Building the foundation — how statistical methods shape health outcomes: A Q&A with Dr. Peter Austin
“The best thing about being a statistician is you get to play in everybody else’s backyard.”
Dr. Peter Austin is a senior scientist with ICES and Sunnybrook Research Institute. He is also a full professor in the Institute of Health Policy, Management and Evaluation at the University of Toronto. In 2024 Peter was named a highly-cited researcher by Clarivate Analytics (formerly Science Citation Index – Web of Science), an honour he has received for the past decade. Only 6,886, or 0.1%, of the world’s scientists and social scientists earned this distinction last year.
We sat down with Peter to talk about his career and what being named a highly-cited researcher means to him.
Q: Can you give us a 30-second elevator pitch describing your area of research?
A: My research focuses on developing and evaluating statistical methods for analyzing large administrative databases. I conduct research in methods for a number of areas: estimating the effects of treatments and exposures on patients’ health outcomes; predicting the risk of specific health outcomes, such as death, for individuals with heart disease; following individuals over time and identifying factors that increase or decrease the risk of specific outcomes occurring over time; analyzing data that have an inherent clustered structure, such as patients clustered or nested within hospitals; and addressing the presence of missing or incomplete data.
Q: What motivated you to pursue a career in statistical methodology?
A: While completing my PhD in mathematics, I came to the realization that I didn’t want to pursue a career as a mathematician. Mathematical research is often a solitary pursuit. Instead, I wanted to use my quantitative skills to work collaboratively with others to address concrete problems and issues in society.
Q: You’ve been with ICES since the early days of the organization – what are some key changes you’ve witnessed in the field of health research and data analytics?
A: Two key changes that I’ve observed are the increase in breadth and depth of data and in the sophistication of the analytic methods used in studies. When I first arrived at ICES, I believe that our data holdings were limited to the CIHI DAD, OHIP, RPDB, and ODB, and I seem to recall that we had just received ODB shortly before my arrival. The analytic methods in health services research tended to be relatively simple, with most analyses being no more complex than a conventional logistic regression model or a Cox proportional hazards model. Both the range of statistical methods and their complexity have increased substantially over time.
Q: What does being named a highly cited researcher (10 years running!) mean to you?
A: It is gratifying to see that my research has had an impact on the research of others and that my methodological studies are being incorporated into their work.
Q: What’s the paper or project you’re most proud of? And why?
A: I don’t think that I can identify a specific paper or project which I’m most proud of. I would group my research into different themes. Of those themes, I’m most proud of my research on propensity score methods. This is the area in which I’ve published the most and have made a solid contribution to that field over a sustained period of time. Many of these papers are highly cited, highlighting that other researchers value my contribution to this area of research.
Q: Statistics and methods don’t often get the same attention or recognition as other types of research, but they are integral to the work ICES does. How does your research impact the health of Ontarians?
A: I think that a helpful analogy is to think of statistical research as performing a function similar to that provided by civil and structural engineers in society. Provision of clean water is something that we often take for granted but which is necessary for a healthy and flourishing society. Similarly, the foundations of large buildings are hidden from sight, but the strength and stability of the foundations are essential to the stability of the entire building. In a similar way, research on statistical methodology provides a solid foundation on which to conduct applied ICES research. My research impacts the health of Ontarians by facilitating the research of applied health researchers and by ensuring that correct conclusions are drawn from the analysis of ICES data.
Q: As you look to the future, what are you most excited for in the field of statistical methodology? And what are you most concerned about?
A: I’m not very good at forecasting the future and am reluctant to predict where the field is moving. Science often makes leaps forward, but it can be difficult to predict where these advances will be made. What we do often see is the development of new methods to address new problems and issues that arise. If I had to make one observation, it is that modern statistical methods have grown in tandem with the increases in computing power that we’ve experienced over the past several decades. I think that as we continue to see increases in computing power, we will continue to see increases in the complexity of the statistical analyses that we can conduct. That is particularly important when considering the size of ICES’ data holdings and the anticipated growth in the size and complexity of these data holdings.
One thing that I am cautious about is our tendency to jump on bandwagons and think that a new method or technique is going to solve all our problems or be applicable everywhere. One can see this to some extent with machine learning methods. Machine learning methods offer very strong advantages in the analysis of image and text data. Whether they offer a consistent advantage in the analysis of rectangular static datasets is debatable.
Q: What advice would you give to someone just getting started in this field?
A: The famous statistician John Tukey said that the best thing about being a statistician is you get to play in everybody else’s backyard. One of my former professors, using a sporting analogy, said that it’s more important what team you play on than what stadium you play in. My advice would be five-fold: one, get a good foundation in mathematical and applied statistics; two, take every opportunity to get experience in applied data analysis; three, strengthen your statistical programming skills; four, decide on a ‘backyard’ that you would like play in, since it is the rare statistician who can play in everybody’s backyard; and five, find and join a team of people who are asking good questions and who are doing excellent research.