|
Abtin Ijadi Maghsoodi Information Systems and Operations Management Faculty of Business and Economics, Business School University of Auckland, Auckland, New Zealand |
|
Jessica Leung Discipline of Business Analytics The University of Sydney, Sydney, Australia Department of Econometrics and Business Statistics Monash University, Australia |
|
| Professor Soroush Saghafian Professor Soroush Saghafian is currently an Associate Professor of Public Policy (Management, Leadership, and Decision Making) at the Harvard Kennedy School, Harvard University. He is the founder and director of the Public Impact Analytics Science Lab (PIAS-Lab) at Harvard University, which is devoted to advancing and applying the science of analytics for solving societal problems that can have public impact. He also serves as a faculty affiliate for the Harvard Ph.D. Program in Health Policy, the Harvard Centre for Health Decision Science, the Harvard Mossavar-Rahmani Centre for Business and Government (M-RCBG), the Harvard Data Science Initiative, and is an associate faculty member at the Harvard Ariadne Labs (Health Systems Innovation). |
We recently had the opportunity to speak with Professor Soroush Saghafian from the Harvard University about the role of Operations Research (OR), Management Science (MS), and Machine Learning (ML) in Healthcare Analytics and Healthcare Policy Planning, and their impact on Diversity, Equity, and Inclusion (DEI).
Could you tell us what makes healthcare analytics an exciting area of research?
The short answer is that you can save a lot of lives. As such, the public impact that you can have in healthcare is huge. Healthcare is one of those areas where you can reach a lot of people and create a large-scale impact. Most countries have significant challenges in healthcare. In the US, healthcare is a mess. We spend about 20% of the GDP on healthcare, and the outcomes are not good compared to other developed countries. Healthcare, essentially, is an area where challenges are both complex and lingering.
What do you think are some of the biggest research and practical challenges right now in healthcare analytics?
At different levels, there are different questions. At the policy level, many governments across the world are struggling with improving healthcare as a sector. But at a slightly different level, we have issues with healthcare decision-making where physicians are trying to come up with the best treatments for patients and they are usually not able to do that because of many reasons. Analytics can play a vital role (along with advancements in technology), can enable physicians to make better decisions and assist policymakers to find superior policies.
How has healthcare analytics changed in the last ten years, and where do you see it going in the future?
One of the main things we are seeing these days is in terms of using smart devices and smart technology to keep track of patients. Mobile health (mHealth) technologies are rapidly developing. If you think about traditional ways of delivering care, you needed the patient to be in a doctor’s office or in the hospital. With mobile health devices, because now everybody has a smart device, you have the ability to reach them and monitor them to recommend treatments, even when they are busy with other activities. So, using smart devices, and sensors that can track patients wherever they are, we can allow physicians to come up with better decisions through what is called “continuous monitoring” (i.e. by supplying them with continuous information about their patients). Say, I’m sitting now too much for this interview; my phone may tell me that: “you should stand for your meeting" or “you should go for a 10-minute walk after the meeting". These are all enabled via technology, but the technology itself is not important unless there is useful analytics behind it. Using suitable algorithms along with detailed data can immensely reduce malpractices in the healthcare industry.
Do you see potential challenges in sharing that level of information with that level of transparency?
There are definitely challenges. We have just published a paper in the National Academy of Medicine, where we discuss the main scientific and regulatory challenges behind mobile health technologies. Because there are chances of data breaches and/or devices not working well with the hospital IT systems, there are several concerns that the governments are facing. For example, should they allow these mobile health apps to talk to each other? Even with the COVID contact tracing apps, there are tons of concerns regarding privacy and related issues that are still lingering. I think the regulatory challenges are still an important part of the picture. In addition, as we discuss in our paper published by the National Academy of Medicine, there are various scientific challenges that need to be resolved before these technologies can reach their full potential.
Some of your work has concerned the public policy within healthcare analytics. Could you please explain what is and what will be the impact of healthcare analytics on DEI?
Unfortunately, there are severe disparities in healthcare, not only among patients but also among hospitals. A particular piece of research that we did (now published in Operations Research) aimed at studying ways to make healthcare more transparent. For instance, when you are shopping for a car you can go to various websites and you can compare the quality of different cars. What we have found is that public reporting can help in terms of equity, but it can also hurt various other social outcomes if it is done incorrectly. So, what we recommend is the information should not be only publicly available but should also be publicly understandable so that an average person can digest it and make better choices.
What would you consider as an equitable decision model or decision system in terms of the allocation of scarce resources?
Let me give you a specific example regarding that. We have collaborated with the Massachusetts General Hospital trying to develop some algorithms for them to assist them in better allocating their new technology for treating cancer patients. This new technology is known as proton therapy, which is much more advantageous than X-ray therapy, in the sense that you can control the energy and release it once it reaches the tumor. Unlike X-ray, proton therapy does not ruin the surrounding parts of the tumor, because you have much more control over where and when to release the energy. Since the technology is very new, very few hospitals in the US have it. Massachusetts General Hospital is one of the hospitals that have the technology, and even them only have a few machines. That is, the resources are very scarce, but the demand for using this technology is incredibly high. We have been trying to help the hospital by developing algorithms that predict how much benefit a patient is going to gain by using this technology (versus the other options that are out there) and assign the capacity to patients who are going to benefit most from the technology. This is an example where you have very scarce resources with a lot of demand. Clearly, you have to make decisions. One of the possible ways is to think about who is going to benefit the most from it and give it to them. On the other hand, there are ethical issues with that. For example, consider a person who is 98 years old, and has applied to use this technology. The algorithm estimates that the benefit for this person is not much, because whether we give it or not, this person is going to die in two years. The question is: is it ethical for the algorithm to deny treatment because of that? The answer depends on your theory of ethics (e.g. utilitarian ethics, deontological ethics, or virtue ethics). But one useful and practical view is this: If there are two people, and one of them is going to gain more benefit than the other (all else equal), then, yes, give it to that person. However, don’t deny the treatment to somebody, just because they don’t have insurance or just because they don’t have the money to pay for it, or just because they don’t have the resources to get into the hospital. Some hospitals, unfortunately, have people who they cannot deny because they donate significant amounts of money to that hospital. I think it’s very important for these algorithms to try to remove those biases and make sure that scarce resources are used in the best way.
When you are training the algorithm, are there any rules overseeing the training sample such that it is a diverse and unbiased training sample?
One of the problems is that people think the algorithms are biased. But the algorithms are often, not biased, as they are just trying to optimize an objective function. What is problematic is that we train them on biased data, and then we expect them not to be biased.
So while we have decision making algorithms to help us with allocating scarce resources, it is also important to treat what goes into the algorithm and what comes out of the algorithm with cautious?
Absolutely. I discussed this in my class with my students. With the algorithms and with the analytics, you aim to remove bias, address inequality, and make more fair decisions. We are human decision-makers, and by nature, we all have hidden spots that make us biased one way or the other. With the algorithms, you have the ability to remove such hidden spots and that’s what’s nice about analytics and algorithms. On the other hand, you have to be careful to avoid training them on data that is biased.
How can we use analytics to improve the effectiveness and efficiency in public policy implementation?
I think this is another area where analytics can play a substantial role. When you have expensive resources, you must use them carefully and efficiently, and this is very important in public policy implementations. So a lot of effort is devoted into trying to make the system more efficient and to reduce waste. At the end of the day, the hope is that analytics can beat politics, such that how resources are allocated or investment are made are not much affected by the political views of the person who is in charge. Instead, they should be based on what we learn from the data, what we learn from evidence, what we learn from analytics, and what unbiased algorithms tell us.
If a student were interested in studying these fields, what recommendations would you give to them? How can they start the conversation? And, where to begin?
In different programs, we try to teach our students different things in terms of how they can try to improve the healthcare sector. So, I think they should go to the right program, read the right books and articles. I’m not just talking about scientific articles, but sometimes reading daily newspaper also helps students to see where the problems are and what the sector is essentially struggling with. These are all good exercises for students to understand the main issues. And when they are in a good Ph.D. program, they will have good advisors, good people to work with, and good resources, all of which allows them to do more impactful research.
One of the challenges coming from a background of industrial engineering and starting a healthcare analytics research program is not being familiar with the healthcare and medical dictionary. Do you have recommendations for students with manufacturing and industrial backgrounds?
Actually, that can be an advantage as you know the manufacturing settings. There are tons of things that have gone through the manufacturing processes to make them efficient. Unfortunately, the healthcare sector has been behind. So there is an opportunity to bring ideas from manufacturing to healthcare. A lot of people have been trying to do that as well to make the healthcare sector more like manufacturing settings where everything is standardized. I have been working with a lot of hospitals, and I have observed first hand that while you can get access to tons of datasets, there is not much data stemming from time studies. Time studies in manufacturing settings are ample. In contrast, in healthcare, we do not know how the most expensive resources, such as physicians, are spending their time. Every physician says “I’m super busy” and there are lots of patients waiting. People die in emergency rooms just because of long waiting times. Since there is no data telling us how exactly physicians are spending their time, it is not easy to offer solutions. In our research, for instance, we had to develop an iPhone application so that physicians can time themselves. Time studies like this in the manufacturing setting were done more than 150 years ago. I think the optimistic view is that it is actually an advantage to know how manufacturing processes work and try to bring useful ideas to the healthcare sector. At the same time, I think domain knowledge is incredibly important. One of the things I see students struggle with is that they do not have the domain knowledge. While you can bring the main ideas from manufacturing to healthcare, you have to understand physicians’ perspectives, and what they have to deal with. Have domain knowledge, and then you can bring new ideas and shape them so that they become implementable and useful in the healthcare sector. I think it is also essential for students to spend time in hospitals. When I was a student, I shadowed a lot of physicians; I spent a lot of time just looking at what physicians do, and it helped me better understand their work.
Is there anything else you would like to share with our readers, and in particular, the students who are thinking about healthcare operations research and healthcare analytics?
I would like to encourage them. People with an analytical background understand how to optimize and improve things and I think they can do a lot for the healthcare sector. I also encourage them to spend more time trying to understand the medical side of things. In essence, you have to be bilingual and understand both the analytics and the medical side. Once you become bilingual, create an impact.
Once again, we would like to thank Professor Saghafian again for sharing with us. We are sure our readers are inspired by these thoughts and will tremendously benefit from it. Thank you!
|