Can we really predict which babies will become obese children?

Some headlines from November 29, 2012:

Childhood obesity ‘can be predicted by check at birth’ – BBC

Formula could predict your baby’s chances of growing into a fat child – Mail Online

Obesity scorecard can identify high-risk cases at birth, say scientists – Guardian

Online calculator to predict childhood obesity – The Telegraph

For the past few years, I have been involved in research aiming to identify infants at an increased risk of later obesity, using only data from their first year of life (an abstract; a report). Given my work in this area, I read these stories, and the PLoS One paper they are based on, with great interest.

I have several serious problems with this work, particularly with how the research has been communicated, by both the media and the paper’s authors. I will outline my more serious objections, before providing more context on the statistical analyses underpinning this research.

 

We cannot predict which babies will become obese. We can only make predictions.

Looking at the above headlines, some suggest that we can predict which babies will become obese. We can’t. We can only make predictions, which is not the same thing. The former phrasing suggests infallibility - the later wording correctly conveys that predictions can and will be wrong. Why does this matter?

 

Telling parents that their kid might become obese has consequences.

So what happens if you tell new parents that their infant is “at risk” of becoming obese? Here is a quote from the paper, suggesting what the authors think might happen:

“Parents of newborns are particularly sensitive to information given about their child’s health. Once informed of their baby’s increased risk for obesity, they might be more receptive to routine advice provided from birth during the first two years of life within population-wide prevention: breast-feeding, feeding on demand, weaning no earlier than the sixth month with recommended meal patterns and food portions, avoiding of television and sugar-sweetened beverages.”

Seems reasonable. Here is a related quote from our own paper:

“If a suitable tool was available, care would be needed by the primary care practitioner to ensure that the risk assessment was done with sensitivity to the perceptions and concerns of the parent. While most practitioners may be experienced, there would be a clear need for further training in the use of a tool, as the outcome from an obesity risk assessment of an infant may be difficult to predict. Hopefully, the parent would seek guidance and, if appropriate, some reassurance from their healthcare provider. However, other responses may occur. Some parents might be so stigmatized and antagonized that they would refuse to co-operate with the primary care practitioner, and might be driven to disregard the advice provided. Another, more unpredictable, outcome could be that the parents could become so alarmed by a high-risk assessment that they embark on a totally inappropriate regime that drives the infant down into the lowest weight centiles, leading to a poor later lifestyle and associated morbidity.”

In fairness to the authors, they also make this point in the paper, saying:

“Encouraging strategies aiming at significantly decreasing energy intake in infants should be avoided however, both because of the well known difficulties encountered by parents in doing it and because of potential, unknown harmful effects of an early caloric restriction.”

This didn’t, however, stop the study’s lead author from saying this, in the Guardian.

“This test takes very little time, it doesn’t require any lab tests and it doesn’t cost anything.”

In my opinion, there are potential risks (or costs) inherent in providing such a risk prediction, which leads me to the following:

 

Providing a risk calculator without any additional information is bad.

I think this is what I am most offended by. They have provided their risk calculator online. It has no information on how to use the calculator, how to interpret the result, and what actions you might take given the result. Perhaps they were too busy and just wanted to get this out as soon as possible? That said, I found this quote quite interesting, from the lead author.

“Nobody has tried to make an equation to predict this before and I don’t know why.” (Link)

The reality is that there are several groups who have done work on at least some aspect of this problem. Some examples include this, this, this, this, this, this, and this.

I can only answer by saying that we haven’t promoted any such equations because of the concerns raised above. We are currently in the process of applying for funding to explore not only the development of an obesity risk tool, but also an evaluation of it costs, risks, and how it might be actually applied. I strongly feel that releasing such a tool without more detailed evaluation is irresponsible.

 

More later on the statistics of prediction:

What is a “predicted probability”?

Why a  tool with a high AUC value doesn’t necessarily lead to useful, cost-effective predictions.