5.2: Harmo­nization of method­ological approaches to derive Nutrient Reference Values (8a.2)

An initial organizing framework for deriving NRVs, developed in the 2005 workshop(King and Garza, 2007), is depicted inFigure 8a.2.

>

Figure 8a.2: An initial organizing framework for deriving NRVs. Adapted from King and Garza (2007).

Of the two core reference values, the AR is defined as the intake needed by 50% of a pop­ula­tion sub­group (based on age, gender, and physio­logical status) to meet a specific criterion of nutrient ade­quacy. NASEM (2018) emphasize that the AR should:

• Be based on the mean nutrient intake of a specific pop­ula­tion;
• Be established for all essential nutrients and food components that have public health relevance;
• Include acceptable macro­nutrient distribution ranges for carbohydrates, protein, and fat that reduce chronic disease risks associated with the intake of these macro­nutrients;
• Consider nutrient-nutrient interactions* and quantify them, if possible; and
• Consider sub­pop­ula­tions with special needs, keeping in mind, however, that reference values are intended for appar­ently healthy** indi­viduals.

* Some nutrient-nutrient interactions are now known to alter require­ments (e.g., calcium-protein-sodium, protein-energy; vitamin E and polyunsaturated fatty acids).

** The term “appar­ently healthy” has come into question because of the implication of rising global prevalence of overweight and obesity and the corresponding increase in chronic disease risk. The NASEM (2017) report on developing dietary reference intakes based on chronic disease recommends that for future NRVs, the health status of the pop­ula­tion should be characterized in terms of who is included and excluded for each NRV.

The second core reference value — the UL — is defined as the highest level of usual daily nutrient intake that poses no risk of adverse health effects in most indi­viduals in the pop­ula­tion and is based on a toxico­log­ical risk assessment model (IOM, 1998). For details of the derivation of the AR and UL, see Sections 8a.2.4 and 8a.2.5.

Based on the deliberations of a second workshop held in 2017 that reviewed the strengths and weak­nesses of the methods currently in use, a new framework for harmonizing the process for deriving the two core nutrient reference values (i.e., AR and UL) was developed. This is presented in Figure 8a.3 that shows the four major steps required to estimate the key NRVs, with the components needed to complete the steps itemized. The four steps are:

• Choosing the appropriate tools and resources;
• Collecting relevant data from the tools and other resources;
• Identifying the best approach for the nutrient under consideration; and
• Deriving the two core nutrient reference values, the average require­ment (AR) and the tolerable Upper Limit (UL).

Figure 8a.3: Framework for harmonizing the process to derive NRVs. Modified from Russell et al. (2018b).

The feasibility of this proposed harmo­nization framework was tested using three exemplar nutrients — zinc, iron, and folate, nutrients of concern for young children and women of reproductive age; for more details, see NASEM (2018b).

8a.2.1 Choosing the appropriate tools and resources

The primary tools and resources needed to develop the NRVs are: systematic reviews, compre­hensive data­bases, and information about relevant local and regional factors that can influence the NRVs, as shown in Figure 8a.3. The EURECCA network applied the systematic review process to identify data relevant to the derivation of NRVs, using six micronutrients as examples; see Hooper et al., 2009; Van 't Veer et al., 2013; and Dhonukshe-Rutten et al., 2013 for more details. Sub­sequently, systematic reviews of bio­markers of status for vitamin B₁₂, zinc, iodine, copper, riboflavin, magnesium, vitamin D, poly­phenols, n-3 long-chain poly­unsaturated fatty acids, and selenium were published (Allen et al., 2009; Pérez-Jiménez et al., 2010; Witkowski et al., 2011). These existing systematic reviews can be updated, or new systematic reviews initiated, where necessary.

Efforts have been made to harmonize the protocols for systematic reviews (Moher and Tricco, 2008). The first step is to use the PICO/PECO model to define the search terms and concepts for the systematic review; this model is described in Box 8a.1.

Box 8a.1: The PICO/PECO Model

The elements in the model are:

• P = Pop­ula­tion: How would you describe the pop­ula­tion sub­group? What are the most important characteristics of the pop­ula­tion?
• I/E = Intervention/Exposure (I/E): What primary intervention or exposure are you considering?
• C = Comparison: What is the main alternative to compare with the intervention?
• O = Outcome: What is the out­come or effect being considered?

Modified from NASEM (2018b)

In addition to using the PICO/PECO model, construction of a predefined analytic framework (e.g., a causal pathway) is also recom­mended to help identify systematic review questions. An example for a generic analytic framework for NRVs is depicted in Figure 8a.4.

Figure 8a.4: Generic analytic framework for a systematic review of studies on the association between a nutrient and health out­comes. Modified from Russell et al. (2009) and NASEM (2018b).

The representation includes putative associations between an exposure (e.g. a nutrient) and dietary bio­markers of intake (e.g., status bio­markers such as serum or tissue nutrient concentrations, (non-validated) intermediate bio­markers (possible predictors of health or clinical out­comes), (valid) surrogate bio­markers (predictors of health or clinical out­comes), and health or clinical out­comes. The solid arrows represent established associations among factors. Line thickness represents the relative directness of an association and the strength of the relation with the health or clinical out­come. Dotted lines represent associations to surrogate bio­markers for which there is no good evidence of an association. Surrogate bio­markers are often used when the study duration is too short to show an effect on health or clinical out­come.

The analytic framework describes the relationships between “exposure” (i.e., nutrient intake) and out­comes of interest, and helps to emphasize what aspects are known and unknown. Note that the analytic framework should be modified to reflect the under­lying biological factors associated with a specific nutrient, the life-stage group, and the reference value of interest (i.e., AR or UL). For example, if the AR is the reference value of interest, then the out­come could be a clinical or health condition or a surrogate bio­marker (preferably a functional bio­marker of nutrient status) associated with deficiency of the nutrient, whereas for the UL, the clinical or health condition or surrogate bio­marker associated with nutrient excess would be selected; see Hooper et al. (2009) and Calder et al. (2017) for details on the selection of reliable surrogate bio­markers. In most cases, a single out­come that serves as a measure of exposure is selected for the AR or UL for a specific nutrient, sex, and life-stage group.

The review of the literature for the systematic review should be guided by the completed PICO elements, associated inclusion criteria, and the analytic framework for the nutrient, life-stage group, and NRV of interest. Quality assessment instruments (QAIs) must be used to evaluate the evidence at the indi­vidual study level, the choice depending on the study design. Numerous QAIs are available. SIGN 50 is an example of a QAI that can be used for both randomized controlled trials and observational studies. Alternatively RoB 2, a revised Cochrane risk-of-bias tool for randomized trials, can be used (Sterne et al., 2019).

The strength of the body of evidence generated from a systematic review must also be evaluated using QAIs. Examples include PRISMA , AMSTAR2 (Shea et al., 2017), and GRADE (Balshem et al., 2011). The latter has been used by both Australia/New Zealand and WHO in their revisions of Nutrient Reference Values. These tools have quality descriptors for criteria such as study selection and data exclusion, risk of bias, sources of funding, inclusion of nonrandomized studies, and issues around heterogeneity, etc. Application of QAIs at both the indi­vidual study level and systematic review level serves to enhance the scientific rigor and transparency of the decision- making process for deriving the NRVs. Note: nutrition-specific QAIs are under development and will be available in the future.

Readers interested in conducting a systematic review are advised to consult the Systematic Review Data Repository (SRDR). This is an open access, Web-based repository of systematic review data compiled by the U.S. Agency for Healthcare Research and Quality and is available free to users worldwide.

8a.2.2 Collecting data from the tools

The second step is to collect the data generated from the tools that are essential for selecting the bio­markers of status, surrogate out­comes, and health/clinical out­comes. Data on the dietary factors with the potential to influence nutrient bio­avail­ability and the health factors (e.g., infection) that can affect nutrient require­ments must also be included.

8a.2.3 Identifying the best approach for deriving the NRVs for the nutrient of interest

Once the relevant data have been collated, the evidence appraised and integrated using the appropriate resources, and any sources of uncertainty identified, the third step is to identify the best approach for deriving the NRVs for the nutrient under study. The decision on which approach to use depends on the availability of data, and the types and quality of studies reviewed (Section 8a.2.1). Three approaches are frequently used: factorial approach, balance studies, and an intake (dose)-response assessment. Limitations of each of these approaches were noted by Claessens et al.(2013). Table 8a.2 provides examples of the types of studies used in the development of the U.S / Canadian Dietary Reference Intakes.

Table 8a.2: Types of studies and examples of their application in the development of the U.S / Canadian Dietary Reference Intakes. Modified from Yates. (2007).

Type of study	Measurement	Examples
Nutrition intervention studies (randomized, placebo- controlled studies)	Functional out­come	Calcium fracture rate with increased calcium intake via supplements or placebo
	Biochemical measurements	Red blood cell folate response to varying levels of folate
Depletion/repletion studies	Biochemical measurements	Leukocyte ascorbate concen- trations for vitamin C Urinary excretion of 4-pyridoxic acid for vitamin B6
Balance studies	Controlled intake and excretion	Protein require­ments
Factorial estimation	Measure losses + bio­avail­ability	Iron & zinc require­ments
Epidemiologic observational studies	Estimate intake and measure losses	Iodine intake and excretion
	Functional out­come	Vitamin A and night-blindness
Observed intakes in healthy populations	Dietary intake data	Vitamin K

The factorial approach is currrently used to determine NRVs for iron and zinc, as in these cases the relationship between exposure (nutrient intake) and surrogate bio­markers or a clinical/health out­come is weak or non-existent and cannot be derived mathematically. Factorial estimates are derived from the sum of obligatory losses (i.e, through fecal, urine, skin, menses etc) plus any additional needs for growth and development (fetus, pregnancy, lactation etc), adjusted by a bio­avail­ability factor to convert the physio­logical require­ment into the dietary require­ment (Fairweather-Tait and Collings, 2010). The Average Requirement (AR) is derived from a resultant pooled estimate of needs, taking into account the bio­avail­ability (Dhonukshe-Rutten et al., 2013). State-of-the-art isotope techniques are now used to measure nutrient fluxes and “true” absorption and retention rates. The potential influence of diet- and host-related factors on nutrient bio­avail­ability was reviewed by Gibson (2007).

Balance studies are used when no reliable bio­marker representative of actual nutrient status exists. They measure input and excretion — when they are equal it is assumed that the body is saturated. Additional assumptions are that the size of the body pool of the nutrient is appropriate, and that increasing the levels of nutrient intake do not provide additional benefit. Limitations of balance studies have been reviewed by Mertz(2007). Balance studies are used to determine protein and, in some cases, mineral require­ments (e.g., calcium and molybdenum).

Intake (dose)-response modeling, usually based on randomized controlled trials (RCTs) and epidem­iological studies, describes how a known physiological out­come changes according to the exposure (i.e., intake) of a nutrient. It is used when there is a clear relationship between the exposure (i.e., intake) of a nutrient and the physio­logical relevant out­come. The latter may be a bio­marker of function, disease, or other health out­come (see Figure 8a.4). These may differ for a specific nutrient from one life-stage group to another because the critical function or the risk of disease may be different. EURECCA conducted a series of systematic reviews (Hooper et al., 2009; Van 't Veer et al., 2013; and Dhonukshe-Rutten et al., 2013)examining the intake-response relationship for several micronutrients in relation to health/clinical out­comes or surrogate out­comes associated with deficiency in order to establish a causal relationship between nutrient intake and a deficiency disease. Such relationships can be complex and influenced by many confounders and effect modifiers (e.g., ethnicity, life-stage group, body mass index, acute illness, and genotype etc). As noted earlier, surrogate out­comes are often used in RCTs when the study duration is too short to show an effect on health or clinical out­comes.

In addition to the focus on the prevention of nutrient deficiency, intake-response modeling can also be used to determine a safe upper intake level (i.e., ULs) (Section 8a.2.5) and chronic disease endpoints (Section 8a.3).

The final stage is to establish the two core NRVs for the nutrient(s) of interest and for the defined sex and life-stage group using the approach selected based on the available evidence. The two core NRVs are the Average Requirement (AR) together with the associated distribution of the require­ment, and the UL. Strenuous efforts should be made to establish the AR for each nutrient because it is the basis for the multiple reference values; details are given below.

8a.2.4 Deriving the Average Requirement (AR) for a nutrient

The require­ments for a specific nutrient vary from indi­vidual to indi­vidual and thus form a distribution of require­ments. For most nutrients except iron, this variation in require­ments is assumed to follow a normal symmetrical distribution as shown in Figure 8a.5, valid for a defined level of nutrient ade­quacy and a specific sex and life-stage group. Consequently, a coefficient of variation (CV) (i.e., the standard deviation divided by the mean × 100%) can be used to estimate the variation. In cases where require­ments are not normally distributed (e.g., iron require­ments for mens­truating adolescent girls and women of child bearing age), data are transformed to achieve normality.

Figure 8a.5: Frequency distribution of the indi­vidual require­ments (mg) of nutrient X in women 30–50y, reflecting variability in require­ments among indi­viduals. Adapted from King et al. (2007).

Inter-indi­vidual variability in require­ments is affected by numerous diet- and host-related factors (Gibson, 2012). However, many of these factors are difficult to measure, or their impact is unknown. Hence, for many nutrients, the distribution of require­ments is unknown, and instead is assumed to have a CV of about 10% (i.e., the standard deviation is about 10% of the mean require­ment), assuming a normal distribution (King et al., 2007). EURRECA explored the biological variation in require­ments associated with single nucleotide polymorphisms (SNPs) on the metabolism of five micronutrients (Claessens et al., 2013) but to date no information related to SNPs has been used in the derivation of NRVs due to a lack of relevant data.

The median of the require­ment distribution represents the Average Requirement (AR) for that particular group of indi­viduals (Figure 8a.6) and is used to assess the prevalence of nutrient ade­quacy within that group (see Chapter 8b for more details). Therefore, by definition the AR is:

“the amount of nutrient that is estimated to meet the nutrient require­ment of half the healthy indi­viduals in a specific life-stage and sex group”.

Figure 8a.6: Average require­ment (AR) for a nutrient. The nutrient require­ments are defined in relation to a frequency distribution of indi­vidual require­ments. RI or the equivalent is defined as two standard deviations above the AR.

8a.2.5 Deriving the Safe Upper Levels of Intake

The Safe Upper Levels of Intake (ULs) represent daily intakes, that if consumed chronically over time, will have a very low risk of causing adverse effects. Intakes from all sources are considered: food, water, nutrient supplements, and pharmacological agents. The UL is defined as:

“The highest level of habitual nutrient intake that is likely to pose no risk of adverse health effects in almost all indi­viduals in the general pop­ula­tion (King and Garza, 2007).”

The ULs are based on a toxico­log­ical risk assessment model involving a four-step process:

Box 8a.2: Four-step toxicological risk assessment model

• identification of risk of toxicity;
• dose-response assessment;
• assessment of the prevalence of intakes outside the reference values; and
• characterization of risks associated with excess intake

(WHO/FAO, 2005).

The dose-response assessment is built upon three toxico­log­ical terms: no-observed-adverse-effect level (NOAEL), lowest-observed-adverse-effect level (LOAEL), and uncertainty factor (UF). These terms are defined in Box 8a.3.

Box 8a.3: Dose-response assessment

• NOAEL is the highest continuing intake of a nutrient at which no adverse effects have been observed in the indi­viduals or groups studied. In some cases, it may be derived from experimental studies in animals. When sufficient data are not available to define a NOAEL, then a LOAELis defined.
• LOAEL is the lowest continuing intake at which an adverse effect has been identified.
• UNCERTAINTY FACTORS (UFs) are used to address gaps in the data and incomplete knowledge (e.g., variability in response within the pop­ula­tion; extrapolation from experimental animal to human data). The estimate for the magnitude of the uncertainty factor for each nutrient should be based on the approach of Renwick et al. (2004).

From (IOM, 1998).

At present, ULs are only set for those nutrients for which there is strong, high quality evidence. Like ARs, ULs vary by age and sex and tend to be lower for young children and pregnant women. The goal is to have less than 5% of a pop­ula­tion sub­group with an intake greater than the UL, including intakes from supplements and fortified foods (IOM, 1998). This goal can sometimes be challenging for pop­ula­tion sub­groups consuming fortified cereals (Arsenault and Brown, 2003). For more details on the derivation of ULs, refer to WHO/FAO (2005).