Periodic Reporting for period 1 - PRAESIIDIUM (PHYSICS INFORMED MACHINE LEARNING-BASED PREDICTION AND REVERSION OF IMPAIRED FASTING GLUCOSE MANAGEMENT)
Reporting period: 2023-01-01 to 2024-06-30
The incidence of undiagnosed diabetes accounts for 36% of all diabetes cases among European adults, while 541M adults worldwide live with levels of glucose “higher than normal but not higher enough” to be considered in the range of Type-2 Diabetes – also called “prediabetes” - which constitutes an important risk factor for further diabetes development. Prediabetes itself is not an extensively studied condition compared to the overt diabetes, but – most importantly - it is a condition that can be reversed starting with lifestyle intervention such as diet and physical activity, without medications to not proceed into diabetes.
The aim of the project is to develop a prototype tool for the real-time prediction of the prediabetic risk based on a series of patient-specific mathematical models (firstly developed during a previous Euroepan-funded project) that simulate metabolism, pancreas hormone production, microbiome metabolites, inflammatory process, and immune system response. The prediction algorithm will be based on a “physics-informed machine learning”. Datasets of real-life data will be combined with mathematical models to overcome the limits of a “black-box” machine-learning approach, while reducing the computational time for simulating the solutions of heavy mathematical models and improving its prediction performances.
The consortium will collect the necessary training data from already existing clinical studies and databases which are representative of the real-life scenarios of a prediabetes/diabetes risk insurgence in adulthood: family history, Metabolic Syndrome, Liver disease and morbid obesity. Two new pilot prospective observational studies will be also conducted, during which we will also equip the participants with wearable sensors (e.g. glucose monitoring, bioimpedance, heart rate, accelerometer) to obtain real-life and real-time data.
The consortium is committed to state-of-the-legislation principles for which concerns the fair enrolment of participants in the clinical studies through informed consent, protecting sensitive data and ensuring the transparency, reproducibility and explainability of the machine learning algorithms developed during the project lifespan.
At the end of the project, the algorithm will be implemented in a web-based application where doctors can insert, manage, and monitor data inputs from all the patients, who will be equipped with a mobile app able to collect data also from wearable sensors. While there are some already-available commercial solutions (e.g. web and/or mobile apps) for monitoring diet and physical exercises directed towards metabolic syndrome and/or diabetic people, as well as few commercially available wearable sensors for glucose monitoring, none of these applications are based on real-life data-trained prediction algorithm.
Improvement in prediabetes risk prediction and its progression represents the main goal of the project, therefore the significance of the project is summarised by the consequent reduction of mortality and complications, improvement of patient’s health status and quality of life, reduction of costs supported by individuals, health systems and businesses and an improvement of prediabetes risk management in time of crisis. At the same time, the project will create positive impacts on research community, governments, medical units or associations, and society as a whole.
The aim of the project is to develop a prototype tool for the real-time prediction of the prediabetic risk based on a series of patient-specific mathematical models (firstly developed during a previous Euroepan-funded project) that simulate metabolism, pancreas hormone production, microbiome metabolites, inflammatory process, and immune system response. The prediction algorithm will be based on a “physics-informed machine learning”. Datasets of real-life data will be combined with mathematical models to overcome the limits of a “black-box” machine-learning approach, while reducing the computational time for simulating the solutions of heavy mathematical models and improving its prediction performances.
The consortium will collect the necessary training data from already existing clinical studies and databases which are representative of the real-life scenarios of a prediabetes/diabetes risk insurgence in adulthood: family history, Metabolic Syndrome, Liver disease and morbid obesity. Two new pilot prospective observational studies will be also conducted, during which we will also equip the participants with wearable sensors (e.g. glucose monitoring, bioimpedance, heart rate, accelerometer) to obtain real-life and real-time data.
The consortium is committed to state-of-the-legislation principles for which concerns the fair enrolment of participants in the clinical studies through informed consent, protecting sensitive data and ensuring the transparency, reproducibility and explainability of the machine learning algorithms developed during the project lifespan.
At the end of the project, the algorithm will be implemented in a web-based application where doctors can insert, manage, and monitor data inputs from all the patients, who will be equipped with a mobile app able to collect data also from wearable sensors. While there are some already-available commercial solutions (e.g. web and/or mobile apps) for monitoring diet and physical exercises directed towards metabolic syndrome and/or diabetic people, as well as few commercially available wearable sensors for glucose monitoring, none of these applications are based on real-life data-trained prediction algorithm.
Improvement in prediabetes risk prediction and its progression represents the main goal of the project, therefore the significance of the project is summarised by the consequent reduction of mortality and complications, improvement of patient’s health status and quality of life, reduction of costs supported by individuals, health systems and businesses and an improvement of prediabetes risk management in time of crisis. At the same time, the project will create positive impacts on research community, governments, medical units or associations, and society as a whole.
The project activities and main achievements are summarised below:
(I) Physics-Informed Machine Learning: the machine-learning surrogate of the previous complex diabetes risk model based on organs' simulations, including a novel in-context learning paradigm for model-free system identification. Partners have also investigated via mathematical equations the effect of physical activity on blood glucose level via mathematical models.
(II) Deployment of the models: At the present, the app is intended as an electronic case report file restricted to data collection purposes only for projects’ clinical studies.
(III) Acquisition and generation of longitudinal datasets representative of prediabetes: given the difficulty in obtaining a relevant number of real-life data on prediabetes (type 2-diabetes is far more studied and data acquisition became far more standardized), the consortium adopted a multicentric approach of real-life data collection from already-existing databases including large studies on healthy subjects, data from metabolic syndrome patients and from morbidly obese patients (having the highest-risk of developing diabetes). Two pilot observational studies for data collection are also being conducted to follow up to 150 participants for four months. Each participant has been provided with the app for remote data collection and been equipped with sensors (e.g. continuous glucose monitoring and Fitbit wrist band).
(IV) Ethic objective: the consortium worked on the design of the first ethics self-assessment protocol. Healthcare professionals will be more involved in all steps of the development as possible, as the consortium unanimously reckoned that clinicians’ oversight (human-in-the-loop) is pivotal for future meaningful and safe decisions.
(I) Physics-Informed Machine Learning: the machine-learning surrogate of the previous complex diabetes risk model based on organs' simulations, including a novel in-context learning paradigm for model-free system identification. Partners have also investigated via mathematical equations the effect of physical activity on blood glucose level via mathematical models.
(II) Deployment of the models: At the present, the app is intended as an electronic case report file restricted to data collection purposes only for projects’ clinical studies.
(III) Acquisition and generation of longitudinal datasets representative of prediabetes: given the difficulty in obtaining a relevant number of real-life data on prediabetes (type 2-diabetes is far more studied and data acquisition became far more standardized), the consortium adopted a multicentric approach of real-life data collection from already-existing databases including large studies on healthy subjects, data from metabolic syndrome patients and from morbidly obese patients (having the highest-risk of developing diabetes). Two pilot observational studies for data collection are also being conducted to follow up to 150 participants for four months. Each participant has been provided with the app for remote data collection and been equipped with sensors (e.g. continuous glucose monitoring and Fitbit wrist band).
(IV) Ethic objective: the consortium worked on the design of the first ethics self-assessment protocol. Healthcare professionals will be more involved in all steps of the development as possible, as the consortium unanimously reckoned that clinicians’ oversight (human-in-the-loop) is pivotal for future meaningful and safe decisions.
The project has so far developed:
• Standardised protocol of clinical/biochemical data on the health-to-prediabetes transition.
• A novel in-context learning paradigm for model-free system identification
• A novel mathematical model of blood glucose level regulation via physical activity
• Discovery of new circulating biomarkers involved in metabolic syndrome and liver diseases
• An Ethics-by-design methodology
• Standardised protocol of clinical/biochemical data on the health-to-prediabetes transition.
• A novel in-context learning paradigm for model-free system identification
• A novel mathematical model of blood glucose level regulation via physical activity
• Discovery of new circulating biomarkers involved in metabolic syndrome and liver diseases
• An Ethics-by-design methodology