Researchers developed a computer program that can predict whether someone might develop Alzheimer’s disease up to 10 years before symptoms appear, using information from their medical records. The study looked at health data from nearly 20,000 people who developed Alzheimer’s and over 111,000 people who didn’t. The computer program was able to correctly identify who would get Alzheimer’s about 80% of the time. The findings suggest that certain health conditions—like heart disease, sleep problems, mood disorders, and vitamin B12 levels—might be early warning signs. If this approach works in real-world settings, doctors could potentially help people prevent or delay Alzheimer’s before it starts.

The Quick Take

  • What they studied: Whether a computer program could use information from medical records to predict who would develop Alzheimer’s disease at least 10 years before they actually got sick.
  • Who participated: The study included medical records from 19,473 people who were diagnosed with Alzheimer’s disease and 111,922 people who never developed it. The records covered at least 10 years of medical history for each person.
  • Key finding: The computer program correctly predicted Alzheimer’s disease about 80% of the time using medical records from 10+ years before diagnosis. The program identified that certain health conditions—including heart disease, sleep problems, mood disorders, inflammation, pain conditions, and low vitamin B12—were associated with higher Alzheimer’s risk.
  • What it means for you: This research suggests that doctors might someday be able to identify people at risk for Alzheimer’s years before symptoms appear, potentially allowing for early prevention or treatment. However, this is still early-stage research, and the computer program hasn’t been tested in real doctor’s offices yet. Talk to your doctor if you’re concerned about Alzheimer’s risk.

The Research Details

Researchers used a type of artificial intelligence called a ‘random forest model’—think of it as a computer program that learns patterns by looking at many examples. They fed the program medical information from over 131,000 people’s health records, including details about diseases, medications, test results, and procedures. The program looked at 2,499 different pieces of health information to find patterns that appeared in people who later developed Alzheimer’s.

To make sure the program worked fairly, the researchers split their data into two parts: they used 75% to teach the computer program what patterns to look for, and then tested it on the remaining 25% of people it had never seen before. This is like studying with practice problems and then taking a test on new problems to see if you really learned the material.

The study looked backward in time, examining medical records from at least 10 years before people were diagnosed with Alzheimer’s. This allowed researchers to see what health conditions or patterns appeared years before the disease was officially diagnosed.

This research approach matters because Alzheimer’s disease develops silently in the brain for many years before symptoms appear. By the time someone gets diagnosed, significant brain damage has already occurred. If doctors could identify people at risk years in advance, they might be able to start treatments or lifestyle changes early enough to prevent or slow the disease. Using medical records that doctors already keep means this approach could potentially be added to regular healthcare without requiring new tests.

This study has several strengths: it used a very large number of people (over 131,000), looked at real medical records over a long time period (10+ years), and used a well-established computer learning method. However, the study was published on a preprint server (medRxiv), which means it hasn’t gone through the full peer-review process that published research typically undergoes. The study also looked at historical data, so it hasn’t been tested to see if the computer program actually works when doctors use it in real life. The program’s accuracy of 80% is good but not perfect—it would miss some people who would develop Alzheimer’s and incorrectly flag some people who wouldn’t.

What the Results Show

The computer program achieved an 80% accuracy rate (measured as an ‘area under the ROC curve of 0.80’), which means it correctly identified people at risk for Alzheimer’s about 4 out of 5 times. This is significantly better than random guessing, which would be 50% accurate.

The program identified several health factors that appeared more often in people who later developed Alzheimer’s. These included age and sex (which we can’t change), but also modifiable factors like cardiovascular disease, inflammation, pain conditions, sleep disorders, and mood disorders like depression and anxiety. The program also flagged certain medications (diuretics, which remove excess water from the body), vitamin B12 deficiency, seizures, and digestive system issues.

Interestingly, the program found that trauma history was associated with later Alzheimer’s development, as were other neurodegenerative diseases (diseases that damage nerve cells). This suggests that brain health problems might cluster together or share common causes.

The research identified that multiple health conditions working together were more predictive than any single condition alone. This suggests Alzheimer’s likely develops from a combination of factors rather than one single cause. The findings also highlight the importance of cardiovascular health, since heart disease was among the top predictive factors. Sleep and mood disorders were particularly notable because these are conditions that can sometimes be treated or improved.

This is described as the first study to predict Alzheimer’s 10 years in advance using electronic health records and machine learning. Previous research has identified individual risk factors for Alzheimer’s (like heart disease, sleep problems, and depression), but this study is novel in combining many factors together using a computer program to make predictions years before diagnosis. The 80% accuracy is promising compared to other early prediction methods, though direct comparisons are difficult because previous studies used different approaches.

Several important limitations should be considered: First, this study looked at historical data, so it hasn’t been tested in real doctor’s offices yet. Second, the study population came from electronic health records, which means it may not represent all people equally—some groups might have more complete medical records than others. Third, the program’s 80% accuracy means it would miss about 1 in 5 people who would develop Alzheimer’s, and incorrectly identify some people who wouldn’t get it. Fourth, the study doesn’t prove that these health conditions cause Alzheimer’s, only that they’re associated with it. Finally, the program was trained on data from a specific time period and population, so it might not work as well in different groups or in the future.

The Bottom Line

Based on this research, the most evidence-based approach is to focus on modifiable risk factors: maintain cardiovascular health, manage sleep disorders, treat mood disorders like depression and anxiety, maintain healthy vitamin B12 levels, and manage inflammation and pain conditions. These recommendations have moderate confidence because while this study identifies these as associated with Alzheimer’s risk, it doesn’t prove they cause it. Do not make major health decisions based solely on this research—talk to your doctor about your personal risk factors.

This research is most relevant for people with a family history of Alzheimer’s, people over 65, and people with cardiovascular disease, sleep disorders, or mood disorders. Healthcare providers and researchers should pay attention to this work as it develops. People without risk factors don’t need to panic—this is early research that hasn’t been tested in real clinical settings yet. This research is NOT a substitute for medical advice from your doctor.

If this approach is eventually used in clinical practice, benefits would likely take years to appear. Any preventive interventions would need to be started years before Alzheimer’s symptoms would have appeared anyway. Realistic expectations would be measured in years to decades, not weeks or months. This is about prevention and early intervention, not immediate treatment.

Want to Apply This Research?

  • Track modifiable Alzheimer’s risk factors: sleep quality (hours per night and sleep disturbances), mood (using a simple daily mood scale), cardiovascular health markers (blood pressure, heart rate if available), and vitamin B12 levels from annual blood work. Create a monthly summary to share with your doctor.
  • Users could set app reminders to: schedule annual blood work to check vitamin B12 and cardiovascular markers, maintain consistent sleep schedules, track mood patterns and seek help if depression or anxiety worsens, and monitor cardiovascular health through regular activity and blood pressure checks. The app could provide educational content about each risk factor and connect users to resources.
  • Establish a baseline of current health metrics, then track changes over 3-6 month periods. Share trends with your doctor during annual checkups. Focus on modifiable factors (sleep, mood, cardiovascular health) rather than unchangeable ones (age, family history). Use the app to identify which factors are improving or worsening over time.

This research is preliminary and has not been peer-reviewed or tested in real clinical settings. The computer program’s predictions are not a diagnosis and should not be used to diagnose or treat Alzheimer’s disease. This information is for educational purposes only and should not replace professional medical advice. If you have concerns about Alzheimer’s risk or cognitive changes, please consult with a qualified healthcare provider. The study identifies associations between health conditions and Alzheimer’s risk, but does not prove that these conditions cause Alzheimer’s disease. Individual risk varies greatly based on genetics, lifestyle, and other factors not captured in this study.