AI Used to ID Risk of Heart Disease in Diabetes Study

University of Gothenburg researchers were able to use artificial intelligence and conventional statistical methods to identify the most important indicators of elevated risk for cardiovascular disease and death.

Artificial intelligence is constantly being used in new and different applications in healthcare. A research team from the University of Gothenburg, is now using the power of AI in combination with conventional statistical methods in a study of risk factors in type 1 diabetes.

The study’s objective was to identify the most important indicators of elevated risk for cardiovascular disease and death.

"What's unique about this study is that we've included machine learning analyses - that is, algorithms for AI - to assess strength of association for cardiovascular risk factors," Aidin Rawshani, PhD, of Sahlgrenska Academy, University of Gothenburg, said in a release. Dr. Rawshani is the corresponding author of a new article in the journal Circulation.

The study is based on register data concerning 32,611 people with type 1 diabetes for whom the mean duration of the disease had been 18 years. Follow-up time averaged just over 10 years. Alongside traditional statistical analysis, AI was used: Autonomous learning enabled the computer software to improve its ability to predict death and cardiovascular events.

When the relative contribution of 17 risk factors was studied, five emerged as the strongest predictors: high long-term blood sugar (glycated hemoglobin) levels, kidney dysfunction, duration of type 1 diabetes, high systolic blood pressure (the first, higher figure of the two measured) and an excess of what is popularly known as "bad cholesterol" (low-density lipoprotein, LDL).

Here are some detailed findings from the study:

For three variables - blood sugar, systolic blood pressure and LDL - levels below those currently recommended in national guidelines proved to be associated with lower risks of cardiovascular disease and death.

Another finding in the study was the association between albuminuria (elevated levels of protein in the urine) and two- to fourfold risk elevation for the outcomes studied. Along with long-term high blood sugar, albuminuria was the factor that most clearly predicted these outcomes.

According to machine learning models, high blood sugar is believed to contribute to the development of the other cardiovascular risk factors. In addition, the researchers found a clear interaction effect between risk factors that cannot be influenced (age and duration of diabetes) and those that can (long-term high blood sugar, systolic blood pressure, LDL cholesterol and albuminuria).

Filed Under
500 characters remaining