A person newly diagnosed with diabetes may be told about a long list of possible future problems, from kidney disease to heart attacks and nerve damage.
But knowing which problem is most likely to happen next for that particular person is much harder.
Researchers at the University of Maryland School of Medicine have developed a new computer-based calculator designed to make that prediction more personal. It can estimate short-term risks for nine diabetes complications and repeatedly update those estimates using information collected during normal medical care.
The study was led by Dr. Rozalina G. McCoy and was published in Nature Communications. The researchers developed the Diabetes Complications Risk Calculator, known as DCRC, using health data from more than 400,000 adults with newly diagnosed diabetes across the United States.
Diabetes occurs when the body cannot properly regulate blood glucose. Persistently high glucose can gradually injure large and small blood vessels, which helps explain why the disease can affect so many different organs.
Damage to large blood vessels can contribute to heart attacks and strokes. Damage involving smaller blood vessels and nerves can harm the kidneys, eyes and feet, while diabetes treatments themselves can sometimes cause dangerously low blood sugar.
Doctors already use risk calculators in many areas of medicine. A common example is estimating a person’s chance of developing cardiovascular disease over several years so that the patient and doctor can decide how aggressively to manage cholesterol or blood pressure.
But diabetes creates a more complicated forecasting problem. A patient may be at risk of several conditions at once, and the most immediate danger can change as new illnesses develop, medications change or laboratory results improve or worsen.
The Maryland team designed its calculator with this changing picture in mind. Instead of producing only a fixed long-term score, the system can generate updated estimates month by month and when new clinical encounters add information to the medical record.
The models use details that doctors commonly collect anyway. These include a patient’s age, diagnoses, medications, laboratory results and measures related to kidney function and other health conditions.
Researchers used machine learning to discover patterns within this large amount of information. In simple terms, the computer examined past patients to learn which combinations of characteristics tended to appear before particular complications developed.
The DCRC can estimate the likelihood of nine different problems, including heart and blood vessel disease, stroke, kidney disease, nerve damage and serious high- or low-blood-sugar events. Looking at these outcomes together may provide a more realistic picture of the choices clinicians face.
To test the system, the researchers first examined how well it performed in the large national dataset used for development. They then tested it separately with patients treated at Mayo Clinic, an important step because prediction tools can sometimes perform well where they were created but poorly in a different healthcare system.
The models showed good to strong predictive performance for many of the outcomes. However, the researchers reported that performance was not equally strong for every complication.
The study also revealed that complications were far from rare during the first years after diagnosis. Around one in three patients developed at least one complication within a year, and the proportion rose above 40% within two years.
Risk was higher in people with characteristics such as older age, high blood pressure, poorer kidney measures and additional medical conditions. Importantly, an individual’s risk was not fixed and could change as health and treatment changed.
This changing risk is where the calculator could become especially useful. Rather than telling someone only that diabetes may cause kidney or heart problems someday, a doctor might be able to identify which complication deserves the greatest attention during the coming months.
That information could influence how often a patient is monitored or which preventive measures are discussed first. It might also make conversations more concrete by helping patients understand why a particular medicine, test or lifestyle change is being recommended.
University of Maryland researchers describe the system as a decision-support tool rather than an automated doctor. Clinical judgment remains essential because risk scores cannot fully account for a person’s priorities, symptoms, social circumstances or other details that may not be captured in electronic records.
The study also has limitations that matter when judging how widely the calculator can be used. The development data came from insured patients, which may leave out important experiences of people with limited or inconsistent access to health care.
Electronic health records can also contain missing information and reflect how frequently someone visits a doctor. A person who receives more medical care may have complications detected earlier simply because clinicians have more opportunities to find them.
Before the DCRC becomes part of routine diabetes care, researchers plan to test it in real clinical workflows. The key question is not merely whether a computer can correctly predict risk, but whether giving those predictions to patients and clinicians leads to better care and fewer complications.
There is also a practical challenge in presenting multiple risk estimates. Too many alerts can create information overload, so future versions will need to deliver predictions in a way that helps rather than distracts busy healthcare teams.
The study is nevertheless strengthened by its very large patient population, its use of information collected in everyday medical care and its testing in an independent Mayo Clinic population.
The ability to update predictions over short periods also makes the approach more closely match the changing nature of diabetes than many traditional long-term calculators.
The findings should therefore be seen as an encouraging step rather than proof that the tool is ready to guide treatment on its own.
If further studies show that it improves decisions and outcomes, the DCRC could help move diabetes care from broadly warning about every possible complication toward identifying the problems that are most urgent for each individual patient.
If you care about diabetes, please read studies about bananas and diabetes, and honey could help control blood sugar.
For more health information, please see recent studies about Vitamin D that may reduce dangerous complications in diabetes and plant-based protein foods may help reverse type 2 diabetes.
Source: University of Maryland School of Medicine.
