We have spoken a lot about what data is and how it is joined to other data sets, but data on its own is of limited use – we have to do something with it to make it useful – it’s like a shopping basket full of ingredients. What we do with it will depend on what we are trying to find out.

Sometimes we want to be able to predict if an individual is going to develop a disease in the future. This is difficult to do – people are complex, with lots of different characteristics about their health, well-being, demographics etc. To illustrate how we might predict if someone will develop a particular disease, let’s think about onion soup.

If I tell you that I am going to make onion soup this evening you could probably guess some of the items in my shopping basket this afternoon – onions, olive oil, butter, garlic… You could even look up a recipe online and have a good guess at exactly what was in my basket.

But what if you have no idea what goes into onion soup and the recipe isn’t available online? How could you figure it out?

You could ask everyone in town what was in their shopping basket today and if they are going to make onion soup tonight. Hopefully everyone who is going to make it will have bought the same ingredients, so now we know what goes into onion soup! Now, when you walk around the supermarket and look at other people’s baskets, you can predict if someone is going to make onion soup tonight.

Thinking about people again, if instead of ingredients we think of combinations of things like if they are a smoker, their blood pressure, where they live etc. and instead of onion soup we consider diseases like diabetes or a stroke, then we can start to see how they are linked. We could then say that if certain of these characteristics are present in an individual then we think they will go on to develop e.g. diabetes. We are starting to be able to predict what is going to happen to an individual in the future, based on what we know about them now.

But, life, like onion soup, is not that straightforward. What if I am buying ingredients to make onion soup tomorrow, not tonight? We would have to change our prediction from “this person is definitely going to make onion soup tonight” to “this person is going to make onion soup in the near future”. It’s the same with people – someone may have all of the characteristics which means they will likely develop a disease, but we can’t always tell when that is going to happen.

Some people may even buy the ingredients with the intention of making onion soup, but never get around to it. Similarly, having all the characteristics which are associated with a particular disease does not always guarantee it will develop. Humans are very complex.

To make matters even more complicated, often people shop for more than one meal at a time. Imagine looking at a basket with all the ingredients for onion soup and onion rings in it – without knowing this was intended for two meals, could we predict that onion soup was going to be one of those meals? Similarly, people often have characteristics which could lead to different diseases.

I have perhaps laboured the onion soup analogy a bit too much here, but hopefully you can see that we need to use data to understand the characteristics of people who we know have a given disease so we can than predict the likelihood of other people developing that disease who have the same characteristics. We usually refer to these as risk prediction models. Generally speaking, more data should lead to more accurate predictions.

We do a lot of this kind of work in the ARC, one example being the ACMI (Anti-Cholinergic Medication Index) project, which has taken characteristics around frailty in older people and the medications they take, to look for an association with being admitted to hospital due to a fall or delirium. By using historic data, we have quantified this association so now we can predict the likelihood of other people being admitted to hospital for the same reason.

I have highlighted just one potential use of health care data here, there are many more ways we use data, which we touch upon in this blog series.

Dr Olly Butters, Care and Health Informatics theme