Grant: Operating Grant: Data Analysis Using Existing CDN Databases and Cohorts

Sleep disorders are a common concern for Canadian children. Such disorders can be associated with cognitive, behavioural, emotional and physical health problems. To diagnose such problems, clinical sleep studies are the gold standard method. However, sleep studies are resource-intensive, have very limited capacity, and therefore lead to long wait times ranging between 8 months to 3 years for Canadian children referred to the sleep clinic. These wait times result in critical delays in diagnosis especially for infant and toddlers in early stages of development, and may have other health consequences. Recent advances in machine learning and sensor technology may help reduce the clinical burden by determining a simplified sleep study setup as well as developing automatic algorithms for sleep data labeling. As the first step towards this long-term goal, we need to identify the minimum data requirements for sleep analysis. Our research team will access and analyze clinical sleep study data from thousands of patients available from the BC Children's Hospital. We will train machine-learning algorithms to classify sleep stages and detect common sleep disorders. We will compare the performance between algorithms trained with full sleep study data and reduced data channels, then optimize to identify the minimum data required to achieve sufficient accuracy for clinical diagnostics. Through this data-driven investigation, we will gain insight into potential ways to reduce the complexity of clinical sleep studies and generate preliminary automatic algorithms to reduce clinical burden. In a future step, we can apply these insights to develop accessible home-based systems for screening of sleep disorders outside of the clinic, which will substantially reduce wait times. In the long term, this line of research will help ensure that children with potential sleep disorders can be assessed in a timely and efficient manner.