Summary: A new artificial intelligence “digital twin” model shows that psychological and social factors—especially loneliness, insomnia and poor mental health—are far stronger predictors of future Type 2 diabetes than commonly appreciated. The study analyzed lifestyle and health data from 19,774 UK adults followed for up to 17 years.
Unlike conventional medical risk tools that rely primarily on blood tests, body mass index (BMI) or wearable data, this model focuses on behavioural, lifestyle and psychosocial information. The findings suggest that emotional wellbeing and social connection deserve equal emphasis alongside diet and exercise when preventing Type 2 diabetes.
Key Facts
- The Power of Three: Loneliness, insomnia and poor mental health were each associated with an estimated 35 percentage point rise in absolute Type 2 diabetes risk in the model’s simulations. When all three factors coexist, the model estimated a combined increase of about 78 percentage points.
- Stress and Biological Pathways: Researchers link these effects to prolonged stress responses: sustained elevation of stress hormones can drive chronic inflammation and impair the body’s insulin regulation.
- Diet and Stress Eating: The model found a connection between stress-related factors and dietary patterns known to increase diabetes risk—higher intakes of processed meats, salty foods and sugary cereals were identified as amplifiers within modeled causal pathways.
- Ethnic Disparities: The AI highlighted pronounced differences across ethnic groups: participants of South Asian, African and Caribbean ancestry showed notably higher estimated risk than White participants, underscoring the need for culturally sensitive prevention strategies.
- Cost-Effective Screening Potential: Because this digital twin framework uses self-reported lifestyle and psychosocial data rather than routine blood tests or continuous wearable streams, it could be applied more widely and affordably in underserved settings to identify people at high risk.
Source: Anglia Ruskin University
Overview
Researchers from Anglia Ruskin University in collaboration with Cranfield University, the University of Portsmouth and Intelligent Omics Ltd used retrospective data from the UK Biobank—19,774 adults followed up to 17 years—to build and test a transparent, simulation-capable digital twin framework. The results were published in Frontiers in Digital Health.
The digital twin model simulates individual health trajectories and tests hypothetical changes in behaviour or environment to estimate their long-term impact on Type 2 diabetes onset. Rather than depending on live sensors or laboratory tests, it uses historical lifestyle, behavioural and psychosocial inputs to forecast risk and to run “what-if” scenarios for prevention.
Key findings include a strong, independent role for psychosocial stressors—loneliness, sleep disruption and mental health problems—in elevating estimated diabetes risk. The model also linked stress-related factors to less healthy dietary choices, which in turn further amplified risk. In counterfactual simulations, improving psychosocial conditions was estimated to reduce predicted Type 2 diabetes risk by around 11.6 percentage points in the cohort examined.
The authors emphasise that the model does not establish causal proof in the real world but uses causal inference tools—such as directed acyclic graphs and counterfactual simulations—to make transparent, testable predictions about how changes in behaviour or context could alter risk. The framework achieved strong predictive performance in the sample (C-index = 0.90, SD = 0.004) and distilled 90 initial candidate predictors down to a final set of 14 variables used for personalized risk trajectories.
Quotes from the research team
Professor Barbara Pierscionek, Deputy Dean for Research and Innovation at Anglia Ruskin University, said the study highlights limitations of traditional risk models that focus mainly on BMI, age and blood pressure. She noted that digital twin systems can test tailored prevention strategies but that many current models depend on wearables, which may exclude underserved communities. The approach used here offers an alternative that is more accessible and scalable.
Dr Mahreen Kiran, the lead author and a postgraduate researcher at ARU, emphasised the importance of including psychosocial variables in health datasets. She said these often-overlooked factors carry meaningful signals about future disease risk and can make AI-based prevention more accurate and equitable.
Key Questions Answered:
A: Loneliness acts as a physiological stressor. Prolonged social isolation can trigger sustained cortisol release, prompting the liver to produce more glucose and making cells less responsive to insulin—processes that, over time, increase the risk of Type 2 diabetes.
A: A digital twin is a virtual model of an individual’s health profile. By combining personal data—sleep, stress levels, ethnicity and daily habits—the AI runs simulated scenarios to forecast likely health trajectories and to test which interventions could lower risk efficiently.
A: The model observed a modest protective association for cheese in some simulations, but this benefit was reduced when psychosocial stressors were present. This suggests that the protective effects of single foods can be outweighed by the harmful influence of chronic stress and poor mental health.
Editorial Notes:
- Edited for clarity by an editor.
- Journal paper reviewed in full by the reporting team.
- Additional context added by staff.
About this AI and diabetes research news
Author: Jamie Forsyth
Source: Anglia Ruskin University
Contact: Jamie Forsyth – Anglia Ruskin University
Image credit: Neuroscience News
Original Research: Open access. “A digital twin framework for predicting and simulating type 2 diabetes onset using retrospective lifestyle data” by Mahreen Kiran, Ying Xie, Graham Ball, Rudolph Schutte, Nasreen Anjum, and Barbara Pierscionek. Frontiers in Digital Health. DOI: 10.3389/fdgth.2026.1710829
Abstract
A digital twin framework for predicting and simulating type 2 diabetes onset using retrospective lifestyle data
Introduction:
Type 2 diabetes is a growing global challenge driven largely by modifiable lifestyle and psychosocial factors. Most current prediction tools emphasise biomedical markers and often rely on real-time wearable data or electronic health records, which limits utility in low-resource settings. This study presents a digital twin framework that uses retrospective lifestyle, behavioural and psychosocial data to forecast Type 2 diabetes onset and to simulate the potential effects of preventive interventions.
Methods:
Researchers used data from 19,774 UK Biobank participants with up to 17 years of follow-up. A penalised Cox proportional hazards approach estimated individual time-to-event risk trajectories from an initial pool of 90 candidate predictors. Predictors were refined through univariate screening, multicollinearity checks and variance filtering to produce a final model of 14 variables. Causal inference methods, including directed acyclic graphs and counterfactual simulations, were applied to explore how changes in behaviour or psychosocial conditions could alter disease progression.
Results:
The model showed strong predictive accuracy (C-index = 0.90, SD = 0.004). Psychosocial stressors—loneliness, insomnia and poor mental health—emerged as powerful independent predictors, each associated with about a 35-percentage-point increase in modeled absolute Type 2 diabetes risk and about a 78-percentage-point increase when combined. Diet reinforced these pathways: high processed meat, salt and sugary cereal intake amplified risk, while cheese showed a modest protective association that weakened under psychosocial stress. Counterfactual scenarios suggested that improving psychosocial conditions could lower estimated risk by roughly 11.6 percentage points in this cohort. The model also highlighted marked ethnic disparities in estimated risk.
Conclusion:
This work introduces a transparent, simulation-enabled digital twin approach that integrates psychosocial and lifestyle data to estimate Type 2 diabetes risk and to test behavioral interventions without depending on live data streams. The framework supports interpretable, personalized prevention planning and could help scale equitable, behaviorally informed public health strategies—particularly in underserved or low-infrastructure environments.