Standardisation of nutrient data across EPIC centres

In this multi-centre international study, the investigation of diet-disease relationships at the nutrient level requires standardised methods to be used across participating countries. This is needed to obtain comparable nutrient estimates. Standardisation between countries needs to be done at every level, including data collection, aggregation and coding of foods, and when applying food composition data (Deharveng et al, 1999). If this process is inadequate, measurement error can lead to artificially weakened or strengthened relationships observed from the EPIC study.

The errors associated with food composition databases relate to the quality of data, completeness and degree of documentation. Deharveng et al (1999) have compared the food composition tables from nine of the European countries participating in EPIC: Denmark, France, Germany, Greece, UK, Italy, the Netherlands, Spain and Sweden. They were compared in terms of availability, definition, analytical methods and nomenclature of nutrients. Although most of the nutrients were analysed and expressed in a compatible way, some nutrients were not comparable because due to lack of commonality in methods and definitions of food components (folate, dietary fibre) or nomenclature (energy, protein, carbohydrate, carotenes, vitamins A & E). For example, two databases could both have analytical values obtained from the same chemical method, but provide different results if one expressed the results for raw foods and the other for foods as consumed. Furthermore, some tables were compiled with data produced over 20 years ago with outdated analytical methods, and many foods reported within the EPIC study are not included in the existing European food composition tables.

EPIC Nutrient Database Project (ENDB)

In the absence of a pan-European food composition database, EPIC developed a method to improve the comparability of the nutrient databases among its 10 participating countries – the EPIC Nutrient Database (ENDB) Project. The main objective of the ENDB project was to provide a standardised nutrient database for calibrating the EPIC dietary data and investigating diet-disease relationships at the nutrient level.

The EPIC-SOFT 24h recall dietary data available in the EPIC study was a valuable starting point for the ENDB project, because it is standardised between countries in terms of food nomenclature, classification, description and quantification. Also, the ingredients of mixed recipes were treated as foods; this makes the food intakes more comparable among countries and is easier for the attribution of nutrient values (Charrondière et al, 2002). The detailed EPIC-SOFT 24h recall data was used to provide food lists for the new database. Then each of the 10 countries participating in EPIC identified the subsets of the national databases that matched the foods reported by the 24h recalls in EPIC (called national datasets). These were then used in the compilation and standardisation of the ENDB; the following steps were involved:

1. Construction of empty standardised ENDB matrices using food entries reported from the EPIC-SOFT 24h recalls in the vertical axis. A set of nutrients were allocated to the horizontal axis according to the EPIC research priorities.
2. Matching of the EPIC foods to those available in the national databases (the subsets of the national databases that match the foods from the EPIC-SOFT 24h recalls), or alternatively, defining how to calculate or adjust them (for example using recipe calculations, cooking weight changes, mineral/vitamin losses).
3. Standardisation of the ENDB nutrient values.
4. Imputation of missing values by national compilers.
5. Calculation of country-specific nutrient values for EPIC foods not available in the national datasets.
6. Building of the ENDB by merging foods derived from the national datasets and those generated by calculation.
7. Export of ENDB and national dataset’s nutrient values and documentation in a common format. 

The final ENDB contains 26 nutrients for between 550-1500 foods per country; this totals food composition for about 10,000 foods consumed across Europe.

Experiences from EPIC are relevant to other large epidemiological studies. ENDB offers a unique opportunity to provide qualitative and quantitative insights on the current strengths and limitations for harmonising nutrient databases from different countries, and provides useful end-user recommendations to EuroFIR. 

Relationship with EuroFIR

The ENDB project was developed before the EuroFIR Network of Excellence was set up. The experiences of developing, standardising and using nutrient databases in the ENDB project offer important lessons for the plans and priorities undertaken as part of EuroFIR.

The main problems encountered during the ENDB project, and their ad hoc solutions, are presented below:

Selection and prioritisation of nutrients and other food components of interest

It is important to collect nutrient data on those nutrients that, although they are not directly of interest, are possible confounders (factors that are associated with the exposure of interest and known to effect risk of the disease outcome) for other diet-disease associations and so need to be adjusted for in analyses. Because there was a lack of certain nutrients in national databases, laboratory analyses for some nutrients of interest, for example specific fatty acids and folic acid, were needed.

Food coding or food matching

There were several discrepancies between the foods reported by EPIC study subjects and those available in national food composition databases. For example ‘spinach, fresh, stir fried’ might have been recorded by a study subject, but only ‘spinach, raw’ was available in the national food composition data. It was therefore necessary to link reported foods to the restricted foods available in the food composition databases using a technique called food coding, or food matching. This involved grouping foods in terms of their similarity in nutrient content. EPIC-SOFT facilitated the standardisation of this food coding/matching process. The main food characteristics to be considered when matching were defined within the matching system.

Mixed recipes and food cooked with fats

A number of homemade or commercial recipes, or multi-ingredient products were recorded by the EPIC study subjects. In order to match these, and also foods cooked with fats, to items in the national food composition databases, these recipes were broken down and their ingredients classified separately.

Cooked foods

In nutritional epidemiology it is felt that foods should be recorded and analysed as consumed, i.e. the cooked versions of foods if consumed as cooked. In cases where there were no cooked equivalents in national databases, the recorded foods were matched to the raw foods and then the values adjusted for weight changes and vitamin/mineral losses.

Documentation, standardisation and evaluation of nutrient values

Specially designed software was used to support the documentation, standardisation, evaluation and export of the national datasets; this is needed to allow clear identification, standardisation and evaluation of food and nutrient values. All food component values were expressed in the same standard units and mode of expression across national datasets.

Missing values

The different levels of completeness of nutrient values in databases can lead to measurement error. In the ENDB project, missing values corresponding to logical zeros or trace values were systematically identified. These were then assigned zero or corresponding limit values. This cleansing process then identified the real missing values. National compilers could then impute or borrow values for the same or similar food available in the evaluated national or foreign datasets. Via this process, the completeness of all datasets improved to above 98%. It was judged more important to approximate missing values than leave them as missing.

Recommendations for EuroFIR

The ENDB project can be viewed as an end-user initiative to harmonise nutrient databases for epidemiological research. The overall aim of this project was to improve the comparability and completeness of existing national databases for use in international research.

A full evaluation of the ENDB experience has been provided to EuroFIR. These insights contributed to the elaboration of EuroFIR’s strategic plans, priorities and guidelines. The end-user recommendations made to EuroFIR, have helped to strengthen many of the ongoing processes taking place within the Network of Excellence. The recommendations included:

1) Strengthen the multi-disciplinary nature of activities within the project by integrating different expertises and partners.

2) Improve the documentation of the existing national databases.

3) Reduce the gap between foods available on the market and those reported by study subjects, and those that exist in national databases.

4) Improve the comparability and standardisation of foods across databases, using a reference description and classification system.

5) Integrate the current knowledge on biomarkers of diet in the criteria for prioritisation of the EuroFIR nutrients and other food components.

6) Provide support to the different end-users on the way to use EuroFIR data and compile their end-user databases.

7) Identify fortified or enriched foods systematically in the databases and track their nutrient content changes over time.

References

Charrondière UR, Vignat J, Möller A et al (2002) The European Nutrient Database (ENDB) for nutritional epidemiology. J Food Comp. Anal.15: 435-451.

Deharveng G, Charrondière R, Slimani N et al (1999) Composition of food composition tables available in the nine European countries participating in EPIC. Eur. J. Clin. Nutr. 53: 60-79.

Slimani N, Charrondière UR, van Staveren W et al (2000). Standardisation of food composition databases for the European Prospective Investigation into Cancer and Nutrition (EPIC): General Theoretical Concept. J. Food Comp. Anal. 13: 567-584.

More information

For a more detailed report of this work please see:

See Slimani N, Deharveng G, Unwin I et al (2007) Standardisation of a European end-user nutrient database for nutritional epidemiology: what can we learn from the EPIC Nutrient Database (ENDB) Project? Trends in Food Science & Technology doi:10.1016/j.tifs.2007.02.005 (electronic reference)