Developing Algorithms to Predict Type 1 Diabetes in Children

Type 1 diabetes is an autoimmune disease induced by the body's own immune system attacking the insulin-producing beta-cells of the pancreas. As a result, affected children become dependent on external insulin. For many years, researchers focused their efforts on finding ways to prevent or delay the outbreak of type 1 diabetes. The first drug that can effectively delay the onset of the metabolic disease was approved in the USA last year. Immunotherapy using the monoclonal antibody Teplizumab can give the affected children up to three more years free of symptoms. Especially during childhood and adolescence, more time without having to manage the blood glucose levels day in, day out, can be a major factor when it comes to quality of life for the whole family. For optimal effectiveness treatment with Teplizumab needs to be started in a pre-symptomatic stage of the disease. So called islet autoantibodies serve as biomarkers in blood for early-stage type 1 diabetes. Screening programs like the Fr1da study led by Helmholtz Munich test for islet autoantibodies in the general population to identify these children before they develop symptoms. However, identifying a critical time point to start preventive treatments is still challenging, as some children have a fast progression while for others it may take years until the development of their first clinical symptoms. With advanced statistics and machine learning Helmholtz Munich researchers can make these predictions easier. Their work is supported by the Deutscher Diabetiker Bund e. V. (DDB).

Predicting Disease Progression with Machine Learning

Using immunological, genetic and metabolic data of Fr1da participants, Helmholtz Munich researchers established a progression likelihood score, that can predict the progression from substages of type 1 diabetes. Thereby, the researchers can identify the timepoint at which a particular child will benefit from prevention measures. For the study, the researchers analyzed data from 447 children included in the Fr1da study population having multiple autoantibodies, who had been enrolled between 2015 and 2021. “From previous studies we knew, that around 0.3 % of German children have pre-symptomatic type 1 diabetes and will most likely develop clinical type 1 diabetes within the next 10 years. The objective of our study was to develop a sub-staging strategy that identifies children within a general population who are at high risk of developing clinical type 1 diabetes within the next 2 years,” explains José Maria Zapardiel Gonzalo, bioinformatician within the Fr1da study group.

Vision for the Future

“With our work, we want to accelerate the recruitment of high-risk children into clinical trials of late-stage pre-symptomatic type 1 diabetes through population-based islet autoantibody screening,” explains Andreas Weiss, biostatistician at the Helmholtz Munich Institute of Diabetes Research. The progression likelihood score allows sub-staging of children with stage 1 type 1 diabetes into children with fast, medium and slow progression. This already comes to practice within the F1da study. “The Fr1da study is representative of the general population because it involves children who participated in a public health screening program. They were not preselected based on genetic or familial diabetes risk,” explains Prof. Anette-Gabriele Ziegler, director of the Helmholtz Munich Institute for Diabetes Research. As soon as therapy like Teplizumab is available in Europe, their research might help to find the perfect time point to start treatment. In their ongoing work, the researchers use machine learning and artificial intelligence to develop algorithms that are even more accurate and stable in predicting the progression of early-stage type 1 diabetes in children.