The many applications of estimates of energy and nutrient requirements can be grouped into two main categories. In diagnostic applications the estimates are used to judge the probable adequacy or inadequacy of observed intakes. In prescriptive applications the estimates are used to suggest what intakes should be. Although these uses may seem quite similar, and in some situations may be almost identical, it has frequently been pointed out in this report that for some purposes different estimates of energy requirement might be used (e.g., assessing the existing situation given existing body sizes and activity patterns versus predicting the energy needs that would be associated with a desired new situation).
Depending upon the mode of application chosen, the user may attach different degrees of interpretational confidence to some of the assumptions made in deriving estimates. If is beyond the scope of this report to consider these differences in detail; the user might consider them in the context of particular uses.
All applications involve consideration of both requirement estimates and food intake. Three obvious conditions must be satisfied for such comparisons to be meaningful:
In order to meet these conditions, a number of statistical and biological principles need to be taken into account. It is the purpose of this section to draw attention to these issues, and again it is recognized that the preferred approach to their resolution may vary with the user's specific application.
Protein and energy raise different biological and statistical questions and therefore have to be treated separately. In the following discussion the various application issues are grouped under specific headings, and within each section protein and energy are treated sequentially but in parallel. Many of the principles highlighted in this section have also been presented in section 2 and have been amplified in various sections of the report.
Food intake data may be available at the level of the individual, the household (or institution), or the population as a whole. For purposes of comparison, requirement estimates must be derived or adjusted to the same level of aggregation—requirements of individuals, households, or populations. When comparisons are made at higher levels of aggregation, it must be recognized that without further information, no inferences are possible about lower levels (e.g., about individuals in the household). To make such inferences, assumptions must be made about the nature of the distributions. The present discussion refers only to the aggregation of requirement estimates. The question of distribution models is discussed in section 11.5.
It was pointed out in section 2 that requirement estimates are derived from studies on individuals. Their application, then, should relate most closely to individuals; at other levels of aggregation the estimates must be derived. It was pointed out also that there is variation in requirements among seemingly similar individuals and hence interpretation must be on the basis of probability. This implies that there is a variance attached to each estimate of requirement and that any application must take into account this “error term”. The issues to be addressed under the heading of levels of aggregation relate to the manner in which these variance estimates should be handled and the impact this has on biological interpretation.
If information is available about an individual's usual intake of utilizable protein per kg of body weight, the interpretation and application of requirement estimates is relatively straightforward. As discussed in section 2 and shown in Table 54, it is possible to make a probability statement about the adequacy of the intake to satisfy the criteria that have been used to define requirement (section 5). If data are available for a large number of individuals, similar or dissimilar, then the probability statements generated for each individual can be used to predict the prevalence of inadequate intakes. Clearly, although this may suggest how many individuals have inadequate intakes (intakes below their true requirements) it will not identify which individuals have inadequate intakes. For the purpose of prescription, one can recommend levels of intake that carry varying degrees of confidence that they are adequate for the random individual. The current convention included in the concept of the safe level of intake is that the probability of adequacy should be 0.975 (adequate for all but 2.5% of individuals).
Table 54 is appropriate when protein intakes are expressed per unit body weight, the same unit of description as the estimate of protein requirement. If intakes and requirements are expressed as g per person per day without reference to the individual's actual body weight, then the variance of the requirement estimate will be increased and the intervals shown in Table 54 will have to be adjusted accordingly. The method of deriving the new variance estimate is presented in Annex 9(B).
|Observed intake of an individual in relation to requirement||Proportion (%) of a group of similar individuals whose intake is inadequate||Probability that an individual's intake is inadequate|
|25% above average requirementb||2.3||0.02|
|12.5% above average requirement||15.9||0.16|
|At average requirement||50.0||0.50|
|12.5% below average requirement||84.1||0.84|
|25% below average requirement||97.7||0.98|
a See text for description of approach when intakes are expressed as g/day.
b This level of intake is described as the safe level of intake. Accepting CV of requirement = 12.5%.
When the household is the unit of observation for food intake, and when demographic and anthropometric information is available about household members, it is necessary to generate an estimate of the average requirement of the household unit and of the variance associated with this estimate. Conceptually, average requirement here means the average that would be expected from a large number of similar households; the requirement of the individual household falls within a distribution of requirements among similar households. An appropriate description of that distribution must be made (mean and variance) before interpretations can be drawn.
The mean of the household requirement distribution will be the sum of the average per caput requirements for each of the individuals within the household. If there is no reason to believe that there is any correlation between the requirements of individuals within the household, then the variance of the distribution may be estimated as the sum of the variances associated with the requirement estimates of the individuals. However, if there were evidence of a genetic influence on requirement, such that related individuals within a household might occupy similar relative positions in their requirement distributions (e.g., all have high or low requirements) then this would have to be taken into account. The effect would be to raise the estimated variance of the household unit. At present there is no direct evidence on this point.
Table 55 presents a hypothetical example of the development of estimates of average requirement and associated variance for a household unit. This example assumes no correlation of requirements among individuals within the household. It illustrates the general phenomenon that would be expected: the variance of the requirement of the household unit, expressed in proportion to the mean requirement, is less than that of a group of individuals (CV of 6% rather than 12.5%). If positive correlation of requirements exists, it would tend to move the estimate of household variability towards that found for individuals.
With these new estimates, the approach shown in Table 54 can be applied to the household unit (adjusting the intervals in the table in accordance with the new CV). Probability statements can be made about the adequacy of the intake of the household unit, and a safe level of intake for the household unit can then be derived (it would be 142 g/day rather than the 158 g/day that would be obtained by adding the safe levels of intakes for individuals).
Although these estimates and probability statements can be derived mathematically, the question is, what do they mean biologically? How should one conceptualize a household unit and attach meaning to the adequacy of intake of the unit? Clearly there is no implication that all members of the household share the same status as the household unit; a low probability that protein intake is inadequate for the household unit does not mean an equally low probability of inadequate intake for all members of the household. It could be argued that across a large number of households, the estimates of prevalence derived from the probabilities for individual households would reveal the prevalence of inadequate intakes among individuals (not clustered within specific households). This argument would not hold if there is any reason to believe that behavioural factors systematically influence the distribution of intakes within households1. Information on this subject is hard to obtain. Some data are available from Papua New Guinea on food distribution between members of the household and differences in the mixture of foods consumed(1).
1 One such factor may be the relative energy needs of individuals within the household. It is reasonable to assume that most individuals in the household shown in Table 55 would consume the same mixture of foods but in a different total quantity, in proportion to their energy needs. It is then reasonable to suppose also that protein intake within the household will be distributed approximately in proportion to energy needs. Energy need and protein requirement are not linked together across individuals. It should not be expected that protein intake will be distributed in proportion to protein requirement (i.e., systematic factors may be expected to operate). This may suggest the need for more elaborate analytical models before biological interpretations can be made. (See also discussion of the PE ratio in section 10.)
When the unit of observation of intake is the total population, the average requirement estimate and the associated estimate of variance can be derived by rather complex calculations. For total populations, these variances (reduced to the expression of CV) would be expected to be relatively small. The most difficult problem is not in obtaining these estimates, but rather in interpreting them once calculated. How does one conceptualize a population unit rather than the individuals that make up that population? For protein, this issue is so complex that there would seem to be no reason to carry out the calculation.
|Household member||Protein requirement|
|Child, 10 years||30||23||12.5||8.27|
|Child, 7 years||24||19||12.5||5.64|
|Child, 4 years||15||12.5||12.5||2.43|
a Mean and variance (standard deviation2) of distribution of requirements for specified class of individual with known body weight.
An assumption that underlies most of the discussion on the interpretation of protein intake in relation to protein requirement is that, after controlling for body weight, there is very little correlation between intake and requirement. This assumption is supported by limited experimental data (2). For energy, there is considerable empirical evidence to suggest that energy intake and energy requirement (here defined as actual rather than desirable energy expenditure) are positively correlated. This may be seen as either a regulation of intake to satisfy energy need or an adjustment of expenditure to match intake or a combination of both. The existence of this correlation makes the interpretation of energy requirement different from that of protein. In this section emphasis will be given to differences in interpretation; procedures of aggregation of requirements are generally similar for protein and energy.
A second important feature of energy requirement is that the physical activity of the individual as well as age, sex, and body size, affect the requirement so that, in contrast to protein, there is another source of variation of requirement that is of major importance.
At the level of the individual, because of the correlation expected between energy intake and requirement, the simple approach shown in Table 54 does not apply to energy. In theory, if the correlation can be reasonably approximated, statistical estimates of the probability of inadequacy and prevalence can be generated. In practice, at the level of the individual it is feasible to draw only the most generalized interpretations from comparisons of observed energy intake and estimated requirement. Perhaps one of the most common errors in the interpretation of dietary surveys has been the direct comparison of intakes with estimates of average energy requirements and classification of the intakes as adequate or inadequate on this basis.
At the level of a group of individuals, each of whose dietary intake is known, it is appropriate to compare mean energy intake and mean energy requirement. If the correlation between intake and requirement is strong, the two distributions should be similar and the two means close. A departure of mean intake from mean requirement suggests inadequate or excess intake for the group and perhaps for most individuals within the group (depending upon the strength of the correlation).
For energy, it is difficult to go further towards obtaining estimates of the prevalence of inadequacy. The extent of correlation is one problem; another is the magnitude of the variance of requirement, unless there are detailed studies of the actual activities of the individuals, which is unlikely. The estimate of the variability of energy requirement in section 4.5 refers to the variability of BMR and perhaps also to the approximate variability of energy expenditure for people performing known activities. It is an underestimate of the variability of requirement for people of differing body weights performing different activities. For children, where energy needs are expressed per unit body weight, the approach would be analogous to that suggested for protein. For adolescents and adults, the estimate of energy requirement and its associated variance is generated for a specific body weight, since the requirement per kg changes with body weight. Therefore a different statistical approach has to be developed. Annex 9(C) suggests the potential impact of variations in activity on the variance estimate, when the individual's actual activity is not known.
If normative judgements are incorporated into the estimation of energy requirement—that is to say, desired levels of activity rather than actual levels (section 4.2)—then the statistical considerations of correlation and variability become even more difficult. However, comparison of mean intake and mean normative requirement may still be useful in the prescriptive type of application or for judging “shortfalls”. It is emphasized again that in making any assessment of observed energy intake, it is essential to specify the actual or expected level of physical activity. Any statement about the adequacy of observed intake must be qualified—adequacy to do what?
In the same way as for protein, the mean energy requirements of individuals can be added to obtain an estimate of the mean energy requirement for a particular household. The variance of this estimate can be examined using the variance estimates for the individuals (for the adults, an average activity profile might have to be adopted and the associated variance estimate increased—see Annex 9(C). For reasons already stated, it is just as difficult to make probability statements about a household unit as it is about an individual. It is only reasonable to assume that a moderate correlation will exist between food intake and energy need of the household unit.
If information is available about multiple household units, as in a community, then comparison of the mean intake of the group of household units with the mean requirement of the group may be informative, the interpretation being analogous to that for groups of individuals. In contrast to the situation with protein, there is more reason to assume a correlation between the distribution of energy intakes and requirements within the households provided that total household intake is not unduly constrained. Nevertheless, until a judgement is made about the likely strength of that correlation, very limited inferences can be made about the situation of individuals within the household.
When the population is the unit of observation for food intake, an estimate of the mean energy requirement can be obtained from demographic, anthropometric, and activity profile data. As for protein, its interpretation is very doubtful. It is known that the distribution of intakes within populations is not uniformly proportionate to need; acute malnutrition exists in populations that appear to have sufficient food to meet the estimated energy needs for the country as a whole. Even though a correlation exists between intake and requirement, it is not perfect and inferences cannot be made about the situation of individuals from a knowledge of intake and requirement at the population level.
For energy more than for protein, the aggregate requirement estimate at the population level may be a useful marker in studies of trends. It may be a meaningful way of taking account of demographic changes and differences in comparisons across populations. It is not a useful index of satisfaction of need or a meaningful target for production.
In section 2 it was emphasized that the requirement estimates referred to amounts needed over moderate periods of time. Although, in keeping with current conventions, requirements are expressed as a rate of intake per day, there is in general no implication that these intakes are to be ingested each day. The requirement estimates do not take account of short-term, intra-individual variation although this variation does, of course, increase the difficulty of measuring the requirements. In many societies there are also longer-term seasonal cycles that affect requirements. A clear example of this would be the cycle of agricultural labour and its effect upon energy requirements in agrarian societies. These cycles, and the human response to them, may be of great importance in assessing the situation of individuals and families living in these areas. It follows that the time-frame of examination should be such that these cycles are revealed rather than masked. Much necessary information may be lost if food intake data are available only at the population level, or if survey data for different individuals or households collected over a span of a year are treated as a single sample without examination for seasonal or other cyclical effects.
The other aspect of time-frame studies relates to the methods of measuring intakes. There must be assurance that the measurements provide estimates of “usual” intake; that is, intake persisting over reasonable periods of time, rather than intake on a particular day. The problem here is one of intra-individual variation in intake. A number of studies in industrialized countries suggest that the coefficient of variation of daily energy intake in the same individual is about 25% of the mean (3–8). Unpublished studies suggest a similar order of magnitude for rural populations consuming a monotonous diet. If the error of the estimate of usual intake is not reduced by repeated observations (increased numbers of days observed), there can be serious errors in applying the probability approach to assessment of intake (misclassification error). The problem is similar for protein, although the intra-individual variation may be slightly lower than for energy.
Section 7 provides a discussion of the effects of digestibility and amino acid composition on the utilization of dietary protein and of digestibility on the utilization of dietary energy.
When intakes are being compared with requirements, there are two possible ways of applying these corrections. One way is to adjust the estimates of requirement, according to the actual foods consumed. Thus if the digestibility of the protein in a particular diet is 90% that of milk or egg protein, the requirement would be increased by a factor of 100/90. This was the approach taken in the report of the 1971 Committee (9; e.g., in Table 25). The second way is to adjust the intake data to describe effective intake. Thus in the example given above, the effective intake would be 90/100 multiplied by the observed intake. This approach may be more logical; it would make it easier to aggregate information from different diets that require different correction factors.
When intake data are collected at a level other than that of the individual, a problem arises in adjusting for amino acid score. In theory this adjustment should be made almost on a meal-by-meal basis, although there is evidence that complementation of amino acids persists over a period of hours; certainly it would be desirable to do it on at least a day-by-day basis; it must also be done on an age-specific basis (section 7.3). If intake data are collected at the household level it may be almost impossible to adjust the observed protein intake according to the part consumed by specific individuals. In this situation, the only feasible approach may be to adjust the requirement estimate for the affected age groups according to the composition of the mixed foods consumed by the household unit.
In the preceding discussion prominence has been given to the very real limitations in interpreting comparisons of intake and requirement at the household and population levels. It has been emphasized that no inferences can be made about the adequacy or inadequacy of intakes at lower levels of aggregation unless some assumptions are made about the distributions of intakes (and requirements) within the unit observed. It is this problem that has given rise to attempts to model these distributions, and attempts to improve the models will undoubtedly continue. It is hoped that some of the points discussed in this section, emphasizing the need to consider both statistical and biological principles, will be helpful for this purpose.
In such models numerous assumptions have to be made, inferred from other information such as income distribution. It will often be too expensive to test all the assumptions, but it should not be too expensive to carry out sensitivity testing, to determine which assumptions are critical to interpretation and to identify those assumptions that warrant more direct examination through appropriate data collection and analysis. A priori emphasis is given to the need for more information about the nature and determinants of the distribution of intra-household food intake. Assumptions about this are an implicit or explicit part of all models that rely on studies of household or population intake, yet there appears to be very little information on which to base these assumptions. Clearly, there is also a need for more information about patterns of activity, in terms of categories of individuals and situations that are of interest to planners and other users.