Emerging Uses of Data Science and AI: DLSPH Kicks Off First Interdisciplinary Seed Projects
Six teams of interdisciplinary researchers have won seed funding from DLSPH to investigate new and emerging projects in data science and artificial intelligence.
The award is part of the School’s commitment to a new research model that brings together broad teams with disciplinary depth to tackle some of the most complex problems in public health and health system performance.
“To grasp the revolutionary opportunities of new technologies such as AI, public health must change,” says Prof. France Gagnon, DLSPH’s Associate Dean of Research. “We must create funding structures that bring together depth and breadth of knowledge and reward collaboration. Interdisciplinary research clusters are a first step.”
The teams include researchers from DLSPH and affiliate institutions with a special emphasis on trainees and new investigators. Each was awarded $20,000. The DLSPH Data Science Interdisciplinary Research Cluster supported five projects, and a sixth – with a focus on child health equity – was funded through a generous partnership with the Edwin S.H. Leong Centre for Healthy Children.
The winning projects:
Using decision tree machine learning to identify worker movement typologies
Under the leadership of DLSPH Assistant Professor Aviroop Biswas, with co-investigators Kathleen Dobson (PhD Candidate, DLSPH), Stephanie Prince Ware (Research Scientist, Public Health Agency of Canada), Faraz Shahidi (Postdoctoral Fellow, Institute for Work and Health) and Peter Smith (Associate Professor, DLSPH), this project will explore the many and complex variables that affect working Canadians’ ability to engage in physical activity. Using the Canadian Health Measures Survey, the team will apply machine learning to over 11,000 respondents’ movement data, cardiometabolic markers, and socioeconomic and physical environment information. The work will determine if an AI approach offers different findings from those obtained through traditional statistical methods, and aims to identify groups in the population that might most benefit from targeted interventions.
A multicentre database of patients hospitalized with diabetes in Ontario across 30 hospitals
DLSPH Master’s student Michael Colacci, together with clinician scientists Michael Fralick and Fahad Razak from University of Toronto’s Faculty of Medicine and eight co-investigators representing five different affiliate hospitals, will link data from the Ontario Diabetes Database and the General Medicine Inpatient Initiative platform (GEMINI database) to create a cohort of approximately 200,000 people who have been hospitalized with diabetes in Canada. The new cohort will offer an incredibly rich source of data to evaluate the nature of inpatient care for diabetes – a disease affecting one in seven adults in this country. The team will use the new database to evaluate which proportion of patients received the most up-to-date care, and apply an artificial intelligence-based prediction model to identify which groups of patients might be at highest risk of negative outcomes.
The evolution of potentially inappropriate prescribing in persons with dementia
DLSPH PhD student Abby Emdin and Professor Susan Bronskill (DLSPH, IHPME and ICES), together with six co-investigators with expertise across geriatric medicine, epidemiology and rehabilitation sciences are collaborating on a project that will apply network analysis to linked health administration data to understand prescription patterns for older adults newly diagnosed with dementia. Using a study population of approximately 25,000 people in Ontario diagnosed between 2015 and 2019, the work will aim to identify the prevalence and circumstances related to patterns of over-medication. The findings may be used to improve health system efficiency by detecting medication overuse, and also prevent inappropriate prescribing – and associated negative health outcomes – among people with dementia. The use of network analysis methodology to patterns of drug prescribing is virtually unexplored, and this project will help identify the utility of this tool for future research in the field of pharmacoepidemiology.
Building Fair Machine Learning Models: Using Big Data to Explore Inequities in Risk Assessment at the Centre for Addiction and Mental Health
Sean Hill, Director of CAMH’s Krembil Center for Neuroinformatics, together with DLSPH Assistant Professor Daniel Buchman and five co-investigators with expertise across multiscale modeling, experimental and emergency psychology, translational science and bioethics, will evaluate whether biases might be built into the way risk assessment takes place in the context of psychiatric care. Recognizing that both conventional risk assessment and different machine learning tools may be subject to bias, the team will use a mixed methods approach that combines an analysis of Electronic Health Records, participant observation, and interviews with clinicians and patients, to identify how ML tools might amplify existing inequities, such as racial bias, because they are trained on biased datasets. Findings from the project will provide critical context and pilot data for future work using machine learning to redress these biases through fairer models.
ExplAIn 2 Kids: Engaging Children & Youth in Ar@ficial Intelligence in Pediatric Healthcare
In this project supported by the Edwin S.H. Leong Centre for Healthy Children, DLSPH Assistant Professor Melissa McCradden, together with co-investigators Dolly Menna-Dack (Holland Bloorview Kids Rehabilitation), Randi Zlotnik Shaul (SickKids), James Anderson (SickKids) and Elizabeth Stephenson (SickKids), will explore how children and youth understand and feel about the use of artificial intelligence in healthcare. The work aims to remedy the fact that there currently exist no explorations of the views of children and youth about the tools employed in their own care. The group brings expertise across qualitative health research, youth engagement, explainable AI, Bioethics, health law and research ethics. The project will generate new guidelines for clinical discussions with children about AI-assisted decision-making, create template language for informed consent forms for AI research studies, and launch a new agenda for healthcare AI engagement for children and youth.
A Predictive Model for the Presence of Occult Cancer in Family Practice Patients
DLSPH Professor Steven Narod, together with PhD candidates Vasily Giannakeas (DLSPH) and Victoria Sopik (Institute of Medical Sciences), and Assistant Professor Jennifer Brooks (DLSPH) aim to generate findings that might help develop a tool which can be used to predict the presence of cancer based on routine complete blood count tests that can be performed in any family doctor’s office. In earlier work, this group linked data from the Ontario Laboratory Information System and the Ontario Cancer Registry to analyse the risk of cancer occurrences in over nine million adults in Ontario; they found that high platelet counts were associated with a more than five-fold risk of a solid cancer diagnosed within six months of the blood test. In this new project, the team will evaluate more measures from original blood tests, incorporate additional variables such as age, sex and ethnicity, and apply statistical learning to improve the predictive values of their findings. The goal will be to create a tool that researchers might subsequently evaluate in prospective studies.
Learn more about DLSPH’s commitment to interdisciplinary research.