Estimating Food Composition Data |
|
| | Updated 2015-08-15 |
| Deriving food composition data for food components where only few analytical values exist |
Compilation of food composition data often leaves the data compiler
with the headache of many missing values, especially for foods that are
rarely analysed or eaten. In order to avoid missing values, the compiler
is therefore forced to estimate or guess the value ("guestimate"). It is
important that the estimated values are created on a sound basis and
documented in such a way that the estimation can be tracked.
One of the basic papers on the estimation of nutrient values is
- Schakels S., Buzzard S., Gebhardt S.E.:
Procedures for Estimating Nutrient Values for Food Composition Databases
Journal
of Food Composition and Analysis 10, 102–114 (1997)
It gives a good and thorough description on how to estimate nutrient
values when analytical data are not available. The authors basically
divide the subject into six main categories
- Using Nutrient Values from a Different, but Similar, Food
- Calculating Nutrient Values from a Different Form of the Same Food
- Calculating Nutrient Values from Other Components in the Same Food
- Calculating Nutrient Values from Household Recipes or Commercial Product Formulations for Multicomponent Foods
- Converting Nutrient Values from Nutrient Label Information of a Commercial Food Product
- Calculating Nutrient Values from a Product Standard
and give an in-depth description of the process of validating the
estimated values as well as documenting and referencing estimated
values.
| The Biologically or Technically "Standardised" foods |
For some foods/food groups there are reasonable and sound solutions to overcome the challenge of estimating values. These foods are characterized by the fact that the amounts of some components is
related to the content of other components.
Especially in foods of animal origin such relationships can be
established without much effort using using common
mathematical/statistical tools, like linear regression. For example,
there are completely linear relationships between the dry matter
(moisture) content and the fat content in meat - a relationship that is
widely used in the meat industry as the fat content can be estimated by
analysing the dry matter content in the meat, a much simpler (and
cheaper method).
The biologically or technically "standardized" foods are foods that from
nature or via processing contain certain amounts of components - and
often - there is an (internal) relationship between several of the
components.
The foods in this category are raw and prepared meats, i.e. meat from
animals like cattle (beef, veal),
pigs (pork) and
sheep (sheep, lamb).
Similarly, also
dairy products (milk and cheeses) show such relationships. Less
apparent, but still existing are also the relations between components in
fish.
The challenge is to find these internal relations in the foods,
e.g. with the help of linear regression models (single or nultiple). On the underlying pages, some of these relations are described.
The fact that we are working with calculated/estimated data values
and not actual results of chemical analysis calls for extreme caution.
First of all, the calculated values must represent values that are
realistic in the compiler's environment, i.e. the relationships from
which the values have been calculated should be derived from analytical
values representative for the compilers' environment as far as possible.
Furthermore, the relationships/hypotheses on which the estimations are
based should be checked by actual chemical analysis of samples, if
possible.
- Schakels S., Buzzard S., Gebhardt S.E.:
Procedures for Estimating Nutrient Values for Food Composition Databases
Journal
of Food Composition and Analysis 10, 102–114 (1997)
Compilers' Toolbox™
