Mathematical Model Forecasts Diabetes Complications 15 Years Out, Confirms Legacy Effect of Early Control

Researchers at the University of Chicago have developed a comprehensive mathematical model that predicts diabetes complications up to 15 years in advance, offering clinicians a powerful new tool to understand how early treatment decisions shape patient outcomes.

The Diabetes Outcome Model of the US (DOMUS), published in Diabetes Care, draws on data from 129,000 patients with type 2 diabetes (T2D) and tracks 17 distinct complications, from traditional cardiovascular events to increasingly recognized conditions such as depression and dementia.

The model addresses a critical gap in diabetes care planning, the authors noted. Unlike existing prediction models that rely heavily on data from the United Kingdom Prospective Diabetes Study, which included minimal representation of African American, Asian, Latino, and American Indian populations, DOMUS uses real-world data from Kaiser Permanente Northern California's multiethnic patient population. "We wanted to build a model that represented the people we actually treat in the US--a socioeconomically and racially diverse population," lead researcher Neda Laiteerapong, MD, MS, professor of medicine and chief of general internal medicine at the University of Chicago, said in a statement. The cohort reflects the demographic diversity of contemporary US practice, with 74% White, 24% Asian, 17% Hispanic, and 9% Black patients, followed quarterly over 12 years.

Laiteerapong said the project began with a patient's simple question: does early treatment really matter? "I wanted to say, 'Yes, absolutely,' but I didn't have any evidence to support that," she recalled. "The challenge with diabetes is that the benefits of treatment, like controlling blood sugar, blood pressure, cholesterol, weight, or quitting smoking, often don't show up until many years later." The DOMUS model now provides that evidence. "First-year HbA1c did, in fact, help predict long-term complications. So yes, those early months matter," Laiteerapong said. First-year hemoglobin A1c emerged as a predictor for most long-term complications, suggesting that the initial months after diagnosis carry lasting consequences--a finding with clear implications for the "wait and see" approach some clinicians take with newly diagnosed patients.

The research team built DOMUS as a microsimulation model with quarterly cycles, tracking patient trajectories through four interconnected modules: diabetes medications, seven biomarker trajectories (including HbA1c, blood pressure, cholesterol levels, and body mass index [BMI]), 17 diabetes-related health outcomes, and mortality. The model predicts not only established complications such as end-stage kidney disease, myocardial infarction, stroke, and amputation, but also conditions that have gained prominence in diabetes care: atrial fibrillation, depression, and dementia.

To ensure accuracy, the team split their data into 3 samples for development (50%), calibration (25%), and validation (25%). The validation process demonstrated that simulated and empirical data aligned closely across biomarkers, events, and mortality predictions. Cumulative failure curves for all complications and deaths fell within the 95% confidence intervals of observed data, confirming the model's predictive accuracy across diverse patient subgroups.

The model incorporates social determinants of health alongside clinical factors. Neighborhood deprivation index appeared as a predictor in heart failure, ischemic heart disease, and foot ulcer equations. English language proficiency influenced risk for myocardial infarction, angina, atrial fibrillation, and nonproliferative retinopathy. These findings underscore how structural inequities shape diabetes outcomes independent of individual clinical characteristics.

Researchers used instrumental variable analysis to account for treatment selection bias, employing primary care physician prescribing patterns as instruments. This methodologic approach helps disentangle the effects of medications from the clinical factors that drive prescribing decisions. The model tracked 15 common medication regimens that represented 99.5% of treatment choices in the cohort.

The median follow-up time reached 21 quarters, generating 748,582 person-years of data. Annual event rates varied widely, from 0.0003 for blindness to 0.0181 for nonproliferative retinopathy. At baseline, 25% of patients had depression, 8% had ischemic heart disease, and 5% each had atrial fibrillation, angina, and myocardial infarction.

Among the model’s limitations the authors noted the Kaiser Permanente population tends to be healthier than the general US population, with participants more than 10% less likely to be overweight or have hyperlipidemia and 9% less likely to have hypertension compared with national averages. The system's commitment to equitable care may underestimate how structural inequity affects diabetes progression in other settings. Additionally, the cohort's medication patterns predate widespread use of GLP-1 receptor agonists and SGLT2 inhibitors, though the model's structure allows researchers to incorporate newer agents by adjusting biomarker equations based on clinical trial data.

The limited 13-year follow-up constrains extrapolation beyond this timeframe, though the researchers used current age rather than time since diagnosis to measure time in survival models, which should improve predictions outside the sample period. External validation in other healthcare systems represents an important next step.

"Historically, when we ask policy questions about diabetes, we often can't do it using real people in real time," Laiteerapong explained. "We have to estimate or simulate the outcomes using mathematical models." DOMUS offers researchers, policymakers, and health system managers a tool to evaluate how interventions affect population health and health disparities. The model can inform cost-effectiveness analyses, guide clinical decisions about treatment intensification, and help quantify the long-term value of diabetes prevention programs. As healthcare systems work to eliminate persistent racial and ethnic disparities in diabetes outcomes, the study authors concluded that DOMUS provides a foundation for understanding how early clinical decisions and social determinants combine to shape patient trajectories. "This model can be used by scientists, policymakers, and health system managers to better understand how choices can affect population health and health disparities, including the broad diversity of U.S. races and ethnicities," they wrote.