Skip to navigation Skip to main content Skip to footer

Approved Research

Early identification of chronic non-communicable diseases and disorders using machine learning

Principal Investigator: Dr Qingqing Mao
Approved Research ID: 94816
Approval date: November 11th 2022

Lay summary

Chronic diseases, also referred to as non-communicable diseases (NCD)s, are on the rise globally. NCDs can manifest with a wide range of symptoms and outcomes varying by the origin of the condition (neurodevelopmental or physiological). We will examine individual disorders from each of these categories (neurodevelopmental and physiological). Chronic physiological NCDs, such as cardiovascular disease and cancer are responsible for 4.1M-<18M deaths annually (71%) worldwide. Chronic neurodevelopmental NCDs such as autism spectrum disorder (ASDĀ  impact more than 1 in 100 children worldwide. Like the diseases of physiological origin, neurodevelopmental NCDs can have severe repercussions such as social and cognitive deficits that impact functioning and quality of life. Research has indicated that ASD contributes to premature mortality at a rate of 2-10 times more than individuals without ASD. Timely diagnosis and treatment can improve outcomes for NCDs of multiple origins. However, diagnosis is often a complex process. Machine learning (ML) holds the potential to fill a gap in care by providing timely assessments based on readily available patient data to predict disease onset or complications, outcomes, and guide treatment approaches.

ML is a form of artificial intelligence that can use readily-available patient data from electronic health records (EHRs) to provide an assessment or prediction on health outcomes. ML has delivered high predictive power for both acute and chronic health conditions and often demonstrates improved performance over standard risk assessment tools used by clinicians. The use of readily available data makes ML ideal for employing in resource-scarce clinical settings, as it can be incorporated into most EHRs and does not require any inputs by the clinician. Thus, it does not interrupt the clinical workflow. We intend to develop, train, and test ML models to separately examine multiple distinct NCDs of physiological and neurodevelopmental origins. Our goals are: 1) uncovering the relationship between genetics and NCDs; 2) predicting NCD onset or complications; 3) diagnosing NCDs; 4) classifying NCDs by severity, stage, or subcategory. Through our work, we will contribute to advancing the understanding of how ML can be an important clinical tool for NCDs of neurodevelopmental and physiological origin. Due to ML accessibility and use of personalized data, this may facilitate earlier diagnosis, which may lead to timely initiation of treatment and improved patient outcomes.