Since the world’s first insulin was injected into a human a century ago, efforts focused mainly on improving insulin treatment for diagnosed patients. Today, preventive therapy is being researched as the next milestone, and an increasing number of countries are discussing strategies to identify the target population for such treatments. To this end, existing data sets from observational cohort studies can be used together with known risk predictors such as islet autoantibodies, human leukocyte antigen (HLA) haplotypes and genetic risk scores. The challenge now is to translate this knowledge into effective screening strategies.
A recent “For Debate” article published in Diabetologia highlights the potential of artificial intelligence (AI) to transform the landscape of type 1 diabetes (T1D) screening and early diagnosis. The article summarises the position of several researchers and clinicians active in the field, who met in an open forum organised by the ASSET partnership, sponsored by the Swedish Innovation Agency. ASSET also emphasises the importance of AI for developing holistic screening strategies that extend beyond disease prediction.
From prediction to prevention
The ASSET initiative is taking a broad approach, focusing on AI to help inform screening programmes, test preventive therapeutics in a clinical setting, and evaluate the ‘implementability’ of such practices in healthcare systems,” the consortium sums up the outline of the project in their statement. As a first step, ASSET is building a machine learning pipeline based on data from the TEDDY cohort, which includes 8,640 high-risk children from birth to the age of 15 or with diagnosis of T1D. Once completed, their model will need to face an unseen data set and comparison with clinical outcomes to assess its sensitivity and specificity.
“AI has significant potential to revolutionise screening for autoimmune diseases, including T1D, by analysing vast amounts of data to identify patterns and risk factors that might not be apparent through other methods,” says Pedro F. Teixeira, first author of the paper. Gun Forsander, EASD member and co-senior author, adds: “Integrating AI in screening programmes could lead to more personalised monitoring strategies and enhance the operational feasibility, cost-effectiveness, and acceptance of such programmes at the population level. Early detection and individualised monitoring of affected individuals should prove effective in providing timely interventions, which may prevent or delay the onset of the disease.” Thus, a proactive approach driven by AI-based strategies should also be helpful in reducing T1D-related complications.
Four areas for AI in T1D prevention
To contribute to the vision of preventive diabetes care, ASSET has identified four areas where AI can be applied, outlining the broad scope from T1D screening to potential future treatment approaches: (1) risk stratification, (2) individualised follow-up programmes, (3) identification of predictive and treatment efficacy markers, and (4) optimisation of healthcare resources and cost-effectiveness of screening (figure 1).
Unlocking AI’s potential to screen for type 1 diabetes
Artificial intelligence is going to revolutionise also the medical field, especially when it comes to harnessing existing data sets for practical applications. A Swedish consortium has now published its approach to developing an AI-based population-wide screening strategy for type 1 diabetes. However, their proposal goes beyond mere disease prediction.
In this broad approach, AI can be used to identify individuals at risk of developing T1D or in the early stages of the disease by assessing genetic susceptibility, family history, environmental exposures, and behavioural factors. It can also predict disease progression, differentiate between slow and fast progressors, and provide personalised intervention suggestions for clinicians, thereby optimising the use of healthcare resources. Finally, AI may lead to the identification of new markers or provide valuable insights into clinical trial design and population selection, thus supporting research into predictive drug development. Taken together, the authors state that these multiple opportunities offered by AI-based screening design could help lay the foundation for a precision medicine approach to T1D.
Balancing gains and challenges
“Finding individuals at risk for the disease allows them to be monitored, to be involved in clinical trials of preventive therapeutics and, in the worst case, to be prepared for a diagnosis avoiding an acute clinical presentation,” the ASSET consortium highlights the benefits of timely screening. However, “both operational and ethical considerations need to be taken into account before implementing the proposed technology on a large scale,” Teixeira points out.
“The key issue,” says Åke Lernmark, EASD member and second co-senior author, “is whether AI would allow the research community, together with industry, to capitalise on large publicly available data repositories to design screening programmes not only for the early detection of individuals at high risk, but also to identify tailored preventive therapies.”
Overcoming the various challenges will require extensive collaboration between researchers, clinicians, and AI experts, as well as rigorous validation of AI models in order to gain acceptance not only from the target population but also from healthcare providers, regulatory authorities, payers and other relevant stakeholders, the authors conclude on the many dimensions taken on by their ambitious project.
Key Points:
- Researchers in the field of T1D are optimistic about the possibility of adding AI to the toolbox for predicting individual disease risk.
- The Swedish ASSET consortium, a collaboration of public and private experts, aims to implement screening strategies for T1D by using artificial intelligence (AI).
- The design of screening programmes including risk-predictive biomarkers, the periodicity of monitoring, organisation and resource allocation are tractable tasks when using AI.
- The application of AI to the several aspects of a screening programme is still very limited, and challenges remain.
Author: Hanna Gabriel, BA MSc. Any opinions expressed in this article are the responsibility of EASD e-Learning.