers matter biologically.
The Bottom Line
If you’re over 50, ask your doctor to check your thyroid function (TSH), vitamin B12 levels, and HDL cholesterol as part of routine care. If any of these are abnormal, discuss with your doctor whether additional monitoring or treatment is appropriate. This is a screening tool to identify risk, not a diagnosis. (Confidence: Moderate—this is promising research but needs real-world testing before widespread use.)
Anyone over 50 with a family history of dementia should pay attention to these findings. People with thyroid problems, B12 deficiency, or low HDL cholesterol should discuss dementia risk with their doctors. Healthcare systems and researchers should care because this could lead to a practical, low-cost screening tool. People without these risk factors shouldn’t panic—many people with abnormal markers never develop dementia.
If you have abnormal markers, changes won’t happen overnight. Correcting vitamin B12 deficiency or thyroid problems might take weeks to months. Any reduction in dementia risk would likely take years to become apparent. This is about long-term prevention, not immediate effects.
Frequently Asked Questions
Can blood tests predict if I’ll get dementia?
Machine learning analysis of routine blood tests can identify dementia risk with 80% accuracy, particularly by measuring thyroid hormones, vitamin B12, and cholesterol. However, abnormal results don’t guarantee dementia will develop—they indicate increased risk requiring monitoring and discussion with your doctor.
What blood markers show dementia risk?
Three key markers emerged from the research: thyroid-stimulating hormone (TSH), vitamin B12 levels, and HDL cholesterol. When these are abnormal in medical records over time, dementia risk increases. Regular monitoring of these markers during routine checkups may help identify at-risk individuals early.
How accurate is this dementia prediction model?
The XGBoost model achieved 80% accuracy (AUC of 0.802) in identifying dementia risk. This means it correctly identified risk about 4 out of 5 times. While promising, the model needs real-world testing in clinical practice before becoming standard screening.
Should I get tested for dementia risk if I’m under 50?
This research focused on older adults, and dementia risk increases significantly after age 50. If you’re younger but have family history of dementia or thyroid/B12 problems, discuss personalized screening with your doctor rather than assuming you need testing now.
Can I prevent dementia if my blood markers are abnormal?
Correcting abnormal thyroid function and vitamin B12 deficiency may reduce dementia risk, though this study didn’t prove prevention. Treating these conditions has other health benefits regardless. Discuss specific interventions with your doctor based on your individual results.
Want to Apply This Research?
- Log your TSH, vitamin B12, and HDL cholesterol levels from annual blood work. Track these three markers annually and note any trends (increasing or decreasing). Set reminders for annual blood work if you’re over 50.
- Use the app to set a reminder to discuss these three specific markers with your doctor at your next checkup. If any are abnormal, use the app to track whether you’re taking recommended supplements (B12) or medications (thyroid, cholesterol) and monitor follow-up test results.
- Create a yearly health dashboard showing your TSH, B12, and HDL cholesterol trends. Compare results year-to-year to catch changes early. If any marker becomes abnormal, increase monitoring frequency to every 3-6 months until corrected.
This research presents a promising screening tool but is not a diagnostic test. Abnormal blood markers do not mean you will develop dementia. This study was conducted at one hospital system and needs validation in other settings before widespread clinical use. Always consult with your healthcare provider about your individual dementia risk, interpretation of blood test results, and appropriate preventive measures. This article is for educational purposes and should not replace professional medical advice.
This research translation is published by Gram Research, the science division of Gram, an AI-powered nutrition tracking app.