- Aug 30, 2019
- Jul 02, 2019
Can artificial intelligence (AI) make the diagnosis of diabetes faster? IBM researchers believe so. Recently they announced a screening tool that could identify type 1 antibodies in people's blood.
For the millions of people living with type 1 diabetes, everyday reality involves significant self-control. Without proper supervision and treatment, your pancreas does not produce enough insulin. The function of this substance is to metabolize glucose (blood sugar) to produce energy, which supplies the cells. That is, individuals in this condition need daily doses of insulin to ensure that their blood glucose levels are healthy, which requires these patients to be very vigilant about their health.
About 1.25 million people have type 1 diabetes in the United States alone. And about 40,000 new diagnoses are added to the list each year, according to the American Diabetes Association. Given this, it is intriguing that there is no standardized screening process to diagnose this condition at the outset. Doctors use family history and other known risk factors for type 1 diabetes to appear on the radar. However, until the diagnosis is made, the patient will likely undergo sudden trips to the ER - which will result in a medical report that will catch him unawares.
This is where the AI comes in. At the 79th Scientific Session of the American Diabetes Association in early June, IBM and JDRF (a formerly known Juvenile Diabetes Research Foundation), a nonprofit organization leading the research on type 1 diabetes, revealed a predictive tool using this technology . She mapped the presence of antibodies against type 1 diabetes in the blood to find out exactly when and how the condition could develop. Jianying Hu, the global leader in Artificial Intelligence Science at IBM Research, told Engadget that the AI was powered by data from more than 22,000 people from the United States, Sweden and Finland.
The program identified similarities between people with disease-specific antibodies and the time line of progression of type 1 diabetes.
"One of the biggest potentials of this kind of work in building machine learning models for type 1 diabetes is the ability to better identify who should monitor and how often to do it," said Hu, whose team has worked on this project for more than one year with the JDRF. "With what little we know, these antibodies are widespread in the progression of type 1 diabetes, but no one knows who is most susceptible to developing them and when." According to the scientist, AI models can give doctors "a more personalized time frame" to monitor people and how often they should be tested.
In the past, type 1 diabetes was called juvenile diabetes, because it is usually diagnosed in children, adolescents, and young adults. However, as Utpal Pajvani, an endocrinologist and assistant professor at Columbia University Medical Center, has pointed out, the disease can affect people of any age.
Pajvani, who is not affiliated with this project, explained that the general practice leads to screening only people who are from the "high-risk" group - who has a first-degree family member who has been diagnosed. As this diabetes is more uncommon, it is not justified to subject the general population to screening. He also warned that comprehensive screening methods, such as this, can lead to many false positives.
"If we submit to all people, including those who are at relatively low risk of developing it, who have the antibody but do not have a family history or other autoimmune disease, we will have a much higher rate of identification of people who may have a positive test for an antibody but with a low risk of actually developing the disease, "Pajvani told Engadget.
The risk of applying a comprehensive screening test for a rare disease like this means that you can also make people unnecessarily anxious about a condition they probably will not have. Essentially, there is no perfect screening test that does not have false positives, he added.
Despite the criticism, Pajvani sees a future for this type of technology. People who are at high risk for diabetes, and have the presence of these antibodies, still do not know what the timing of disease progression will be. This kind of artificial intelligence tool could give physicians the roadmap needed to map their course, explained the endocrinologist.
Hu said that soon his team will add more data, once collected in Germany, to be classified by AI. She added that another large part of the project is working with physicians to see how they can apply the tool and how the insights gathered from AI can be used in clinical trials.
"I love what people are thinking about it, and I love the idea of pushing forward the important clinical question of who will develop a condition, especially one as significant as type 1 diabetes," Pajvani added.
He said that, as a clinician, he has not yet seen AI leave the theory to be used in the day-to-day practice of treating patients. Hu believes that the presence of machine learning in medicine will only continue to "accelerate", and that it is "extremely important" work that can develop tools that are indispensable to doctors.
Today, this artificial intelligence does not provide a definitive screening method, but it does provide a way for how machine learning tools could be used for faster type 1 diabetes diagnoses in the future.