What is data science? Part two

At UrbanTide we always talk about the smart city, and health and social care being a core component in delivering smarter cities. Do you think there’s any unique barriers or opportunities in health and social care, when it comes to data science? So I think if you look at smarter cities and health and social care, there’s a great opportunity there. Because we have health care data that’s routinely collected by hospitals or GPs. We have data that’s collected from our phones, from our cars, from ourselves– if you have wearables or trackers. And we also have a huge wealth of social care data about people that live at home independently when they’re older, for example.

But that opportunity is also a barrier, because those different organisations hold the data in separate data structures– basically in separate data places. And that’s a difficulty for harnessing machine learning. Because you have to link the data in order to be able to look at it in interesting ways. And I don’t think that’s a technical barrier. I think it’s something we can actually do something about really easily. I think what we need to do is to have really good data sharing agreements, and protocols that allow ease of sharing the data between the different systems. If we did that, the machine learning and the data science would then allow us to do a lot more interesting things around smart cities and health.

Yep. I totally agree. That’s brilliant. Data sharing agreements, definitely. OK. I’ll start with the barriers. Obviously you don’t want to be able to identify someone specifically, that they’ve given data. And that becomes really a challenging problem. Because once you actually correlate different data from devices, from health trackers, et cetera, that you start to easily identify actual people. And then the question obviously becomes about the privacy. How can you protect that. And there’s some challenge that needs to be addressed pretty much over the last five, ten years. So with regards to opportunities, obviously there are many opportunities. There now many medical devices used.

And using these– for example, variable ones, and that obviously you have some more regular health checks, and so on. And using all these different streams of data, you can identify some things much more conclusive than just one component. So that’s one of the opportunities that needs to be still addressed. So if you look at telehealth and telecare, for example– so that’s more in the social care arena. You’ve got lots of people who have devices in their homes, so fall alarms and pendants.

And one of the things you can do is, using all that data, collecting patterns of that data– if you had the power of data science to look at that in more interesting ways, you can actually do risk stratification. So you can start to identify the people, and the types of people that are more or less at risk, and intervene not just only at the right time but for the right people– to have the maximum impact for the most amount of people. That’s also proving to be a cost effective way to decide who gets certain types of devices and social care packages. So it’s useful for the health care service and the social care service.

John, I was wondering if you had any unique examples of applied data science in the health and social care sector. So a really interesting example of this is where you can build predictive models. That’s what data science allows you to do– it allows you to take data and build a predictive model of that. Well a particularly useful area of that is to predict, for example, how many people are going to come in through the doors of say an A&E department at a particular time of day. Why do we need that? Well then we can make sure that there’s enough staff there for them.

Obviously we don’t want to put too many staff in an A&E department– they won’t have anything to do. But we want to make sure there’s enough staff. So what we can do is that we can take the data– time series data, as we call it, because it’s a series of data points over time– and we can look at the fluctuations per day, for various holidays and things like that, and we can try and predict for every hour of every day what the intake is going to be for a particular department. And then we can know what staff we need to put on.

We might even assign things like certain traumas, for example, occur at certain times of day, or certain times in the week. And so we can have the specialist staff in place at the right time using these predictive models. Fascinating. You could imagine even that elements such as the weather– being predictions of weather, or like you say holidays– can also add to that model. Absolutely. There’s going to be all kinds of factors that we would need to build into there. The time of year, the weather, the time of day– as I say, holidays or not holidays.

And that will enable you to build a good predictive model of exactly how many people are going to be coming in through the doors of the department at that time. There’s a similar example, used in a city in America, where they used predictive models like this based on time series data to predict where crimes were going to occur, and exactly where in the city, so that the police would be there to pick up the criminals before the crime had even happened. That’s the kind of example that predictive modelling– what it can do for you.