Recently, an international team led by Prof. John Speakman from the Shenzhen Institutes of Advanced Technology (SIAT) of the Chinese Academy of Sciences, alongside the International Atomic Energy Agency-DLW database consortium, have derived a predictive model/equation by combining classical statistics and machine learning for total energy expenditure. The model predicts energy from easily acquired anthropology variables, like body weight, height, age and sex without actually measurement, offering a more objective way to assess the validity of food intake records.

The work was published in Nature Food on Jan. 13.

Dietary exposure is closely implicated with the chronic diseases. Nutritional epidemiology aims to link dietary exposures to chronic disease, but present instruments for evaluating dietary intake particularly at the large-population scale greatly rely on the ability of the investigated subjects of noting down or recalling what they have eaten or they are eating. These include methods such as food frequency questionnaires, 24 hour recall interviews and food diaries.

It is pretty well known that such tools are extremely inaccurate because people forget, or may even deliberately falsify their reports.  Great amount of inaccurate data (termed here dietary misreporting) will mislead the decision of nutritional strategy and policy. 

In this study, the researchers used a method called the doubly-labelled water technique, an isotope-based method that measures directly individual's energy needs. In total they pulled together more than 6000 measurements and used classical statistics and machine learning based approaches to derive a predictive model which they then validated in about 600 additional subjects. 

To demonstrate the usefulness of the proposed method, the researchers applied it to two large surveys of food intake data: National Health and Nutrition Examination Survey (NHANES) in the USA and National Diet and Nutrition Survey (NDNS) in the UK. Using the new method they found that 47.6% of food intake records in NHANES and 54.2% in NDNS had unrealistically low levels of energy intake. 

This finding refreshes traditional research concepts in nutritional science. 

“The main issue with the new tool is that it suggests we should throw out large amounts of data, and nutritionists using the dietary instruments may be unwilling to do that. However, continuing on just publishing erroneous data because its too painful to acknowledge it’s flawed, probably isn’t the best way forwards for nutrition science. I think as we go into the future many widely held beliefs that have been based on these problematical methods will need to be revised,” said Prof. John Speakman.