Diet-disease relationship better understood using new data mapping method: study

22 Nov 2018 --- Researchers at Trinity College Dublin, Ireland, have developed a new data mapping method which may improve the quality of dietary data collected by short frequency questionnaires (SFQs).

The dietary assessment tools currently available are not able to precisely identify the correlation between diet and disease, often rendering the results biased and inaccurate. Even relatively simple descriptive analysis of "unhealthy" food intake data can be compromised and bias our understanding of the potential association with chronic disease. The researchers have now developed a data mapping method that combined the matched datasets of two studies to yield more accurate findings. 

“Using this approach to successfully map datasets from different surveys should help improve the quality of the data that can be estimated using SFQs, therefore, improving the potential to identify diet-disease relationships. This new research will interest those professionals involved in understanding dietary intake, including nutrition scientists, dieticians and health care professionals,” says the study’s co-author, Michael Crowe. 

This data mapping method was employed for the first time in a research project in Dublin Dental University Hospital, Trinity College, which tried to determine the association between dental problems in children and risk factors, like psychosocial factors, diet, body weight and general health. 

In this research project, the team used a new method to connect matched datasets from two studies to improve the quality of dietary data collected using SFQs in two large cohort surveys.

Growing Up in Ireland (GUI) is a nationally representative longitudinal study of infants in Ireland which used an SFQ to assess the intake of “healthy” and “unhealthy” food and drink of three-year-old children. The National Preschool Nutrition Survey (NPNS) provides the most accurate estimates available for dietary intake of young children in Ireland using a detailed four days food diary. 

The data were linked to fill in all GUI food groups with information from NPNS’s datafile. A data mapping algorithm was applied using food name, cooking method, and food description to fill all GUI food groups with information from the NPNS food datafile which included the target variables, frequency, and amount. 

The combined data were analyzed to examine all food groups described in NPNS and GUI and what proportion of foods were covered, non-covered, or partially-covered by GUI food groups, as a percentage of the total number of consumptions. The term “non-covered” indicated a specific food consumption that could not be mapped using a GUI food group, according to the study.

“High sugar” food items that were non-covered included ready-to-eat breakfast cereals, fruit juice, sugars, syrups, preserves and sweeteners, and ice-cream. The average proportion of consumption frequency and amount of foods not covered by GUI was 44 and 34 percent, respectively. 

“The difficulties associated with measuring diet are well documented. Having explored the augmentation of a limited SFQ in the GUI study by unidirectional mapping from a more detailed four-day food diary in the NPNS study. The results have highlighted the deficiencies in using SFQs for investigating associations with health outcomes. The SFQ did not capture a substantial portion of habitual foods consumed by three-year-olds in Ireland,” the study notes. 

Crowe tells NutritionInsight that: “The key message is the misappropriate interpretation of results from large cohort surveys that use dietary instruments that cannot capture a large amount of ‘unhealthy’ food and drink consumed.”

The results of the study may improve future SFQs used in research to give a better understanding of the diet-disease relationship. Researchers interested in specific foods may use this method to assess data using the mapped food database as reference, notes the research.

“More research is warranted on automating or semi-automating these methods and to improve the speed and scale of mapping FD to food composition tables,” concludes Crowe.

By Kristiana Lalou