TL;DR
Researchers have developed a new predictive equation using a large dataset of doubly labeled water measurements to accurately estimate total energy expenditure (TEE) from simple variables like weight, age, and sex. This equation allows for the identification of individuals who are likely misreporting their dietary intake in studies, revealing that over 50% of participants in two major surveys showed implausible energy intake. This has significant implications for understanding diet-disease relationships, as misreporting can introduce systematic biases.
What Problem Does This Paper Address?
Nutritional epidemiology relies heavily on self-reported dietary intake data to link diet to chronic diseases. However, these methods, such as food frequency questionnaires and dietary recalls, are notoriously inaccurate. People struggle to accurately estimate portion sizes, have imperfect memories, and may even intentionally falsify their reports. This "misreporting" can lead to flawed conclusions about diet-disease associations. While methods exist to flag potentially erroneous data, they often rely on arbitrary cut-offs or complex measurements. This paper aims to address this by developing a more robust and accessible tool to identify misreporting in dietary studies.
Methods
The core of this research is the development of a predictive equation for Total Energy Expenditure (TEE). The authors utilized a large dataset of TEE measurements obtained via the doubly labeled water (DLW) technique, considered the gold standard for measuring energy expenditure. They employed a classical general linear regression model, incorporating variables such as body weight, height, age, sex, and elevation of the measurement site. Machine learning models (Random Forest, XGBoost, Support Vector Regression) were also explored but did not outperform the linear regression. The key innovation lies in deriving an equation that predicts TEE using easily obtainable variables, and then establishing 95% predictive intervals around these predictions. These intervals serve as a threshold to identify individuals whose reported energy intake falls outside the expected range, thus flagging potential misreporters.
Datasets
- International Atomic Energy Agency (IAEA) Doubly Labeled Water (DLW) Database: This forms the primary dataset for deriving the predictive equation. It contains over 7,500 DLW measurements from individuals aged 8 days to 96 years. For the final model, 6,497 measurements from individuals aged 4 years and older were used, split into an analysis set (5,899 individuals) and a validation set (598 individuals).
- National Diet and Nutrition Survey (NDNS): A UK-based survey used for applying and validating the predictive equation.
- National Health and Nutrition Examination Survey (NHANES): A US-based survey used for applying and validating the predictive equation.
Key Insights and Ideas
- Misreporting is pervasive and systematic: The study highlights that misreporting isn't just random noise; it's a significant issue that systematically biases results, particularly when it comes to macronutrient composition.
- Simple variables are powerful predictors of TEE: The fact that basic anthropometric and demographic data can predict TEE with reasonable accuracy is a significant finding, making the screening tool accessible.
- Macronutrient bias is linked to BMI: The study reveals a strong correlation between reported macronutrient intake (especially protein and fat) and BMI, suggesting that misreporting itself is influenced by body weight and can distort diet-disease relationships.
- Existing cut-offs are less sensitive: The new method appears to identify a higher prevalence of misreporting compared to older methods like the Goldberg cut-off.
Key Results
- A predictive equation was developed that explains 69.8% of the variation in the natural logarithm of TEE using body weight, height, age, sex, and elevation.
- When applied to the NDNS and NHANES datasets, the equation identified that 27.4% of participants had energy intakes falling outside the 95% predictive interval, indicating potential misreporting. (Note: The abstract mentions >50%, but the text and tables consistently point to 27.4% for the combined datasets. I'm using the more specific number from the text.)
- The study found that as the level of misreporting increased, the reported macronutrient composition of the diet became systematically biased, with higher reported protein and lower reported fat intake.
- The discrepancy between reported intake and predicted TEE was negatively correlated with BMI, meaning individuals with higher BMIs were more likely to under-report their energy intake.
Limitations and Future Work
- Exclusion of specific populations: The predictive equation was derived from individuals without specific diseases, unusual physical activity levels, or during pregnancy/lactation. It is explicitly stated that the equation cannot be used for these groups as it underestimates their energy expenditure.
- Assumption of energy balance: The derivation assumes individuals were in energy balance during the DLW measurement. Deviations from this (e.g., rapid weight gain/loss) could lead to misinterpretation of predictions as misreporting when it's actually due to energy imbalance.
- Incomplete predictor data: While the model is robust to missing elevation or ethnicity data, complete predictor datasets yield better predictions.
- Future work: The authors suggest that integrating independent measures of physical activity (e.g., accelerometry) could further improve the model. They also emphasize the need for innovative approaches to dietary assessment that rely less on self-report to overcome the fundamental limitations of current methods.
Summary updated