Optimising an FFQ Using a Machine Learning Pipeline to teach an Efficient Nutrient Intake Predictive Model

Area of Science:

• Nutrition science
• Data science
• Computational biology

Background:

• Food frequency questionnaires (FFQs) are standard, cost-effective tools for nutrition monitoring.
• Extensive FFQs may contain redundant questions, potentially reducing participant engagement.
• Optimizing FFQ length is crucial for efficient and accurate dietary intake assessment.

Purpose of the Study:

• To investigate the impact of reducing questions on predicted nutrient values and diet quality scores using machine learning.
• To identify the most informative subset of questions within the Extended Short Form Food Frequency Questionnaire (ESFFFQ).
• To compare the performance of various machine learning algorithms in predicting dietary intake and quality.

Main Methods:

• Applied machine learning algorithms to different subsets of questions from the ESFFFQ.
• Utilized the PROMETHEE method to compare algorithm performance across multiple metrics.
• Evaluated the predictive accuracy of reduced FFQ question sets for specific nutrient intakes (sugar, fiber, protein) and overall diet quality.

Main Results:

• Machine learning models using smaller question subsets accurately predict sugar, fiber, and protein intake.
• Reduced FFQ question sets, when analyzed with machine learning, generally outperform traditional statistical methods.
• Certain subsets of questions are sufficient for predicting diet quality scores, indicating redundancy in longer FFQs.

Conclusions:

• Machine learning offers a robust method for optimizing FFQ length by identifying critical questions.
• Optimized FFQs can maintain data accuracy while reducing participant burden.
• The proposed approach can guide the development of more efficient and effective dietary assessment tools.