J.A. Barnes
Jane Barnes is a consultant with Foodsense, 103 Cremorne Road, Cremorne NSW 2090.
The 1980s have seen an ever growing emphasis on food and nutrition. National nutrition policies have been formulated, health education programs with major food components have proliferated, the media are full of nutrition information and misinformation, and nutrition is increasingly becoming an area of concern to the Australian consumer. As a result the demand for access to accurate detailed nutritional information about the Australian food supply has escalated.
The demand for the information has been formalised as food legislation requires full ingredient listing on packages and nutritional labelling to justify any nutritional claim made on packaging or advertising. Requests to extend the labelling requirements range from the anxious consumer, concerned about the quality of manufactured foods, to health authorities who believe that full access to nutrient information will enable consumers to make more informed decisions and thus select a more appropriate balanced diet.
The penalties for not providing the labelling information required, or for providing incorrect information, can be so costly that reputable manufacturers go to great lengths to ensure that the required information is provided and is accurate. Indeed, manufacturers often go beyond the formal requirements, answering enquiries from the general public, consumer groups, health professionals and government bodies regarding specific aspects of their products.
To service such enquiries more effectively, one major manufacturing company initiated a project to make nutrition and ingredient information on its food products readily available to the public. This involved the production of a nutritional information booklet generated from product formulas and calculated by means of a modified nutritional computation program. The problems encountered in this exercise, which required close collaboration with the suppliers of the program, form the basis of this paper.
Project options
There were two options available to generate the information, ie analysis or calculation. Considering the magnitude of the project (250 foods), analysis was quickly ruled out as raising too many complex issues such as sampling and analytical methodology, collection of samples from manufacturing outlets scattered round the country and interpretation of the laboratory results. Further, the cost of performing analyses would have been prohibitive. On current commercial prices, simple proximate analysis, plus analysis of sodium and potassium would cost about $200 per food if carbohydrate were calculated by difference. A more comprehensive analysis, including full fatty acid, vitamin and mineral determinations would increase the costs to between $700–$1,000 per food. The number of samples required to obtain a meaningful figure to account for seasonal variations in all the ingredients and the differences which inevitably arise in the manufacturing process meant that a project based on analysis would also not meet other needs, discussed later. It was therefore rejected as a viable option for the project. Calculation of the data from ingredient information was the other option and this option was selected as being both cost-effective and flexible as described later. It was decided to produce the information brochure by computerised recipe calculation similar to the approach of Marsh (1983).
Calculation systems
In a comparison of available calculation methods (Marsh 1983), particular attention was paid to the variations found in the results due to the different retention and yield factors built into the various programs to account for nutrient changes and losses during the cooking processes. No one method of calculation was found to be superior. Indeed the manipulations involved to account for retention and yield factors were considered too complicated to incorporate into a program which would have multiple users. Marsh's recommendation (1983) was to use a simple summation method, ie addition of all ingredient nutrients, with a basic computation built in to account for water losses during cooking.
To investigate the reasons for differing computation results, Frank and co-workers in 1984 had analysed eight representative 24 hour dietary recalls, using two similar software programs containing the same basic USDA nutrient data base extended by the individual operators. This comparison found major differences between the calculated nutrient intakes. The major contributor to the diversity in results was the differing sources of additional nutrient data and the criteria applied to their selection. Taking this kind of comparison one step further, Hoover (1983), reported the results of a project carried out for the US National Nutrient Data Bank Conference series. This project set out to compare the nutrient information generated from a range of computer programs. Proximate nutrients were found to vary by between 11% to 25% of their respective means when results were compared between eight systems. Quality control protocols were introduced to eliminate coding judgements relative to portion size and selection of food items. However, variation between computerised calculation systems remained high. Differences in the nutrient data between data bases were considered to be a source of the variability in the final calculations produced in the study.
A model to review nutrient data base capabilities was subsequently developed (Hoover & Perloff 1983) and this comprised several computing tasks. One task was to verify the accuracy of entry of data for additional food items into the data base; another task was to check the accuracy of calculation of a recipe comprised of nine ingredients. A test of the review model showed that the recipe calculation task produced the greater variability, the final protein composition of the recipe varying from 9.1 to 12.7 g per 100 g. The variability was mainly attributed to differences in the nutrient data in the data bases and to differences in the yield factors used. In a later specific study of data base development, Shanklin and co-workers (1985) again identified data base entries as the cause of analysis variation between programs. More recently Powers and Hoover (1989) evaluated various calculation software programs. These workers acknowledged the shortcomings of the summation method, but their comparison did not enable them to identify and recommend a superior method of computation.
All the available nutrient computation computer programs in Australia are based on the summation principal, so selection of the program depended on its capacity to fulfil other user needs of the client.
Additional criteria for selection of data base computation program
The other requirements considered during the selection of a suitable data base and analysis program for the project were: ease of update when changes occurred in ingredients or formulations; ability to handle new product work particularly in the development of food items to meet specific nutrient guidelines, eg the National Heart Foundation's Food Approval Program (Haddy 1990); formulation of specialised food products; tool in quality control procedures; and, facility to enable the identification of the presence of specific ingredients, eg presence of salicylates or MSG in the food.
The first four requirements would be met using an appropriate calculation type system. To fulfil the fifth criterion a program was required which had the flexibility to extend the fields of the data base by at least nine. Then every food item listed could be coded for the presence or absence of particular ingredients or compounds, such as wheat, egg, milk protein, salicylate etc. When a product recipe was finally calculated a positive listing of all these specific ingredients could be produced.
Package and nutrient data base
The program selected (Diet 1, Xyris Software) was specifically modified both to accommodate the extra data base fields needed and to incorporate an additional calculation option which would extend the range of foods in the nutrient data base, which was NUTTAB87 (Common-wealth Department of Community Services & Health 1987). This data base contains a mixture of both cooked and uncooked foods and as it was primarily developed to be a tool for analysis of diets as consumed, many basic uncooked foods, for which recent Australian nutrient data have been published were not included. Only about 35% of the ingredients used in the manufactured products could be located in the data base. To develop the complete nutrient data base required for the project necessitated extensive additional data input.
The first step was to draw up a list of all the food ingredients used in the range of products to be analysed, thereby identifying the gaps in the data base. These groups were: those foods for which Australian data were available in the literature (eg raw meats); those foods for which published Australian data were available but in an inappropriate form (eg dried fruits and vegetables for which data were only available on the fresh products); those ingredients where incomplete Australian data were available (eg flours); and, new items, for which little information was available on nutrient content or availability (eg thickeners and stabilisers).
Local data available
The major source of published Australian data was a series of papers published between 1981 and 1989 (eg Greenfield & Wills 1985).
Dehydrated items
Many of the items in the product range were dry products (eg soup mixes) which incorporated many dried ingredients. Factory specifications all generally included a moisture specification for these ingredients. It was therefore decided to calculate the dehydrated form from the hydrated data. This was viable as only macro nutrients were being considered and few changes of significance occurred to these nutrients during the dehydration process. A conversion formula was built into the computation program enabling conversion of the data for any food to a different hydration level. A similar calculation was not possible in products with varying levels of other nutrients, eg flours.
The diverse range of baked products to be calculated employed wheats with protein levels varying from 9% in the soft protein types to 15% in the high protein durum wheat types necessary for successful pasta manufacture. NUTTAB only included flour with 12% protein content. Although protein levels were available for the flours, it was not possible to do a simple conversion as with the water content.
Incomplete data
In the case of the flours discussed above, appropriate figures were incorporated from food tables from other countries which had appropriate protein levels. Here ingredient specifications were most important to facilitate selection of an appropriate figure from the consulted data. The major sources of data were Paul & Southgate (1978), especially the supplements (eg Tan & others 1985), the German food tables (Souci & others 1981) and the USDA food tables (eg USDA 1989).
New foods for which no data exist
There appeared to be four options to fill the gaps: commission analyses of the ingredients, or at least only accept data based on analysis; generate all data by calculation from ingredient specifications drawn up by the company food technologists; utilise suppliers' data, even though some would be from analysis and some from calculation; or a combination of the last two mentioned.
With an estimated 300 different ingredients to be considered, the first option would have been prohibitive from the point of view of cost and time. Although a large percentage of these ingredients were governed by strict ingredient specifications from the company, only in a few instances were the specifications found to be adequate to enable the appropriate data to be recalculated. For example for the textured vegetable proteins, moisture and sodium levels could be found, but data were inadequate for other nutrient calculations. This limited the application of the second option. The fourth option was chosen but as manufacturers' data were generated both by analysis and by calculation, it was important to determine which had been used in each case so that the data base could be marked accordingly. Inevitably judgements had to be made as to the best value to ascribe to an ingredient or additive. Since these judgements inevitably contained an element of personal bias, it is easy to understand how the studies cited earlier attributed most of the final discrepancies to differences in the data base. As ingredients for the food industry are continually being modified or developed, new data will always have to be generated for computation programs. The data base content is likely, therefore, to remain the weakest link in the system.
Specific difficulties
Specific difficulties were posed by new materials; by flavours and colours; by confidentiality; and by ingredient-free lists.
New materials
With the increasingly complex ingredient supply, new ingredients and additives are being developed about which little is known, particularly regarding nutrient availability. A major example of these little studied additives is the modified starches.
In the food supply of the 80s, modified starches became increasingly common. Modified starches are particularly helpful in adding texture and organoleptic qualities to foods which have reduced fat content. With the drive towards lower fat foods and the sophistication of chemical technology it is to be anticipated their usage will increase. Therefore the question of how resistant the modified starch is to digestion in the human gut becomes relevant and even critical in determining its contribution to the overall nutrient composition of a product. Particular products, like the gravies in the survey, already contain significant quantities of modified starches. The food labelling regulations permit 10% variation in declared energy level and 20% variation in other nutrients, which may not be generous enough when considering a product with such a major contribution from a nutrient of unknown bioavailability.
Flavours and colours
Undoubtedly flavours and colours are the biggest and most diverse group of additives used within the food manufacturing industry, and this group represented the largest group of ingredients in the project. Over the last few years companies have endeavoured to find flavours which may be listed as “natural” and colours which likewise do not have to carry the stigma of “artificial”. This change has resulted in an even more extensive range of flavours and colours being employed. The carrier for a flavour is important in determining the ways in which the flavour can be successfully used. Huge variations were found in sodium content depending on the carrier present. Initially it was assumed that savoury flavours would contain up to 40 g sodium per 100 g and sweeter flavours never more than 20 g sodium per 100 g. However, further enquiries of suppliers soon revealed that in some instances new, more natural, sodium-free or lower-sodium carriers were available. However manufacturers, even when not inhibited by confidentiality, were often unable to supply the sodium information required for the data base having never estimated this themselves. The figures eventually used were often a mixture of calculated data and manufacturers' estimations, supplied to satisfy our requests. However, in most products the flavours and colours were present in such small quantities that the variations in data would produce minimal impact on the final results.
In products such as soups where flavours represent a more significant percentage of total weight, initial calculations using the generalised sodium estimations proved to be very different from five year old laboratory analysis figures for some of the products. The revised estimates agreed well with sodium analyses of the new products generated by the company laboratory.
The current demand for products with lower sodium levels will generate more requests to flavour manufacturers regarding the levels of sodium in their products and increase demand for their lower sodium products. As a result it should become increasingly easy to collect sodium data on these products. If fuller nutrient analysis calculations are required then further details of the nutrient composition of the carriers will also be necessary.
Confidentiality
Part of the difficulty in obtaining accurate ingredient data is the importance of this information to the manufacturer. Competition within the ingredient supply industry is fierce. Flavour companies now have vast ranges of flavours, many of which are designer flavours developed by the companies in response to individual requests from manufacturers. Many food technologists have individual additive preferences when building up new product flavours. So in a large company it was not surprising to find there were up to seven beef flavours employed which to the uninitiated appeared to be performing the same function.
The flavour manufacturers have to guard the uniqueness of their products, to remain competitive. Often the difference between one company's particular flavour and their competitior's is the flavour carrier. If the nutritional information were freely available, a discerning chemist from the opposition may well be able to work out the formulation. So to respect the flavour house confidentiality at times an educated guess as to the sodium content had to be made.
Ingredient-free listings
Public interest in food sensitivities generates many of the requests to manufacturers. Consumers are particularly anxious to know whether some specific ingredient of concern is present or not in a food. In addition some consumers are interested to know if products contained added sugar or animal fats. It was decided to compile a positive listing of nine ingredients, wheat, gluten, milk protein, lactose, egg, yeast, added sugar, animal fat and MSG. A listing for the presence of salicylates, anti-oxidants and amines was considered but not published at this stage. As each food item was loaded into the data base, the presence of any of these ingredients was recorded, eg flours all had a positive indication for wheat and gluten. Decisions had to be made about, eg whether eggs should gain two positive listings, for egg and animal fat, and, since MSG is a naturally occurring substance in foods such as tomatoes, whether all tomato products should be listed with MSG.
Conclusion
No system is perfect. Recognition of the major limitations empowers the user with the caution not to misinterpret the final data. The quest for more information about the composition of foods in the diet is laudable and to be pursued, and a data base computation system can augment current knowledge. The individual figures can only ever be a best approximation based on the information available. As these tools are further developed it would be hoped that they become increasingly relevant and useful.
References
Commonwealth Department of Community Services & Health. 1987. NUTTAB87. Nutrient data table for use in Australia. Disk or tape format. Canberra: Commonwealth Department of Community Services & Health.
Frank, CG, Farris, RP & Berenson, GS. 1984. Comparison of dietary intake by 2 computerized analysis systems. J. Am. Diet. Assoc. 84: 818–20.
Greenfield, H & Wills, RBH. 1985. A bibliography of UNSW studies on the composition of Australian foods and related areas. Food Technol. Aust. 37: 460–1.
Haddy, B. 1990. Food composition data and the National Heart Foundation's Food Approval Program. Food Aust. 42:8, s14–15.
Holland, B, Unwin, ID, & Buss, DH. 1987. Cereals and cereal products. London: Royal Society of Chemistry.
Hoover, LW. 1983. Computerized nutrient data bases: 1. Comparison of nutrient analysis systems. J.Am. Diet. Assoc. 82: 501–5.
Hoover, LW & Perloff, BP. 1983. Computerized nutrient data bases: 2. Development of model for appraisal of nutrient data base system capabilities. J.Am. Diet. Assoc. 82: 506–8.
Marsh, A. 1983. Problems associated with recipe analysis. Proceedings of the Eighth National Nutrient Data Bank Conference. Washington, DC: National Technical Information Service, US Department of Commerce.
Paul, AA & Southgate, DAT. 1978. The composition of foods. London: HMSO.
Powers, PM & Hoover, LW. 1989. Calculation of nutrient composition of recipes with computers. J.Am. Diet. Assoc. 89: 224–32.
Shanklin, D, Endres, JM & Sawicki, M. 1985. Computer analysis programs. J. Am. Diet. Assoc. 85: 308–13.
Souci, SW, Fachmann, W & Kraut, H. 1981. Food composition and nutrition tables 1981/82. Stuttgart: Wissenschaftliche Verlagsgesellschaft mbH.
Tan, SP, Wenlock, RW & Buss, DH. 1985. Immigrant foods. London: HMSO.
USDA. 1989. Composition of foods. Raw, processed, prepared. Handbook 8–20. Cereal grains and pasta. Washington, DC: USDA.
