Simple Tool Predicts 2-Year Diabetes Risk

Gregory A. Nichols, PhD


June 06, 2018

From Practice to Research

In the world of healthcare, we generally use research to guide clinical practice. But from time to time, health organizations don't have the luxury of waiting for well-designed methodical studies to play out. Such was the case when the Centers for Medicare & Medicaid Services announced in 2016 that Medicare would start paying for diabetes prevention services in 2018.[1]

Kaiser Permanente Northwest (KPNW), an integrated delivery system with more than 550,000 members, and its sister programs—Kaiser Permanente Hawaii (KPH), with roughly 250,000 members, and Kaiser Permanente Georgia (KPG), with about 300,000 members—faced the daunting task of rolling out these new services to a significant number of people. And they weren't looking at just Medicare enrollees.

The new Medicare benefit would be based on the Diabetes Prevention Program (DPP), a lifestyle intervention of healthy eating and moderate physical activity. Given the enormous health costs for people with diabetes[2] and the proven, enduring benefits of DPP,[3,4,5] it made good business sense for Kaiser to provide a relatively low-cost intervention to prevent substantial downstream costs. But offering it to everyone who was eligible right away could quickly overwhelm available resources, as an estimated 38% of adults have prediabetes.[6]

We wondered: Was there an easy way to target these resources to a subset of people with prediabetes who are at greatest risk?

Needed: A Simpler Algorithm

There are several published algorithms for estimating the risk of developing diabetes. For example, the Framingham Offspring Study[7] formulated a diabetes risk score that provided excellent discrimination with a relatively small set of demographic and clinical variables; this was subsequently validated by researchers using KPNW data.[7,8]

However, this scoring algorithm required eight variable inputs to calculate and then estimate risk over 8 years. To be immediately clinically and operationally useful, Kaiser was looking for a much more abbreviated set of inputs that could predict risk within the next year or two. A highly experienced endocrinologist who served on the KPNW diabetes steering committee thought it would be interesting to look at only glycated hemoglobin A1c and body mass index (BMI) for such prediction.

Thus, we collected readily available data from the electronic health records of KPNW, KPH, and KPG to quickly ascertain the usefulness of A1c and BMI for predicting short-term diabetes risk among patients with prediabetes.[9]

This was not a research project, and we were not searching for the best algorithm for predicting who would develop diabetes. We knew that the 5.7%-6.4% range of A1c values that comprise prediabetes was probably overly broad for estimating short-term risk, so we created four categories of A1c values using small increments: 5.7%-5.8%, 5.9%-6.0%, 6.1%-6.2%, and 6.3%-6.4%.

We then created a matrix with those categories and standard categories of BMI (normal weight, < 25 kg/m2; overweight, 25-30 kg/m2; obese, 30-35 kg/m2; and severely obese, ≥ 35 kg/m2).

Using a patient's most recent A1c measurement between 2009 and 2011, we looked forward 2 years to determine who developed diabetes (A1c ≥ 6.5%) and placed them in the simple matrix.

Remarkable Findings

The pattern of the results was not surprising, but the proportional differences between the cells of the matrix were remarkable. The segment of patients with A1c levels in the 6.3%-6.4% range who developed diabetes within 2 years was two to three times greater than that in the 6.1%-6.2% range for all BMI categories, and 10-20 times greater than for the group in the 5.7%-5.8% range.

Of note, this was true even among those for whom BMI was missing because height was not recorded in the electronic health record. That is not to say that BMI was unimportant. As one would expect, the proportion of patients who developed diabetes was greater as the BMI increased.

Also of interest, the findings were quite similar in all three regions of the country.

Clinical Relevance

Motivating patients to participate in exercise programs is notoriously difficult. Most people who are overweight or obese got that way in part because of an aversion to exercise, making them an especially resistant group.

The simple two-variable matrix we created could be printed on a card or posted in an exam room for easy reference. Perhaps pointing to a chart that clearly displays a patient's nearly immediate risk of developing diabetes would be a catalyst for the patient to at least listen and hopefully participate in a prevention program.

As noted, our results were strikingly similar across the three diverse regions of Kaiser Permanente health plans. Because KPNW is predominantly composed of white persons, KPH has a large proportion of native Hawaiian and Asian persons, and KPG has a large proportion of black persons, the results would seem to generalize across racial categories. Moreover, we believe the simplicity of the matrix and the distinct differences in risk across the A1c and BMI categories warrant dissemination.


Worthwhile analyses and potentially important findings need not be the result of clinical trials or even rigorous observational analyses. As shown here, a quality improvement project that originated as an operational question of how to best allocate precious healthcare resources can provide relevant information for the busy clinician.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.