Researchers propose a new model to screen for misreporting in nutrition surveys

An international team led by Prof. John Speakman of the Shenzhen Institutes of Advanced Technology of the Chinese Academy of Sciences has derived a predictive model by combining classical statistics and machine learning for total energy consumption, providing a more objective way to assess the validity of food intake records to be assessed.

The research was published in Natural food on January 13.

Nutritional epidemiology aims to link dietary exposure to chronic disease, but in the past, methods for evaluating dietary intake have relied heavily on subjects’ ability to write down or remember what they ate or what they eat, using tools such as food frequency questionnaires, 24-hour questionnaires, recall interviews and food diaries.

Such tools are widely known to be inaccurate as people can forget or even falsify their reports. Increasing amounts of inaccurate data (here called nutritional misreporting) will mislead the decision on nutrition strategy and policy.

In this study, researchers used an isotope-based method, the double-labeled water technique, which directly measures the individual’s energy needs. They collected more than 6,000 measurements in total and used classical statistics and machine learning-based approaches to derive a predictive model that was then validated in approximately 600 additional subjects.

The resulting equations are currently the most accurate method of estimating energy requirements without making an actual measurement.

To demonstrate the effectiveness of this model, researchers applied it to two large surveys of food intake data: the National Health and Nutrition Examination Survey (NHANES) in the US and the National Diet and Nutrition Survey (NDNS) in the UK . They found that 48% of food intake data in NHANES and 54% in NDNS had unrealistically low levels of energy intake.

“This new model suggests that we need to throw away large amounts of data, and nutritionists who use nutritional tools may not be willing to do that. But continuing to publish flawed data because it’s too painful to admit it’s flawed is probably not the right thing to do. I think as we move into the future, many widely held beliefs based on these problematic methods will need to be reconsidered,” said Prof. John Speakman.