Dalla Lana School of Public Health

DLSPH faculty and students are invited to the Dalla Lana School of Public Health Data Science Methods for Population Health & Health Systems Research Seminar Series that will be held from February 7 to February 17, 2023. 

The seminars will be taking place in person and Zoom. Please see details below. 

Speaker	Date	Title	Research Seminar Series	More Information
Jude Kong	Tuesday February 7th, 2023	Leveraging Responsible, Explainable & Local Data Science Methods for Pop. Health & Health Systems	10:00 to 11:00am HS Building, 155 College Street, 4th floor, Rm. 412 or Zoom: https://utoronto.zoom.us/j/88997604927 Meeting ID: 889 9760 4927 Passcode: DLSPH2023	Feb. 7 – More Information
Alton Russell	Monday February 13th, 2023  *light lunch to follow	Informing targeted public health measures with data-driven decision modeling	9:00 to 10:00 am HS Building, 155 College Street, 4th floor, Rm. 412 or Zoom: https://utoronto.zoom.us/j/83251121498 Meeting ID: 832 5112 1498 Passcode: DLSPH2023	Feb. 13 (Alton Russell) – More Information
Yizhen Xu	Monday February 13th, 2023  *light lunch to follow	Longitudinal Causal Inference with Latent Variable Models	11:00 am to 12:00 pm HS Building, 155 College Street, 5th floor, Rm. 507 or Zoom: https://utoronto.zoom.us/j/85813090216 Meeting ID: 858 1309 0216 Passcode: DLSPH2023	Feb. 13 (Yizhen Xu) – More Information
Tianyuan Lu	Tuesday February 14th, 2023  *light lunch to follow	Leveraging Human Genetics for Improving Population Health: Advances and Challenges	11:00 am to 12:00 pm HS Building, 155 College Street, 6th floor, Rm. 650 or Zoom: https://utoronto.zoom.us/j/82982280291 Meeting ID: 829 8228 0291 Passcode: DLSPH2023	Feb. 14 – More Information
Zihang Lu	Wednesday February 15th, 2023  *light lunch provided	Integrating Multidimensional Longitudinal Features to Discover Disease Phenotypes: A Bayesian Machine Learning Approach	11:00 am to 12:00 pm HS Building, 155 College Street, 6th floor, Rm. 650 or Zoom: https://utoronto.zoom.us/j/85116778015 Meeting ID: 851 1677 8015 Passcode: DLSPH2023	Feb 15 – More Information
Jason Poulos	Thursday February 16th, 2023  *light lunch provided	Targeted learning in observational studies with multi‐valued treatments: An evaluation of antipsychotic drug treatment safety	11:00 am to 12:00 pm HS Building, 155 College Street, 6th floor, Rm. 650 or Zoom: https://utoronto.zoom.us/j/86854754638 Meeting ID: 868 5475 4638 Passcode: DLSPH2023	Feb. 16 – More Information
Camellia Zakaria	Friday February 17th, 2023  *light lunch to follow	Network-centric Mobile Sensing for Smart Health	11:00 am to 12:00 pm HS Building, 155 College Street, 4th floor, Rm. 412 or Zoom: https://utoronto.zoom.us/j/87273104736 Meeting ID: 872 7310 4736 Passcode: DLSPH2023	Feb. 17 – More Information

---

Jude Kong

Tuesday February 7th, 2023  11:00am to 12:00pm

Lunch with Faculty and Students  Light lunch will be provided  HS Building, 155 College Street, 4th floor, Rm. 412

---

Title: Informing targeted public health measures with data-driven decision modeling
Presenter: Alton Russell
Date: Monday February 13th, 2023
Time: 9:00 to 10:00 am
Location: HS Building, 155 College Street, 4th floor, Rm. 412 or Zoom: https://utoronto.zoom.us/j/83251121498
Meeting ID: 832 5112 1498
Passcode: DLSPH2023 

Abstract: Decision-analytic models can inform measures to address important problems in population health and health systems. Traditionally, decision analysts have focused on the aggregate or average impact of measures on a population. Increasingly, policy makers seek to understand the distribution of impacts across diverse populations. This is for two main reasons: to understand the equity implications of policies and to enable targeted public health measures. 

I will describe how integrating data-driven methods like machine learning into decision analysis can improve estimation of the impact of public health measures on diverse populations, informing targeted interventions. I will describe two applications: dispensing the overdose reversal drug naloxone to patients receiving prescription opioids and tailoring the frequency of blood donations to each donors’ estimated trajectory for recovering iron stores. 

W. Alton Russell, PhD, is an Assistant Professor in the McGill School of Population and Global Health and director of the Data-Driven Decision Modeling Lab, or D3Mod lab. The D3Mod lab aims to enable the efficient, effective, and equitable use of finite healthcare resources by developing, assessing, and applying traditional decision modeling methods (mathematical modeling, simulation, optimization) together with data-driven methods (machine learning, Bayesian statistics). Dr. Russell has developed decision analytic models and data-driven analyses to inform multiple areas of health policy and clinical practice, including blood donation and transfusion, managing pediatric kidney disease, opioid use disorder and overdose, and gastroenterology. Dr. Russell received undergraduate training in Industrial Engineering and Public Health at North Carolina State University, Masters and Doctoral training in Management Science and Engineering at Stanford University, and postdoctoral training at the Massachusetts General Hospital Institute for Technology Assessment and Harvard Medical School. 

---

Title: Targeted learning in observational studies with multi‐valued treatments: An evaluation of antipsychotic drug treatment safety
Presenter: Jason Poulos
Date: Thursday February 16th, 2023
Time: 11:00 am to 12:00 pm
Location: HS Building, 155 College Street, 5th floor, Rm. 507 or Zoom: https://utoronto.zoom.us/j/86854754638
Meeting ID: 868 5475 4638
Passcode: DLSPH2023 

Abstract: Antipsychotic drugs are the main treatment tool for those with serious mental illness, and some of the newer (second generation) antipsychotics carry a risk of cardiometabolic morbidity, such as diabetes. Much of the existing evidence on the relative safety of antipsychotic drugs come from observational studies that are limited by the use of regression-based methods and pooled comparisons. This paper provides the first doubly-robust estimates of the relative safety of specific antipsychotic drugs. We employ targeted minimum loss-based estimation (TMLE) with machine learning and demonstrate through extensive simulations an efficiency gain from estimating the correct multinomial treatment model. The standard TMLE implementation used in the literature estimates the treatment model through a series of binomial regressions. We evaluate the relative diabetes or mortality risk of non-random assignment to one of six commonly used antipsychotics within a cohort of adults with serious mental illness. We find a safety benefit of moving from a newer antipsychotic considered among the safest of the second-generation drugs to an infrequently prescribed older drug thought to pose a low cardiometabolic risk. 

Jason Poulos is a Postdoctoral Fellow in Data Science in the Department of Health Care Policy at Harvard Medical School. He received a PhD in Political Science with an emphasis in Computational Science and Engineering from UC Berkeley in 2019. He subsequently held a postdoctoral appointment at the Statistical and Applied Mathematical Sciences Institute (SAMSI), where he participated in the causal inference and deep learning programs. His research focuses on leveraging machine learning for improving causal inference in the social sciences and health data sciences. 

---

Title: Network-centric Mobile Sensing for Smart Health
Presenter: Camellia Zakaria
Date: Friday February 17, 2023
Time: 11:00 am to 12:00 pm
Location: HS Building, 155 College Street, 4th floor, Rm. 412 or Zoom: https://utoronto.zoom.us/j/87273104736
Meeting ID: 872 7310 4736
Passcode: DLSPH2023 

Abstract: The near future brings a bold promise of deep health personalization services and management – seamlessly integrating the Internet of Things (IoT) across individuals, communities, and the environment. However, two key challenges exist in Smart Health efforts in practice. First, personalized care and real-time interventions require holistic user models to process data efficiently and predict accurately. Second, model development must comprise heterogeneous data of individuals in the community. In this talk, I will highlight how employing network-centric mobile sensing of smart devices can overcome these challenges, particularly in inferring behavioral features suitable for predicting sleep and stress, as two common everyday health issues. In doing so, such systems can complement wearable sensing for longitudinal monitoring. I will conclude with a discussion on driving smart health with interconnected sensing systems that can provide more significant value-of-networked stakeholders. 

Camellia Zakaria is a Postdoctoral Research Fellow at the Laboratory of Advanced Software Systems (LASS), University of Massachusetts Amherst, USA, where she continues broadening her dissertation work of developing network-centric sensing systems and behavioral analytics for health applications. The orientation of her research is as affirmative as it is critical in fulfilling the long-term vision “toward revolutionary smart health – one which offers deep personalization to an individual’s health but integrates personal health to drive population and public health efforts.” Grounded in Systems research, her investigations lie in the intersecting subfields of Applied Machine Learning and Human-Computer Interaction. Her research publications are in top-tier conferences such as ACM IMWUT(UbiComp), CHI, and CSCW. Zakaria received her Ph.D. in Information Systems at the Singapore Management University, School of Computing and Information Systems, Singapore. More information about Zakaria can be found at https://nczakaria.github.io/. 

---

Title: Integrating Multidimensional Longitudinal Features to Discover Disease Phenotypes: A Bayesian Machine Learning Approach
Presenter: Zihang Lu
Date: Wednesday February 15^th, 2023
Time: 11:00am to 12:00pm
Location: HS Building, 155 College Street, 5th floor, Rm. 507 or Zoom: https://utoronto.zoom.us/j/85116778015

Meeting ID: 851 1677 8015
Passcode: DLSPH2023

Abstract: Clustering longitudinal features is a common research goal in health studies to identify distinct developmental patterns that reflect disease phenotypes and to facilitate targeted intervention. Compared to clustering a single longitudinal feature, integrating multiple longitudinal features allows additional information to be incorporated into the clustering process, which reveals co-existing developmental patterns and generates deeper biological insight. In this talk, I will discuss a newly developed Bayesian machine learning approach and software package for clustering multidimensional longitudinal features with complex data structures. Results from analyzing large Canadian birth cohort data to discover respiratory phenotypes will be presented and discussed.

Zihang Lu is a tenure-track Assistant Professor in the Department of Public Health Sciences at Queen’s University. He completed an MSc and PhD in Biostatistics from the University of Toronto. Dr. Lu’s research focuses on Bayesian statistical and machine learning methods motivated by complex health data. His research is supported by the Natural Sciences and Engineering Research Council of Canada, the Canadian Statistical Sciences Institute, and the Canadian Institutes of Health Research.

---

Title: Longitudinal Causal Inference with Latent Variable Models
Presenter: Yizhen Xu

Date: Monday, Feb. 13, 2023
Time: 11 am to 12 pm
Location: HS Building, 155 College Street, 5th floor, Rm. 507 or Zoom

Date: Monday February 13 2023
Time: 11:00am to 12:00pm
Location: HS Building, 155 College St., Room 507 or Zoom:https://utoronto.zoom.us/j/85813090216

Meeting ID: 858 1309 0216
Passcode: DLSPH2023

Abstract: Dynamic prediction of causal effects under different treatment regimes conditional on individual’s characteristics and longitudinal history is an essential problem in precision medicine. This is a challenging problem in practice because outcomes and treatment assignment mechanisms are unknown in observational studies, individual’s treatment efficacy is a counterfactual, and the existence of selection bias is often unavoidable. Latent structural models explain observed correlations by making assumptions about the latent causes of the variables; however, when comparing dynamic treatment regimes, the causal interpretation of its coefficients may not be marginal in many cases.

In hope to address some of these challenges, we relax the no selection bias assumption and account for the unobserved heterogeneity in both the treatment assignment and biomarker dynamics using a multivariate generalized linear mixed-effects models and develop a longitudinal causal framework that sequentially updates the latent time-invariant factors, with the method applied to study the effectiveness of an immunosuppressant medication for scleroderma.

Yizhen Xu is currently a postdoc in biostatistics at the Johns Hopkins University, her research interest is in Bayesian latent variable modeling, causal inference, and machine learning. She graduated from the Brown University in January 2020 with a thesis on methodologies in ensemble learning, Bayesian trees, and longitudinal causal inference.

---

Title: Leveraging Human Genetics for Improving Population Health: Advances and Challenges
Presenter: Tianyuan Lu

Date: February 14, 2023
Time: 11:00am to 12:00pm
Location: HS Building, 155 College Street, 6^th floor, or Rm. 650 or Zoom: https://utoronto.zoom.us/j/82982280291 Meeting ID: 829 8228 0291; Passcode: DLSPH2023

Abstract: Since the first human genome sequence was completed in 2003, multi-trillion dollar investments have been made with the goal of utilizing human genetics to revolutionize healthcare systems and improve population health. Despite the identification of numerous genetic variants associated with various health outcomes, there still remain challenges in translating these findings into effective healthcare policies and practices.

In this talk, I will discuss key aspects of my research aimed at leveraging human genetics to improve healthcare for complex diseases. First, I will showcase the development and potential clinical utility of genetic risk scores, with a focus on population-level risk screening and stratification. As an example, I will highlight a genetic risk score for height that, when combined with clinical risk factors, can accurately identify children at risk of adulthood short stature and guide growth-promoting therapies. Second, I will present our efforts to identify risk factors and potential therapeutic targets through genetics-facilitated causal inference methods. I will discuss our discoveries of molecules in circulation, such as proteins and metabolites, that could be used as biomarkers for psychiatric disorders. Finally, I will discuss several limitations and future directions of this field, with an emphasis on mitigating health and healthcare disparities in diverse populations.

Tianyuan Lu is a Senior Research Scientist at 5 Prime Sciences Inc. and Lady Davis Institute for Medical Research at the Jewish General Hospital. He obtained his PhD in Quantitative Life Sciences from McGill University in 2022. He has been leading research programs in developing and implementing statistical genetics and genetic epidemiology methods for improving healthcare for complex diseases in diverse populations. His research has been supported by multiple fellowships and awards, including a Schmidt AI in Science Postdoctoral Fellowship, a Vanier Canada Graduate Scholarship, and a Fonds de Recherche du Québec – Santé Doctoral Fellowship.