EHR-based risk calculator reduces 3-year diabetes risk in patients with prediabetes

Embedding an individualized diabetes risk calculator into primary care visits was associated with a significant reduction in progression to type 2 diabetes among patients with prediabetes, according to a study published in Learning Health Systems.

Patients whose care teams had access to an electronic health record (EHR)–based prediction model were significantly less likely to develop type 2 diabetes within 3 years than a propensity-score–matched cohort who received usual care in the same health system without individualized risk estimates. The three-year risk of developing type 2 diabetes was 19.5% versus 27.6%, respectively, the authors reported.

“By integrating a risk prediction model into routine care, healthcare providers can more effectively target interventions to those at greatest risk, thereby improving both clinical outcomes and resource allocation,” the authors wrote. “The findings suggest that even modest interventions, such as facilitating a simple conversation about individualized risk, may have tangible benefits, possibly because they prompt patients to engage more actively in their own care—even in the absence of protocol-driven intervention.”

The findings come as payers and provider organizations weigh how to allocate limited diabetes-prevention resources across a large prediabetes population. More than one-third of U.S. adults have prediabetes, the authors noted, and individual risk of progression varies widely. With newer pharmacotherapies such as tirzepatide entering the market, risk stratification is becoming increasingly important to make sure care is cost-effective.

“With the current high cost of GLP-1 and GIP receptor agonist medications, payers and provider organizations with risk contracts will need to weigh this expense against the likely clinical benefits, which will depend on each person's baseline risk,” the authors explained. “The present study's results suggest that providing individualized risk estimates and engaging patients has the potential to shift patient selection of preventive interventions to those at higher risk, improving cost-effectiveness and meaningfully impacting outcomes.”

The model, developed by a team at Tufts Medical Center and the American Medical Group Association, uses 11 clinical variables typically available in an EHR and estimates 3-year diabetes risk along with the projected benefit of interventions evaluated in the Diabetes Prevention Program (DPP; either an intensive lifestyle program or metformin). It was developed and validated using data from 2.2 million people with prediabetes in the Optum Labs Data Warehouse and accommodates missing data, according to the study.

Researchers integrated the tool into the workflow at Premier Medical Associates (PMA), a 100-provider multispecialty group in suburban Pittsburgh, Pennsylvania, that is part of Allegheny Health Network. Nurse navigators ran risk scores during chart preparation, and results were shared with clinicians during morning huddles ahead of visits.

The intervention shifted both the volume and targeting of preventive care. Before implementation, no patient had been formally referred to the local DPP, and fewer than 5% had been prescribed metformin, the authors wrote. Afterward, interventions increased and became strongly risk-stratified: 74% of high-risk patients received a DPP lifestyle program referral or metformin, compared with 19% of intermediate-risk and 8% of low-risk patients.

Provider confidence also rose with the new model. Before the tool, 41.6% of providers felt confident or very confident estimating an individual patient's risk; afterward, 92.8% did. Patient confidence was higher at baseline but did not change significantly (63.8% versus 66.9%).

“While the risk model appears to be a useful tool for providers, in terms of improving their confidence, our implementation study also highlights the importance of provider-patient communication in fostering trust and motivating behavior change.”

The authors noted that the intervention was a system-level bundle, including provider education and workflow redesign, which makes it difficult to isolate the effect of the risk conversation alone. They also noted the study was conducted within a single high-performing practice and that the control group was treated in other practices, meaning the outcome analysis was potentially subject to practice-level confounding. Despite the limitations, the study suggests that individualizing risk estimates for this patient population in the primary care setting could help reduce the likelihood of diabetes development for those with prediabetes.

“The outcome study provides evidence that this use of the model, even in the absence of protocol-driven follow-up interventions, may significantly reduce the risk of progression to type 2 diabetes,” the authors concluded. “Given the growing burden of prediabetes, integrating personalized risk prediction models into routine care represents a promising strategy to improve patient outcomes and optimize the use of healthcare resources.”