Surveillance and Post-Market Monitoring of Potential Health Effects of Novel (including GM) Foods: Feasibility Study

Background
In May 1999, a report from the Chief Medical Officer and the Chief Scientific Advisor was published reviewing the health implications of genetically modified foods (Donaldson and May, 1999). It made a number of recommendations, including 'instituting population health surveillance...to monitor population health aspects of genetically modified and other types of novel foods' and '...to examine trends over time to detect any early changes in the incidence of adverse health outcomes, whilst recognising the difficulties in establishing causal relationships' (page 24).

In the light of these recommendations, this FSA-funded study was initiated to investigate the feasibility of using commercially available datasets on household food consumption, and sales through major supermarkets, to carry out nutritional surveillance. Intakes of novel foods were not monitored, only the feasibility of carrying out a long-term food intake surveillance project.

The key questions that this study aimed to address were whether it was possible to use such data as a means of quantifying possible 'exposure' to certain foods and nutrients at population level, and to establish whether there is demonstrable temporal, geographical and/or socio-economic variation in these 'exposures', since otherwise any variation in health outcome would be unexplainable by that exposure.

Methods
Two commercially available datasets were obtained for this study: (1) food purchases collected by a rolling Panel of c.10,000 British households between 1991 and 2000 (provided by Taylor Nelson Sofres (TNS)); (2) total sales of food products from 7 major supermarkets in Great Britain for the year 2000 (provided by Information Resources, Inc. (IRI)). The total cost for these data was £93,412.50.

A major nutrient coding effort was undertaken to convert the product description data provided by the commercial companies into estimates of energy and nutrient content. This revealed various problems with the quality and nature of the available data, including errors in the Panel demographic data and weeks with no record of any food purchases (TNS), and lack of detailed product descriptions and no information on total weight of each product sold (IRI).

While most issues with the TNS data were resolved satisfactorily, the IRI data ultimately proved to be un-useable for the present study. This study also highlighted the limitations of currently available nutrient databases for carrying out coding work on this scale.

Information on household purchases of four different marker products (marker groups 1, 2, 3 and 4) was also extracted from the TNS database to investigate the feasibility of tracing consumption of specific food items in the diet.

Results
A total of 4,550,088 weeks of 'shopping basket' data were available from 105,688 individuals in 33,177 households in the TNS Panel between 1991 and 2000. TNS estimate that these data capture about 70% of total food consumption, since impulse purchases and food eaten outside the home are not included.

Basic demographic checks suggested that the Panel members were broadly representative of the GB population as a whole, although children and households with families were over-represented and there was a slight shortfall of households in the most deprived areas. Comparison with published nutritional studies showed that the TNS data under-estimated total daily energy intake by about 32% on average, which is in line with the assumed 70% capture rate for these data.

Estimated macronutrient intakes were expressed as percentages of total energy intake (%TEI), to help guard against systematic under-reporting biases: the mean and variance of these estimates across households in the TNS Panel were generally similar to those reported in the National Food Surveys and other nutritional studies, giving confidence in the validity of the nutritional information obtained from the TNS 'shopping basket' data.

Statistically significant temporal, geographical and socio-economic differences were detected for all nutrient intakes estimated for households in the TNS Panel. However, even small differences between the mean values in each sub-group will appear statistically significant because the tests are based on very large numbers (total sample size = 33,177 households). There was also considerable variation in estimated nutrient intakes within each sub-group.

Only a small proportion of TNS households ever purchased any of the marker products (less than 4% in each case). However, this proportion varied significantly by socio-economic subgroup and (for marker group 1 and marker group 3 only) by region, suggesting that it would be feasible to detect variations in 'ever' versus 'never' exposure to specific food products using such data. Within the subset of households that purchased each marker product however, there was no strong evidence of geographic or socio-economic variations in the amount of each marker product purchased, partly due to the very small number of households in this subset.

Conclusions
Subject to some enhancements (in particular to the sampling methods, quality control checks and improved availability of information on food compositions to facilitate nutrition-coding techniques), the results of this feasibility study suggest that it would be possible to monitor food-purchasing patterns reliably at household level using the commercially available TNS (or similar) data.

As such, these data could be used to inform nutritional surveillance to provide information on population 'exposure' to novel food products. Surveillance of ingredients such as soy protein (whether or not genetically modified) is not possible using currently available information, but would be feasible using the TNS data if lists (and gram weights or volumes) of ingredients could be made electronically available for each product, in such a way that they could be readily linked to the barcode data.

This feasibility study involved only examination of food product data, and not of any health data. Clearly, any surveillance system will need to correlate food purchase/sales information with health data. Using the sources of food purchase and sales data considered in this study, this could not currently be done at an individual level, as the purchase data are based on a sample of households, and sales data are generated at the level of the supermarket (and then further aggregated for confidentiality).

Ecological analysis could be carried out to relate health outcomes to potential exposures at a group level (e.g. by small area), although problems of confounding and ecological bias will tend to complicate interpretation. Effect sizes would need to be large in order to be reliably detected using such methods, although we believe it would be feasible to use this approach to investigate apparent temporal clusters of health events that were thought to reflect exposure to recently introduced novel foods.

A second approach would be to exploit planned future improvements to data linkage in the NHS that will allow health events such as hospital admissions and GP records, as well as cancer diagnosis and mortality which can be followed up using current systems, to be linked and followed-up using the individual's NHS number (provided proper consent is obtained).

The TNS Panel generated a database of more than 100,000 individuals followed for an average of 2.5 years over the 10 year period considered in this study. By recording the NHS number of these individuals, so that they could be flagged every time they appeared in routine health datasets, it would be possible to monitor short-, medium- and long-term health effects and link them back to dietary information at the household level.

A list of the recommendations resulting from this feasibility study is given below:

Recommendations
1. FSA are encouraged to set up a system similar to the Nutrition Coordinating Centre (NCC) database in the USA for the purposes of maintaining a constantly updated database of nutritional information relating to food products currently available in this country. This system should include a) both nutrient and (where possible) product ingredient information; b) a direct link between barcode information and product names and nutrient data; c) liaison with the food manufacturers to ensure that the necessary ingredient information is provided, and to the required format; d) a direct link with manufacturers/retailers so that their ingredient information can be fed directly into the database; e)annual updates of existing product information so that changes in nutrient (and ingredient) composition can be tracked.

2. FSA should work with TNS to initiate suggested improvements to their database (identified in section 9.4) to make it suitable for long-term nutritional surveillance.

3. FSA are encouraged to set up a nutritional surveillance system based on the TNS Super Panel (or equivalent) to provide a continuous record of foods entering (and leaving) the market place.

4. FSA should approach TNS about recording NHS number of Panel members and obtaining appropriate consents (and possibly on-going questionnaire data) to allow the health experience of Panel members to be followed up longitudinally through routine NHS records.

5. Serious consideration should be given to replacing the NFS with a continuous monitoring system using barcoded household purchase data, run by TNS (or similar), provided the enhancements identified in section 9.4.1 could be implemented. This would allow estimation of national dietary trends and other information currently provided by the NFS, and also provide the basis for an on-going nutritional and health surveillance system.