“I love statistics and like that it permits so much freedom”: An interview with Susan Murphy

Susan Murphy is the H.E. Robbins Distinguished University Professor of Statistics & Professor of Psychiatry at the University of Michigan. She obtained her B.S. in Mathematics from the University of Louisiana and her PhD in Statistics from the University of North Carolina at Chapel Hill.

She is known for her work applying statistical methods to clinical trials of treatments for chronic and relapsing medical conditions, in particular in relation to the SMART model, which stands for Sequential, Multiple Assignment, Randomized Trial, whereby it is assessed when intervention is best for the patient and at what level.

Her current primary interest concerns clinical trial design and the development of data analytic methods for informing multi-stage decision making in health. In particular for (1) constructing individualized sequences of treatments (a.k.a., adaptive interventions) for use in informing clinical decision making and (2) constructing real time individualized sequences of treatments (a.k.a., Just-in-Time Adaptive Interventions) delivered by mobile devices.

In 2013, Professor Murphy was awarded the MacArthur Fellowship or ‘Genius Grant’. She has contributed to numerous publications and regularly gives lectures from the Bradley Lecture to the Wald lectures at last year’s Joint Statistical Meetings in Seattle.

Statistics Views talks to Professor Murphy about her career and the fascinating research that has gained her worldwide recognition as a statistician.

 

1. When and how did you first become aware of statistics as a discipline and what was it that inspired you to pursue a career in statistics?

I had the opportunity to study mathematics in Germany and while I was there, I took a class on stochastic processes and I just loved the material. I wasn’t aware of other areas of applied mathematics and I was looking for ways to use maths in a way that might be applied to real life. This class made me realise that statistics for stochastic processes was definitely something that I wanted to pursue.

2. Today there are far more women in statistics but back when you were a student, were there many other women in your class?

Not really, nor were there many Americans. There were hardly any women, almost all foreign guys! There were a few women from Asia; but pretty much no one from the US or women, they were equally rare!

3. You are currently Professor of Statistics, Professor of Psychiatry and Research Professor at the Institute for Social Research at University of Michigan. Over the years, how has your teaching and research motivated and influenced each other? Do you continue to get research ideas from statistics and incorporate your ideas into your teaching?

Certainly I incorporate research ideas into teaching. A long time ago, I used to work on semiparametric models and in teaching I would bring in ideas underpinning maximum likelihood gained from my work in this field, even into my undergraduate teaching. Of course, work with graduate students does inform my research. Whatever I am working on, I try and teach from that point of view and I think I am a better teacher for it.

4. In 2013 you won a “genius grant” from the MacArthur Foundation for developing the SMART model, which can be used to customize treatment regimens for people with chronic or relapsing disorders. How did you learn that you had won?

It was a big surprise. I was at work and I got this phone call and the caller stated that she was calling from the MacArthur Foundation; immediately I am thinking, “They want me to write a reference letter for someone.” But instead she said I had won a MacArthur Fellowship! I was completely surprised and didn’t anticipate it at all. I don’t know how I came to be considered. It was a shock and changed my life immediately.

5. Have there been any further developments using the SMART model since the research was first published?

Many colleagues and former collaborators as well as former students are working in this area. There are still many open problems with much room for scientists to make advances in how to use the data to inform better clinical decision-making.
One of my colleagues has been working with researchers on serious mental health. They are conducting a cluster-randomised SMART which is very interesting and new. The goal of this trial is to construct an effective sequence of “implementation” interventions. Implementation interventions are interventions that are provided to clinics to encourage the adoption of evidence-based treatments by the clinicians in the clinics In this trial, the units are clinics and a clinic that does not respond to the initial low-cost intervention is re-randomised to a more intensive implementation intervention. There is enormous amount of work to be done on how to analyse data from this kind of trial. There is a lot of room for this kind of trial across different areas of health.

I think an important difference between a SMART clinical trial design and the kind of design most are familiar with (a two-arm randomized control trial design), is that the SMART is a factorial design. Factorial trial designs are used to develop an intervention package, in this case a sequence of interventions, as opposed to the two-arm randomized trial design which has the purpose of evaluating an intervention package. There is much confusion about factorial designs in the health field, likely because there has been some use of factorial designs in drug trials for evaluating multiple treatments. Our use of the SMART factorial design is more similar to the use of factorial designs in agriculture and manufacturing.

The SMART trial is really about building an intervention package, not about evaluating one, which is a different ball game.

6. What are you focussing on currently and what do you hope to achieve through your research?

All the work I do now is in mobile health. I haven’t completely left the world of SMART design – I still have a lot of collaborators as explained and I also advise young clinical scientists whom I mentor.

Mobile health is a setting in which treatments are delivered via a mobile device like a wristband or a smartphone. Statistically this is a sequential decision problem involving sequences of treatments just like in the SMART, but in mobile health, many of the treatments are treatments designed to help you help yourself. Now the sequential decision problem is more real time, for example, what sequence of treatments can help support people as they try to quit smoking? Stress is a big factor in relapsing e.g. is there some kind of support we can offer via a mobile device to help people manage stress and thus prevent smoking relapse? Other areas include physical activity e.g. if someone has had a heart attack, one of the big aims is to improve their physical activity in order to prevent another heart attack – how can we provide activity ideas in real-time to help the person be less sedentary and more active?

This is probably one of the most exciting times in statistics in perhaps 50-60 years and it is all because of the new ways to obtain and analyse data that we never could have before. New ways of collecting large amounts of complex data are being discovered at a very rapid pace. This requires that we, statisticians, develop methods to analyse the data so that the data is maximally useful to society.

7. Your lectures during JSM 2015 were the Wald Lectures. Could you please tell us about your chosen theme for the lectures?

All three lectures were about mobile health and sequential decision making. The first lecture was about clinical trials designs, called micro-randomized trials, that we can use to develop mobile health interventions such as when and where to provide activity recommendations. If your calendar on your smartphone indicates that you are busy then perhaps this isn’t the best time to call you to go for a walk. Indeed such a message when you are busy might aggravate you. So we have to figure out how to collect data to discover the time and setting in which to intervene that encourage rather than annoy you. The second lecture was about how we can analyse this type of data to help us address questions, such as the calendar question I just provided or does the weather have an impact on whether you respond positively to an activity recommendation? If different sensors indicate that you are highly stressed, it is may be best not to ping you and suggest you manage your stress in this way. The third lecture was about the development of data analysis methods that would run in real-time as you experience the mobile intervention so as to personalise the intervention to you. For example say we both want to increase our physical activity level but for whatever reason, you respond well to activity suggestions even when you are stressed, but I don’t. So it may be ok for you to receive activity recommendations when you are stressed but not me.

My main point in these lectures was to indicate how open mobile health is and due to the collection and analysis of real-time data, how absolutely critical it is that statisticians come into these fields. The field of mobile health is dominated by engineers and computer scientists and of course, many behavioural and clinical scientists but this is an area that really needs statistics due to need to provide measures of confidence that other disciplines cannot offer.

8. What would you say to encourage young people to follow a career in statistics?

I love statistics and like that it permits so much freedom. For example, early in my career, I worked mainly in mathematical statistics. Now I work mainly in machine learning and experimental design. I like that freedom and the ability to use mathematics in ways that can benefit society. Statistics provides us with a way to do that. I think if young people want to use quantitative skills in a way that can benefit society, statistics is an ideal career.

9. What do you think the most important recent developments in the field have been? What do you think will be the most exciting and productive areas of research in statistics during the next few years?

Whenever our colleagues figure out new ways to collect data– such as the collection of data on us as we go about our life– a whole series of new interesting problems opens up in our field. That is going on now with the collection of “Big Data”– in medicine on a molecular level, in outer space we can collect large amounts of data, in mobile health we collect data on how we go about our everyday lives, on the Web, people collect data on how we utilise the internet and search – all provide challenges to statistics. This is probably one of the most exciting times in statistics in perhaps 50-60 years and it is all because of the new ways to obtain and analyse data that we never could have before. New ways of collecting large amounts of complex data are being discovered at a very rapid pace. This requires that we, statisticians, develop methods to analyse the data so that the data is maximally useful to society.

10. What do you see as the greatest challenges facing the profession of statistics in the coming years?

I think the greatest challenge is connected to the most exciting development – everyone across the fields of computer science, engineering, humanities, social, behavioural and natural sciences recognise that we have all these new ways to collect data, to measure actual things going on in real life – thus the use of statistics is pervasive throughout all these fields. This is good as scientists in these fields want to learn more about statistics but the challenge for us is to take part, stay abreast of the new developments and to play a leadership role.

11. What is the best book on statistics that you have ever read?

Cox and Hinckley’s Theoretical Statistics which is a beautiful book that is not much used nowadays but the intuition makes it incredibly good. That book made me a big impression on me.

12. Who are the people that have been influential in your career?

The first was a high school teacher, Mrs. Bacala, who was an excellent mathematics teacher and the clarity of her lectures was fantastic. When I went to graduate school, I was really fortunate to take courses from Prof. Leadbetter who works in stochastic point processes and he too lectured with incredible clarity and precision. To teach with clarity and precision means a great deal to me. Consequently both of these individuals have been very influential in my career. Prof. Leadbetter also helped me after I left graduate school in terms of making important contacts.

When I look for someone to whom I want model my career after, I think of the great statisticians of our world like Sir D. Cox, A. Wald, P. Bickel and B. Efron; these people not only worked in the theory concerning the foundations of statistics but they also tackled real problems. Very inspiring!

13. If you had not got involved in the field of statistics, what do you think you would have done? (Is there another field that you could have seen yourself making an impact on?)

I don’t know, I really like statistics! I am kind of a one dimensional person but if I had done anything different, it would have been in applied mathematics because mathematics is such a beautiful way to communicate with clarity and precision. Statistics allows me to use mathematics in a way that might improve our world and that is the reason it is particularly attractive to me. The combination of mathematics and the potential of improving our world are incredibly motivating and what I want to do.