Conference Coverage

New risk-prediction model for diabetes under development


 

EXPERT ANALYSIS FROM THE WCIRCD 2018

– Clinicians treating patients with diabetes rely heavily on the U.K. Prospective Diabetes Study (UKPDS) Risk Engine and the Framingham Risk Score to predict outcomes, but the populations used for developing these tools differ significantly from the current U.S. diabetes population.

MDedge News/Doug Brunk

Dr. Vivian A. Fonseca

“All these risk engines have various degrees of accuracy along with several limitations, including that they are derived from data from various populations,” Vivian A. Fonseca, MD, said at the World Congress on Insulin Resistance, Diabetes & Cardiovascular Disease. “Sometimes the results may not be generalizable. That’s one of the big problems with the risk engines we’re using.”

To address these shortcomings, Dr. Fonseca, Hui Shao, PhD, and Lizheng Shi, PhD, have developed the Building, Relating, Assessing, Validating Outcomes (BRAVO) of Diabetes Model, a patient-level microsimulation model based on data from the ACCORD trial. The model predicts both primary and secondary CVD events, microvascular events, the progress of hemoglobin A1c and other key biomarkers over time, quality-adjusted life-year (QALY) function decrements associated with complications, and an ability to predict outcomes in patients from other regions in the world. The risk engine contains three modules for 17 equations in total, including angina, blindness, and hypoglycemia (Pharmacoeconomics. 2018;36[9]:1125-34). “There are lots of data now showing that if you get hypoglycemia, your risk of a cardiovascular event goes up greatly over the subsequent 2 years,” said Dr. Fonseca, who is chief of the section of endocrinology at Tulane University Health Science Center, New Orleans. “No other risk engine has that.”

When he and his associates applied the UKPDS Risk Engine to the ACCORD cohort, they found that the UPKDS Risk Engine overpredicted the risk of stroke (2.3% vs. 1.4% observed), MI (6.5% vs. 4.9% observed), and all-cause mortality (10.3% vs. 4% observed); yet it underpredicted congestive heart failure (2.2% vs. 4% observed), end-stage renal disease (0.5% vs. 3% observed), and blindness (1.35% vs. 8.1% observed). In the ACCORD cohort, baseline duration varied from 0 to 35 years. “Using left truncated regression, we can piece together the segmented follow-up times for 10,251 patients to a complete diabetes progression track from 0 years to 40 years after diabetes onset,” he said.

Dr. Fonseca said that BRAVO improves four aspects of diabetes risk prediction compared with other tools: It better captures the effects of body weight on cardiovascular risks, cost, and QALYs; it better captures hypoglycemia; it has a globalization module to calibrate regional variation of cardiovascular risks; and it has both utility and QALY equations developed from the same study cohort. Internal validations studies found that BRAVO predicted outcomes from the ACCORD trial, including congestive heart failure, MI, stroke, angina, blindness, end-stage renal disease, and neuropathy. Data from the ASPEN, CARDS, and ADVANCE trials were used to conduct external validation, and the incidence rates of 28 endpoints correlated with that of BRAVO “extremely well.” In addition, BRAVO has been calibrated against 18 large randomized, controlled trials conducted after the year 2000. “Regional variation in CVD [cardiovascular disease] outcomes were included as an important risk factor in the simulation,” said Dr. Fonseca, who is also assistant dean for clinical research at Tulane. Results to date show a high prediction accuracy (R-squared value = .91).

He and his associates are currently examining ways to apply BRAVO in clinical practice, including for risk stratification. “Let’s say you have a large health system, and you want to separate out your patients who have high, medium, or low risk for diabetes and make sure they get they get the right care according to their stratification,” he explained. “A couple of large health systems are trying this out right now.”

BRAVO can also be used as a tool for cost-effectiveness analysis and program evaluation. In fact, he and his colleagues at five medical centers are working with the American Diabetes Association “to see what effect a certain intervention will have on outcomes in people with diabetes over a number of years, and how cost effective it might be.”

Finally, BRAVO can be used for diabetes management in clinical practice. “Based on an individual’s characteristics, the BRAVO model potentially simulates future outcomes such as complications and mortality, providing a transparent platform for shared decision making,” he said.

Dr. Fonseca disclosed that he has an ownership interest in the development of BRAVO.

Next Article: