Artificial intelligence can predict which patients will develop prediabetes or type 2 diabetes by analyzing routine medical data, according to a 2026 study of 21,023 Canadian patients. Gram Research analysis shows that blood sugar levels, body weight, cholesterol, and blood pressure are the strongest predictors of diabetes risk. Early identification using these AI tools could help doctors recommend lifestyle changes like diet and exercise to reverse prediabetes before it becomes permanent diabetes.
Researchers used artificial intelligence to analyze medical records from over 21,000 Canadian patients to predict who would develop prediabetes or type 2 diabetes. By examining routine health information like blood sugar levels, weight, cholesterol, and blood pressure, the AI identified key warning signs that doctors can use to catch the disease early. This matters because prediabetes is reversible—people can prevent diabetes through diet and exercise if caught in time. The study shows that combining everyday medical data with smart computer analysis could help doctors identify high-risk patients before serious problems develop.
Key Statistics
A 2026 study of 21,023 Canadian primary care patients found that artificial intelligence models could predict diabetes risk using four key health measurements: blood sugar levels, body mass index, cholesterol, and blood pressure.
According to Gram Research analysis of this 2026 study, blood sugar levels emerged as the most important predictor when AI models analyzed patterns across 21,023 patient medical records to identify diabetes risk.
The research demonstrates that machine learning can distinguish between patients who will maintain normal blood sugar, develop prediabetes, or progress to type 2 diabetes by analyzing routine clinical data from primary care electronic medical records.
The Quick Take
- What they studied: Can artificial intelligence predict which patients will develop prediabetes or type 2 diabetes by analyzing their medical records?
- Who participated: Over 21,000 patients from Canadian primary care clinics. The study looked at their routine health measurements and medical history stored in electronic medical records.
- Key finding: According to Gram Research analysis, blood sugar levels, body weight, cholesterol, and blood pressure are the strongest predictors of diabetes risk. AI models using this information can identify high-risk patients before they develop the disease.
- What it means for you: If you have prediabetes, your doctor might use these AI tools to assess your risk and recommend lifestyle changes like diet and exercise to prevent full diabetes. Early detection gives you the best chance to reverse prediabetes through lifestyle changes alone.
The Research Details
Researchers gathered medical information from 21,023 patients in Canadian primary care clinics. They used electronic medical records—the digital files doctors keep on patients—and added detailed information about medications patients were taking and whether they smoked. They then trained different artificial intelligence computer programs to learn patterns from this data and predict which patients would develop normal blood sugar, prediabetes, or type 2 diabetes. The AI models were designed to be ’explainable,’ meaning researchers could understand which health measurements the computer thought were most important for making predictions.
This approach is different from just looking at individual patients one at a time. Instead, the AI learned from thousands of patients’ experiences to spot patterns that humans might miss. The researchers compared different AI methods to see which one worked best at predicting future diabetes risk.
This research matters because prediabetes is a critical window of opportunity. Unlike type 2 diabetes, which is usually permanent, prediabetes can be reversed through lifestyle changes. If doctors can identify people with prediabetes early using AI tools, they can recommend diet and exercise changes before the disease becomes irreversible. Using routine medical information that doctors already collect means this approach could be practical and affordable to implement in regular doctor’s offices.
The study used a large sample size (over 21,000 patients), which makes the findings more reliable. The research used real medical records from actual patients, not laboratory conditions. However, this was a retrospective study, meaning researchers looked backward at existing data rather than following patients forward over time. The study was conducted in Canada, so results may need to be tested in other countries and populations to confirm they work everywhere. The researchers acknowledge that future studies should validate these findings with prospective cohorts (following patients forward in time).
What the Results Show
The artificial intelligence models successfully identified four key health measurements that predict diabetes risk: blood sugar levels, body mass index (BMI, a measure of weight relative to height), cholesterol levels, and blood pressure. These four factors were consistently the most important predictors across different AI models tested.
The study demonstrates that machine learning can effectively use routine clinical data to distinguish between patients who will maintain normal blood sugar, develop prediabetes, or progress to type 2 diabetes. By analyzing patterns in thousands of patient records, the AI learned which combinations of these measurements indicated highest risk.
The explainability techniques used in this study revealed that blood sugar levels were particularly important—the AI paid special attention to this measurement when making predictions. This aligns with what doctors already know about diabetes development but confirms it through a data-driven approach using thousands of patients.
The research also examined the role of medication use and smoking status in diabetes prediction. While these factors were included in the analysis, the study emphasizes that the four main health measurements (blood sugar, weight, cholesterol, and blood pressure) were the strongest predictors. The inclusion of detailed medication information expanded the dataset beyond what was previously available, providing a more complete picture of patient health.
This study builds on existing knowledge about diabetes risk factors. Doctors have long known that high blood sugar, excess weight, high cholesterol, and high blood pressure increase diabetes risk. What’s new here is using artificial intelligence to analyze thousands of patient records simultaneously and identify the precise patterns and combinations of these factors that best predict who will develop diabetes. This data-driven approach confirms traditional risk factors while potentially identifying new patterns humans might have missed.
The study analyzed existing medical records rather than following patients forward in time, which limits certainty about cause and effect. The research was conducted using Canadian primary care data, so results may differ in other countries or healthcare systems. The study was relatively small in terms of the number of patients actually used to develop the AI models (though it analyzed 21,023 records). The researchers acknowledge that more features and data from prospective studies are needed to improve the model’s accuracy and ensure it works reliably across different populations.
The Bottom Line
If you have prediabetes or risk factors like high blood sugar, excess weight, high cholesterol, or high blood pressure, work with your doctor on lifestyle changes including diet and physical activity. These changes can reverse prediabetes before it becomes type 2 diabetes. Ask your doctor about using AI-assisted risk assessment tools if available in your clinic. Confidence level: High—this recommendation is supported by strong evidence that lifestyle changes can reverse prediabetes.
This research is most relevant for people with prediabetes, people with family history of diabetes, and people with risk factors like excess weight or high blood pressure. Primary care doctors should care about this research because it shows how to use their existing medical records to identify high-risk patients. Healthcare systems and clinics should consider implementing these AI tools to catch diabetes early. People with type 2 diabetes should focus on management rather than prevention, though early intervention in family members could be valuable.
Lifestyle changes can begin improving blood sugar levels within weeks to months. Most people who make sustained diet and exercise changes can see measurable improvements in blood sugar within 3-6 months. Reversing prediabetes typically takes 6-12 months of consistent lifestyle changes, though individual results vary.
Frequently Asked Questions
Can artificial intelligence predict if I’ll get diabetes?
Yes, according to a 2026 study of 21,023 patients, AI can predict diabetes risk using routine health measurements like blood sugar, weight, cholesterol, and blood pressure. Early prediction allows doctors to recommend lifestyle changes that can prevent or reverse prediabetes.
What are the best predictors of type 2 diabetes?
Research shows blood sugar levels, body weight, cholesterol, and blood pressure are the strongest predictors of diabetes risk. These four measurements, when analyzed together by AI, can identify high-risk patients before symptoms develop.
Can prediabetes be reversed with lifestyle changes?
Yes, prediabetes is reversible through diet and physical activity. The study emphasizes that early identification of prediabetes enables preventive interventions. Most people see improvements in blood sugar within 3-6 months of sustained lifestyle changes.
How accurate is AI at predicting diabetes from medical records?
A 2026 analysis of 21,023 patient records showed AI models could effectively predict diabetes progression using routine clinical data. However, researchers note that future prospective studies are needed to validate accuracy across different populations and healthcare systems.
Should my doctor use AI to check my diabetes risk?
If you have prediabetes or risk factors like high blood pressure, excess weight, or high cholesterol, ask your doctor about AI-assisted risk assessment. These tools can identify high-risk patients early, enabling preventive lifestyle interventions before diabetes develops.
Want to Apply This Research?
- Track your fasting blood sugar levels weekly (if you have a home glucose monitor) or at doctor visits. Also monitor weight weekly and blood pressure monthly. Log any dietary changes and exercise minutes daily to correlate with blood sugar improvements.
- Use the app to set reminders for daily 30-minute walks or other moderate exercise, log meals to track carbohydrate and fiber intake, and record blood pressure and weight measurements. Create alerts when blood sugar, weight, or blood pressure readings trend upward to prompt immediate lifestyle adjustments.
- Create a dashboard showing your four key metrics: blood sugar, weight, cholesterol (from lab results), and blood pressure. Set monthly goals for each metric and track progress over 3-6 month periods. Share trends with your doctor to adjust your prevention plan as needed.
This research describes how artificial intelligence can help predict diabetes risk using medical data. It is not a substitute for professional medical advice, diagnosis, or treatment. If you have concerns about diabetes risk, prediabetes, or any health condition, consult your healthcare provider. The findings are based on Canadian primary care data and may not apply equally to all populations. Always work with your doctor before making significant lifestyle changes or starting any diabetes prevention program.
This research translation is published by Gram Research, the science division of Gram, an AI-powered nutrition tracking app.