Academic Profile: Professor John Frank

John Frank is Chair of Public Health Research and Policy at the University of Edinburgh and leads a project within the Data for Children Collaborative with UNICEF aiming to predict obesity in Scottish children.

 

What is the aim of your project?

We’re aiming to find out if we can identify children who will be persistently obese by the ages of 10 and 12. If those children are shown to be obese at these two ages in a row, we know they’ll almost certainly go on to have long term consequences later in life. The first impact is psychological damage during the teenage years. Then there will be physiological problems later in life, starting with pregnancy and birth complications for the young women. Later, type 2 diabetes, early cardiovascular disease, some cancers and arthritis strike those early who are obese. To do this, we’ll be using the data from the Growing up in Scotland project (growingupinscotland.org.uk), and will use predictive analytics to identify those children some years earlier. Why would we do that? Because the treatments – once a child is obese – aren’t very successful. It can require whole families to change their behaviour, which can be very difficult. People tend to be wedded to food preferences, so it’s very difficult to get them to change behaviours.

 

How much of a problem is childhood obesity in Scotland?

Prior to the present obesity epidemic (i.e. in the 1970s), the overall rate of childhood obesity would be expected to sit around 5%. But in Scotland it’s now around 22%. This used to be the same across all social class groupings, twenty years ago. But now, children in the top and bottom of the Scottish Index of Multiple Deprivation – which is a very good index – shows that children in the top and bottom 20% have separated; children from the most deprived backgrounds sit around 25% whereas children from less deprived are just below 20%. This is what happens in any chronic disease caused by what can be called lifestyle factors, such as poor diet or not enough exercise. Typically, families who have more information at hand and have more options available to them take action to prevent their children becoming more obese, such as changing their diet or enrolling them in physical activities.

 

What is the Growing up in Scotland database?

Growing up in Scotland is a child cohort originally of about 3500 children born around 2003-2004. It’s based on interviews with the parents, then – once children are about 8 – children can also be asked some simple questions. It’s a very good dataset, funded by the Scottish Government as a general resource. It has about seven stages of data collection at different age intervals – with height and weight for each of five stages.

The data is depersonalised and we won’t be combining it with any other data sets. Although our project doesn’t present any ethical challenges, there are still processes to ensure ethical safeguards. Anyone who wants to use it has to go through the appropriate process, with the Scottish Government ultimately granting permission. This can take a while and involves a certain about of paper work. Data access rules mean it will be analysed in a secure fire-walled environment. The data can be linked – which we might do in a separate study later — to wider NHS data on mortality, hospitalisation and prescribed pharmaceuticals from an NHS doctor – but not GP records (yet). So, there aren’t cost barriers – it’s free – but there are barriers in terms of access permission.

The emphasis for our work will be on those risk factors which are routinely collected in entire national populations of children – e.g. in the health care or educational systems. This could enable the creation of national predictive systems for obesity in children, enabling health services to offer early interventions to children at very high risk for full-blown obesity at ages 10 to 12.

 

What are the pre-requisites to set up something like Growing up in Scotland?

It just needs a government to have a survey research centre that can draw a sample representative of the entire population of infants. With new-born samples, typically they aren’t recruited to cohorts until around ten months of age, which is what was done with Growing up in Scotland. Any survey research centre should be able to find those families using a list of all births. After obtaining full consent from the family, interviews obtain the initial set of interview questions and entry of anonymised information to the database. Each child is assigned a unique identifying number, which we as researchers cannot link to any identifiable information on that child or the family.

 

How many years in advance do you hope to be able to predict obesity?

A key factor will be how many years before the ages 10 and 12 we can predict persistent obesity. The first record of weight (at around 10 months) isn’t useful because there’s still infant feeding and catch-up growth (for low-birthweight babies) to take in account. At the next stage – at around ages 1 and 2 – children simply haven’t had enough time in the world to settle into a specific trajectory regarding their weight. Data collection for weight and height around ages four and five is most likely to predict later obesity.

 

Can we deliver an algorithm that accurately predicts risk of childhood obesity?

Yes, there’s no risk around doing that because the data is already collected. It’s been cleaned of errors, and it’s sitting in a public repository. We’ve already got the necessary permissions to analyse it. However, the challenge would be the next stage of running a trial using the algorithm. That would require the randomisation of many hundreds (maybe more) of children to be screened or not screened, and then followed to see whether early detection and offering treatment actually confer health benefits. Such a “screening trial” costs a lot and so wouldn’t be proposed without an effective algorithm. Even just on the basis of the predictive algorithm we develop, however, governments can be advised based on the results and might want to do pilots for interventions, such as taxes on high sugar and fat foods.

 

What are the advantages of UNICEF’s involvement in the project?

UNICEF knows the many challenges faced in developing countries – while many of them now face the “dual burden” of both traditional malnutrition and new obesity in children, using insights learned from other countries can be helpful in dealing with the latter. UNICEF have an interest in making a predictive algorithm available to other countries, where it could be tested in different settings.

Patterns of childhood obesity in many of the developing countries where UNICEF is active are still emerging. Where childhood obesity does occur, it is mainly within the social elite, and there’s still a lot of children with cumulative malnutrition, leading to short height for their age, in other social groups.

Naturally, there will be differences between datasets collected in a post-industrial country such as Scotland and similar data sets from developing countries elsewhere in the world. The problem in Scotland is the reverse of the problem in developing countries where it’s the children from higher socioeconomic groups who tend to be groups where childhood obesity is present. Also, in very poor countries, data collection – especially using child cohorts requiring multiple interviews over many years – can be very difficult; there can be large amounts of population movement, people living precarious lives, high unemployment, etc.

 

What will be the challenges of using the results?

Having been in this field for 40 years, I’ve seen that an inaccurate screening test can do more harm than good. If we screen too early, we might put children into treatment programmes when their obesity might have resolved itself without an intervention, and vice–versa. So we plan to ensure our predictive algorithm for later obesity is sufficiently accurate to justify any screening processes that come later.

My team consists of Paul Bradshaw the Director of ScotCen (the survey centre which has done all GUS interviewing since its inception), and three post-doctoral researchers whom I have trained in the last half-decade, and who have used Growing up in Scotland data before, in relation to obesity predictors up to age 8. The statistical methods we’re using are not novel. However, if we find something that’s promising we can look at bringing AI/machine-learning expertise, which would involve reaching out to other disciplines in the university. I’m more comfortable with doing it that way than letting AI come up with the predictors, which can be a ‘black box’ process where no one is quite sure how it came up with the results.

 

For more information, visit: dataforchildrencollaborative.com