Artificial intelligence (AI) and digital twin technology are converging to advance precision medicine in endocrinology. Emerging research suggests that integrating these technologies may improve disease modeling, enable more personalized treatment strategies, and support clinical decision-making.

Digital twins are virtual, patient-specific representations that integrate data from multiple sources. Mathematical models, AI algorithms, and machine learning techniques then analyze and update these models as new information becomes available. This combination allows researchers to simulate disease progression, predict treatment responses, and test therapeutic strategies in a virtual environment before they are applied to patients.  

“The convergence of AI and digital twin technology has the potential to transform endocrine care by enabling more personalized, predictive, and data-driven clinical decision-making,” says David Ahn, MD, chief of diabetes services for Hoag Hospital in Newport Beach, California, and a member of the AACE Endocrine AI editorial board.  “As AI-driven medicine becomes increasingly integrated into clinical practice, understanding digital twin technology can help endocrinologists optimize treatment decisions and improve patient outcomes.”  

How Digital Twins Work

Originally developed in engineering and manufacturing, digital twin technology creates a virtual representation of a physical object that can be continuously updated with real-world data. In healthcare, a digital twin is a computational model of an individual patient that simulates physiological processes, predicts outcomes, and enables personalized treatment decisions.  

In a recent lecture at the 2026 AACE Annual Meeting, Dr. Ahn used the Apollo 13 film to illustrate how digital twin technology works. After an explosion forces the crew to abandon their mission and return to Earth, the astronauts shut down nearly all onboard systems to conserve power. Before reentry, however, they must restart the capsule’s computers and critical systems. The challenge is that Mission Control has never attempted such a startup sequence under these conditions. Turning on systems in the wrong order could drain the remaining power before the spacecraft is fully operational.  

Ken Mattingly (played by Gary Sinise), who was grounded before the mission, is assigned to solve the problem. Working in NASA’s Apollo capsule simulator—a replica (or twin) of the spacecraft—he spends hours testing different startup sequences within the mission’s strict power constraints. Through trial and error, he identifies the single functional solution and then shares it with the astronauts to bring them safely home.  

Building Digital Twins 

Digital twin platforms are multistep processes that combine patient data, physiological modeling, and artificial intelligence. 

• Data Integration. First, large volumes of patient-specific data from multiple sources are collected. Modern healthcare systems increasingly generate such data through electronic health records, laboratory testing, CGM devices, insulin pumps, smartwatches, mobile health applications, and lifestyle and behavioral data.   

• Physiological Modeling. Researchers then create mathematical models that represent biological systems and depict the underlying physiology and pathophysiology of diseases.  

• Artificial Intelligence and Machine Learning. AI algorithms examine patient data and continuously update the digital twin. Machine learning models detect patterns that may not be apparent through conventional clinical analysis and improve prediction accuracy over time.  

• Simulation and Prediction. Once established, the digital twin can run simulations to evaluate potential treatment strategies, forecast outcomes, and then select and determine the intervention.  

Digital Twins in Endocrinology

Diabetes management represents the most advanced application of digital twin technology in endocrinology. Researchers are building models that integrate continuous glucose monitoring (CGM) data, insulin administration records, dietary intake, physical activity, and metabolic parameters. These can allow researchers to test multiple treatment strategies virtually before selecting the optimal approach for individual patients.  

For instance, a 2025 study published in npj Digital Medicine found that AI-enabled metabolic digital twins significantly improved glycemic control in people with type 1 diabetes. In the 6-month randomized clinical trial involving 72 people with type 1 diabetes, researchers used personalized metabolic digital twins to optimize automated insulin delivery settings biweekly and to allow users to test treatment changes in a virtual simulation before applying them in real life. The digital-twin approach significantly improved glucose control, increasing time spent in the target glucose range from 72% to 77% and reducing HbA1c from 6.8% to 6.6%, while standard feedback reports alone provided no additional benefit. 

A 2025 randomized trial published in The New England Journal of Medicine evaluated a digital twin–enabled precision care program for adults with type 2 diabetes. The intervention integrated CGM, wearable sensor data, laboratory results, and behavioral information to create a personalized digital representation of each patient's metabolism. The platform continuously incorporated new patient data to model individual responses to food, physical activity, sleep, and other lifestyle factors, enabling tailored recommendations to improve glycemic control and reduce reliance on glucose-lowering medications. At 12 months, 71% of patients in the intervention group achieved an HbA1c below 6.5% without medications other than metformin compared with 2.4% of those receiving usual care.  

“Taken together, this body of evidence positions AI-enabled digital twin technology not as a single-endpoint pharmacologic substitute but as a precision metabolic intervention,” said Shashank Joshi, MD, in an interview with AACE Endocrine AI. 

Beyond diabetes, AI and digital twin technology are increasingly being applied across other areas of endocrinology, although most of these applications remain in early development or validation stages, noted Dr. Joshi, a consultant endocrinologist at Lilavati Hospital and Joshi Clinic in Mumbai, India, whose research includes remission of type 2 diabetes using AI. In obesity management, AI-powered digital twins are being developed to model weight-loss trajectories, personalize lifestyle and pharmacologic interventions, and identify patients most likely to benefit from anti-obesity medications or bariatric surgery. 

In thyroid disease, researchers are exploring digital twin platforms that integrate hormonal, imaging, and clinical data to support individualized treatment planning and disease monitoring. In osteoporosis and metabolic bone disease, AI-enabled digital twins can combine bone mineral density measurements, imaging findings, biomechanical data, and clinical risk factors to improve fracture risk prediction and simulate responses to anti-resorptive or anabolic therapies.  

Challenges and Future Directions 

Despite their promises, digital twins face multiple challenges, according to Dr. Ahn. Data quality remains a major concern. Predictions are only as reliable as the data used to construct the model. Incomplete, inaccurate, or biased data can limit performance.  

Model validation is equally important. Digital twins must undergo rigorous testing to demonstrate clinical accuracy and reliability before widespread adoption. Privacy and security considerations are also critical because these systems depend on large volumes of sensitive patient information.  

As digital twin technology becomes integrated into clinical workflows, endocrinologists will likely encounter these AI-powered tools as part of routine decision-making and personalized patient care. For this reason, they should develop a working understanding of how digital twins are built, validated, and applied. “Even clinicians who do not directly use digital twin platforms will increasingly encounter AI-enhanced tools that rely on similar modeling methods,” said Dr. Ahn. “Understanding their capabilities—and their limitations—will be vital for evaluating evidence, interpreting recommendations, and integrating these technologies safely into patient care.”