National Academies Press: OpenBook

The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary (2008)

Chapter: 3 Criteria for Scientific Decision Making: Session 2

« Previous: 2 Conceptual Framework for DRI Development: Session 1
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 63
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 64
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 65
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 66
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 67
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 68
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 69
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 70
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 71
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 72
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 73
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 74
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 75
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 76
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 77
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 78
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 79
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 80
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 81
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 82
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 83
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 84
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 85
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 86
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 87
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 88
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 89
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 90
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 91
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 92
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 93
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 94
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 95
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 96
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 97
Suggested Citation:"3 Criteria for Scientific Decision Making: Session 2." Institute of Medicine. 2008. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary. Washington, DC: The National Academies Press. doi: 10.17226/12086.
×
Page 98

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

3 Criteria for Scientific Decision Making: Session 2 During the planning phase of the workshop, Session 2 participants were requested to take into account the same general questions asked of Session 1 participants (see Box 2-1). However, they were specifically asked to each address different decision-making criteria important to the develop- ment of Dietary Reference Intakes (DRIs). These criteria are components of the “road map” for DRI development as described earlier (see Chapter 1). General questions were asked of each participant: Can we provide more specific guidance to study committees on scientific decision making to help clarify the concepts and tasks and to promote consistency across study committees? Can we provide guidance to study committees on the use of scientific judgment in the face of limited data that would allow such judg- ment to be more transparent and better documented? The second session was moderated by Dr. Robert Russell of Tufts Uni- versity. Dr. Irwin Rosenberg, also of Tufts University and former chair of the Food and Nutrition Board (FNB), opened the session with a talk on the selection of endpoints. Dr. Susan Taylor Mayne, a professor in the Division of Chronic Disease Epidemiology at the Yale School of Public Health, then spoke on the options available in the face of limited dose–response data. Dr. Stephanie Atkinson, a professor in the Department of Pediatrics at McMaster University, discussed the challenges in addressing extrapolations and interpolations for unstudied groups. Dr. Hildegard Przyrembel, from   This chapter is an edited version of remarks presented by Drs. Rosenberg, Mayne, Atkinson, Przyrembel, Subar, and Garza at the workshop. Discussions are composites of input from vari- ous discussants, presenters, moderators, panelists, and audience members. 63

64 THE DEVELOPMENT OF DRIs 1994–2004 the Federal Institute for Risk Assessment in Berlin, spoke on the challenges in addressing adjustment for data uncertainty. Dr. Amy Subar, a research nutritionist at the National Cancer Institute (NCI), gave a presentation on the implications of estimating dietary intake for DRI development. Finally, Dr. Cutberto Garza, provost and dean of faculties at Boston College and former chair of the FNB, closed the session with some highlights of physiological, genomic, and environmental factors that are important to the DRI process. Discussions and comment periods were held throughout the session. SELECTING ENDPOINTS: WHAT ARE THE ISSUES AND WHAT ARE THE OPTIONS FOR CRITERIA? Presenter: Irwin Rosenberg Endpoints play a pivotal role in the DRI process. They are the skeletal structure on which the Estimated Average Requirements (EARs) and toler- able upper intake levels (ULs) are draped. In essence, they are an expres- sion of the targets or goals of the DRI development process. They should be related to quantifiable or measurable attributes that relate to the overall public health goal of the project. The key concerns from the perspective of selecting endpoints are “adequacy for what ends?” with respect to the EARs and “adverse effects as reflected by what?” for the ULs. Experience in Selecting Endpoints Since the 1941 National Research Council (NRC) report (1941), the selection of endpoints for nutrient reference values has evolved in response to changes in nutrition science. These advances sometimes revealed associa- tions between an endpoint and diet and at other times identified possible endpoints through better understanding of metabolic and physiological states. Moreover, approaches for endpoint selection have been variable across the study committees responsible for the reference values. This is to be expected, given the differences in the biology and functions of essential nutrients. Throughout the experience of developing reference values, limited data have often precluded the identification of the most appropriate endpoint for any given age/gender category. This situation in many cases results in the need to extrapolate knowledge about the endpoint used for one group that is better studied (e.g., adults) to a less well-studied group (e.g., children). This is one area of work that needs further exploration (see presentation by Dr. Atkinson in this chapter). Importantly, limited data on dose–response relationships have always

CRITERIA FOR SCIENTIFIC DECISION MAKING 65 made it difficult to compare, consider, and prioritize endpoints for the pur- poses of establishing a reference value. The reliance on studies that exam- ined dose ranges not relevant to adequacy considerations is not a desirable solution. Although meta-analysis studies offer some promises and newer strategies are being developed to deal with limited data (see next presenta- tion by Dr. Mayne), the ideal situation is to have better data. As we have experienced during the past 10 years, endpoints for specific chronic diseases are especially challenging. Although it is desirable to have chronic diseases as the targets for our requirements and thereby reference values, that was possible in only a few instances. However, we need to recognize that the use of chronic disease as a basis for reference values is not a new paradigm. Throughout the history of the Recommended Dietary Allowances (RDAs), chronic disease has been an implicit part of trying to set reference values that were above those necessary to prevent deficiency. The idea of achieving the health of the population—and thereby including the risk of chronic disease as an endpoint—has always been present at some level within the process. Whether this can be done explicitly, as was done for some macronutrients in the last series, will require further discussion. Finally, I would like to make a few quick points on the lessons we are now considering. First, one question raised to workshop participants is the issuing of multiple reference values based on multiple endpoints for a single age/gender group. This is not the issue of study committees considering multiple endpoints before they select one to serve as the basis for refer- ence values, but assigning them the task of issuing values for the various endpoints. Specifying multiple endpoints for a nutrient within a given age/ gender group is not useful or appropriate; in fact, it could be very confus- ing. Rather, a single endpoint for the age/gender group should be selected. Second, the question of whether reference values—EARs, RDA, ULs—are to address essential nutrients only or be expanded to nonessential nutrients, such as fiber and carbohydrate, needs to be considered, particularly in light of our understanding about the interface between the DRI process and food-based dietary guidance. Selecting Endpoints In the past, a number of endpoint types have served as the basis for reference values. These have included clinical signs, measures of develop- mental abnormalities in children, biochemical measures, balance study out- comes, body pool measures, functional measures, and measures of chronic disease risk. A 1994 Institute of Medicine (IOM) document (1994) lists the types of evidence that have been used in establishing RDAs. These include

66 THE DEVELOPMENT OF DRIs 1994–2004 • biochemical measurements that assess the degree of tissue saturation or adequacy of molecular function in relation to nutrient intake; • nutrient depletion and repletion studies in which subjects are main- tained on diets containing marginally low or deficient levels of a nutrient, and then the deficit is corrected with measured amounts of that nutrient; • balance studies that measure nutrient status in relation to intake; • epidemiological observations of populations in which the clini- cal consequences of nutrient deficiencies are corrected by dietary improvement; • extrapolation from animal experiments (although applying animal data to human studies is difficult); and • nutrient intakes observed in apparently normal, healthy people, which was one way of arriving at an Adequate Intake (AI). If one views the stages of nutrient insufficiency as a series or cascade of events that describe the temporal sequence of deficiency of, for example, a given vitamin, the initial stages could be called “subclinical deficiency,” or findings that would occur before symptoms or signs of disease (e.g., low circulating levels of nutrient, decreased tissue levels or desaturation of body pools, and metabolic disruption). The more advanced stages of deficiency, which could be called “clinical deficiency,” encompass the symptoms and/or signs of disease (e.g., reversible changes in the skin and irreversible changes or cell death). An emerging area important to the criteria for selecting an endpoint is the ability to use a biomarker or surrogate as an endpoint reflective of the functional or clinical response of interest. I will conclude my remarks by reviewing the case of vitamin D. The vitamin D case is an interesting example because circulating levels of 25-hydroxy­vitamin D have been shown to be related to intake of vita- min D. Although this is complicated by synthesis in the skin as a result of sun exposure, it is generally a good measure of absorption of vitamin D. However, data may be emerging that relate levels of 25-hydroxyvitamin D to measures of bone density, skeletal disease risk (as in the case of os- teoporotic fracture), and other disease risk (as in the case of extraskeletal cancer and even some immune dysfunctions). Moreover, evidence that 25- hydroxyvitamin D is related to the absorption fraction for calcium suggests that 25-hydroxyvitamin D values have the potential to serve as a target endpoint for an important function and may demonstrate certain conver- gence with other observations—for example, a lower risk of several kinds of cancer, at least in some intervention studies. A regression meta-analysis reported by Bischoff-Ferrari et al. (2005) shows that in a number of studies, a significant decrease in relative risk

CRITERIA FOR SCIENTIFIC DECISION MAKING 67 FIGURE 3-1  The effects of vitamin D supplementation on hip fracture and non- vertebral fracture. NOTE: CI = Confidence Interval. SOURCE: Bischoff-Ferrari et al. (2005). Copyright © (2005), American Medical Association. All rights reserved. of hip fracture is observed in the area of 75–85 nmol/L (Figure 3-1). This raises the question as to whether it is possible to find biological markers or endpoints of this kind that will show a convergence of effects, where multiple goals of preventing fracture and perhaps contributing to the pre- vention of chronic disease can be embodied quantitatively in an endpoint. Many avenues need to be pursued to better specify the selection of endpoints for reference values. This presentation has elaborated on some that may be useful and suggested that certain paths will be more fruitful than others. However, we must remember that one set of criteria or even an algorithm is unlikely to be “one size fits all” because there may need to be different approaches for different nutrients and types of reference values. This process will be an evolution that must be carefully planned. General Discussion A participant commented that the RDA and UL values are frequently close together because a UL is often established using an endpoint that

68 THE DEVELOPMENT OF DRIs 1994–2004 occurs at a low level of intake for public health safety purposes, whereas endpoints selected for adequacy-based reference values tend to be those that occur at higher levels of intake. In response to the participant’s question of whether this should be done in the future, Dr. Rosenberg replied that there should be even more collusion in the process of setting EARs/RDAs and ULs, especially because awareness of the margin between the two values is important. The decision should be driven by scientific data rather than the rote conclusion that the ULs must be as low as possible and the EARs/RDAs as high as possible. An audience member remarked that it would be useful if the DRI study committees considered endpoints more comprehensively within their reports: For example, vitamin C at level X prevents scurvy and at level Y impacts another endpoint of interest. There would be different endpoints for the same nutrient, but endpoints more important in societies other than North America would not be neglected. An audience member commented on Dr. Rosenberg’s pessimism about using disease risk reduction for certain recommendations given the im- portance of reducing the risk of disease as an overall health goal. It was stated that there is a numerical relationship between fiber intake and the onset of cardiovascular disease (CVD). Another participant commented that measures of fiber intake from observational data can be a marker for other dietary and behavioral patterns and therefore may be problematic as a basis for setting DRIs. In response, Dr. Rosenberg noted that there will be instances when there is a direct relationship between dietary intake and a chronic disease response, but in many cases it will be difficult due to lack of specificity and confounding factors. He emphasized the importance of focusing on intermediate or surrogate markers predictive of disease out- come as a way of ensuring a focus on chronic disease risk reduction. A brief discussion took place regarding the process for validating biomarkers for disease. Dr. Rosenberg emphasized the need for sound science and clear validation. DOSE–RESPONSE DATA: ARE THERE OPTIONS FOR DEALING WITH LIMITED DATA? Presenter: Susan Taylor Mayne A more challenging aspect of the DRI process is dealing with limited data on dose–response relationships. The DRI process depends on dose– response data for both EARs and ULs. Even if there are extremely limited data on dose–response for many nutrients, DRI study committees need to establish numeric values. As a consequence, some DRI values are “softer” in reality than what might be expected. This is well illustrated using the

CRITERIA FOR SCIENTIFIC DECISION MAKING 69 example of the dose–response data that were available in establishing the EAR for selenium. Dose–Response Data and the Selenium Estimated Average Requirement (EAR) The study committee considered several possible endpoints or biomark- ers for selenium status, ranging from disease endpoints (e.g., Keshan disease and cancer) to blood or plasma selenium levels to plasma selenoprotein concentration as a biomarker of selenium status. The study committee ultimately chose plasma selenoprotein concentration maximization as the biomarker. Two studies that evaluated maximization of plasma selenoproteins in response to supplemental selenium were available. One was a study of 52 men and women from New Zealand (Duffield et al., 1999), and the second was a study of 45 men from China (Yang et al., 1987). Both populations had low selenium intake. In the New Zealand study, the baseline selenium intake of the subjects averaged 28 µg/day (for comparison, U.S. intakes are about 100 µg/day). Groups were given five different levels of selenium per day for 5 months: 0, 10, 20, 30, or 40 µg/day. The endpoint being monitored was plasma selenium-dependent glutathione peroxidase. All of the groups receiving ad- ditional selenium were found to have increased glutathione peroxidase, but they could not be distinguished from one another due to large variations in response. Because the variation was so large, a dose–response could not be calculated. Instead, the investigators decided that the lowest added intake, 10 µg/day, may be sufficient, so they set an EAR of 38 µg/day, which is the baseline intake of 28 µg/day plus 10 µg/day. In the Chinese study, the baseline selenium intake of the subjects was even lower, 11 µg/day. Groups were given five different selenium doses for 8 months: 0, 10, 30, 60, or 90 µg/day. Although it was difficult to determine a dose–response based on the limited sample size, it was estimated that average maximization was achieved at the added intake of about added 30 µg/day. This gave an EAR of 41 µg/day when combined with the baseline intake of 11 µg/day. With weight adjustment to reflect North American body size, the EAR was increased to 52 µg/day. The IOM study committee simply averaged these two numbers (38 and 52 µg/day), resulting in an EAR of 45 µg/day. As the variation data were difficult to calculate, a coefficient of variation of 10 percent was assumed, and the Recommended Dietary Allowance was set at 55 µg/day. As discussed above, the EAR for selenium was based on fewer than 100 subjects. Dose–response data anywhere in the world were very limited. The only available data were obtained from selenium-deficient populations

70 THE DEVELOPMENT OF DRIs 1994–2004 from outside North America. Important questions are: How relevant is this EAR to the United States and Canada? Are there alternative techniques that we should be employing to try to characterize dose–response using more relevant and statistically powerful data? Solutions to the problem of limited dose–response data can be grouped into two general approaches. The first is the statistical or modeling ap- proach, which applies various models to try to characterize dose–response, such as in relation to chronic disease or mortality (e.g., a large cancer pre- vention trial in the United States with 35,000 men randomized to selenium supplementation or a placebo). The second approach is the biological ap- proach. Both approaches are described below. The Statistical Approach The advantage of the statistical approach is that many studies with large sample sizes are available (both observational and clinical trials). One disadvantage is that the intake data in these large population studies are often susceptible to measurement error. This is nutrient specific; for ex- ample, the intake data are not of good quality for vitamin E and selenium. However, in many of these same studies, we can examine plasma nutrient status as a biomarker for chronic disease risk to estimate the dose–response, which can then be related to intake data using metabolic or other relevant studies. Different statistical approaches are used to analyze nutrients in rela- tion to chronic disease risk. The traditional single-study approach is where one examines nutrient intake or status in relation to a chronic disease end- point. The typical approach is to quantile the intake or status data, then examine the relationships across these quantiles and test for linear trends using statistical testing. Nutrient intake or status can also be examined as a continuous variable. The relationship between intake or status of nutrient X and disease Y can be modeled using regression. Both of these approaches typically assume a linear relationship, which may or may not be a valid assumption. An example to highlight this is found in work from Ulrich (2007) relat- ing folate status to breast cancer risk. Although some studies are finding protective effects with higher folate status, other studies are finding sugges- tions of adverse effects or at least no benefit. Ulrich (2007) has suggested this is because the relationship between folate and breast cancer risk is nonlinear (Figure 3-2). The linearity of a relationship depends on the part of the dose–response curve in which it lies (see dotted and dashed lines in Figure 3-2). This implies that one must be aware of the likelihood that many dose–response associations involving nutrients and chronic disease may be nonlinear.

CRITERIA FOR SCIENTIFIC DECISION MAKING 71 Hypothetical nonlinear Breast Cancer Risk relationship between folate status and breast cancer risk Relationship for area of dose–response studied Relationship for area of dose–response studied Folate Status FIGURE 3-2  Hypothetical nonlinear relationship between folate status and breast 3-2.eps cancer risk as compared with relationships for different areas of the dose–response curve. SOURCE: Modified from Ulrich (2007). One alternative to linear models is restricted cubic spline models, also known as piecemeal polynomial curves. Spline models allow for the ex- amination of nonlinear effects of continuous variables (e.g., nutrient in- take or concentration) in relation to disease risk. Some advantages of this approach are that no functional form needs to be specified; it is available in standard statistical packages (SAS, BMDP); and it can reveal nonlinear dose–response relationships. An example of the use of restricted cubic spline models is from Wright et al. (2006), who examined the relationship between serum vitamin E and all-cause mortality (Figure 3-3). When the best model is fit to the data, as serum vitamin E concentrations rise, there is apparently a reduction in the risk of dying in this cohort up to a particular point; after that, it appears there is no additional benefit and, if anything, the possibility that the risk may start to increase. We might choose a serum vitamin E concentration associated with the minimum risk based on this curve, then determine the nutrient intakes required for half the population to achieve this plasma vitamin E concentration. Combining data from multiple studies and using the data to estimate dose–response relationships are also possible. One standard approach is to

72 THE DEVELOPMENT OF DRIs 1994–2004 Percentile 1 10 20 30 40 50 60 70 80 90 99 1.5 1.4 1.3 1.2 Relative risk 1.1 1.0 0.9 0.8 0.7 0.6 0.5 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Cholesterol-adjusted serum α-tocopherol (mg/L) 3-3.eps FIGURE 3-3  Cubic spline regression for total mortality according to cholesterol- adjusted serum α-tocopherol concentrations. —, Predicted relative risks; ---, 95% confidence interval. The reference value (9.1 mg/L; relative risk = 1.00) corresponds to the median value of the first quintile of serum α-tocopherol concentrations. To convert cholesterol-adjusted serum α-tocopherol concentrations from mg/L to µmol/L, multiply by 2.322. SOURCE: Wright et al. (2006). Am. J. Clin. Nutr. (2006; 84; 1200–1207), Ameri- can Society for Nutrition. take data across multiple, randomized nutrient supplementation trials and perform a systematic review and meta-analysis. Meta-analysis was origi- nally developed for clinical trials to see if an effect is present or not (e.g., do statins reduce CVD risk?). Meta-analysis can also be used to characterize dose–response using data from different trials with different nutrient doses and different achieved plasma concentrations. In Figure 3-1 (page 67), from a meta-analysis looking at vitamin D supplementation and its effects on hip fracture and nonvertebral fracture, the authors performed a meta-regression to fit a linear regression to the data on the relative risk for a chronic disease endpoint as a function of achieved plasma 25-hydroxyvitamin D concentrations. Although they fit

CRITERIA FOR SCIENTIFIC DECISION MAKING 73 a linear model to these data, a nonlinear model may fit better, especially for hip fracture. We could have used these data to fit a nonlinear function, identify a plasma concentration at which lowest risk is observed, and then relate that level back to intake data. Meta-analysis is also used for observational epidemiological studies of nutrients and chronic disease risk, but it was not designed for observational studies, and therefore its application is much more problematic. The dose that corresponds to high intake in one population may be very different from that in another population and in different parts of the dose–response curve (see Figure 3-2). The dose–response meta-analysis across categories can be done, with the caveat already mentioned. An example from the literature is a meta-analysis looking at observational studies on selenium intake and prostate cancer risk (Etminan et al., 2005). The investigators plotted studies of selenium intake (with lowest intake as the reference group) and risk of prostate cancer (Figure 3-4). Finding any dose–response data in this type of study is difficult because of the nonquantitative nature of the data. Another approach to estimate dose–response is to combine data from multiple studies into a pooled analysis, where the original data from mul- tiple studies are obtained and reanalyzed together. The assumption is that intake data across the studies are similarly (quantitatively) assessed, which is an assumption whose validity can be challenged. Validity is nutrient spe- cific, depending on the ability to estimate intake of that nutrient accurately across populations. An example of a pooled analysis is from Hunter et al. (1996), who examined the relationship between percentage of energy from fat in the diet and breast cancer risk (Figure 3-5) and concluded there was no association. However, it is assumed, perhaps not correctly, that when data are pooled from multiple cohort studies that use different dietary instruments, fat intake (along with energy intake) can be measured precisely and similarly across the studies. The statistical approach can also apply to ULs. Instead of risk of inadequacy, risk of excess is modeled (e.g., the risk of hip fracture with high vitamin A intake). Similar approaches as described previously can be applied to ULs (e.g., spline models, meta-analysis, meta-regression), and the nutrient concentrations or intake levels at which risk of adverse effect begins to increase can be evaluated. In terms of using chronic disease endpoints for dose–response estima- tion, although chronic disease data are widely available from U.S. and Canadian populations, causality and confounding (e.g., correlated nutrients from the same foods) are difficult to address. The use of plasma biomarkers is desirable to examine dose–response, but it does not solve the confound- ing problem.

74 THE DEVELOPMENT OF DRIs 1994–2004 Cohort Studies Duffield-Lillicio (13) Lipsky (14) Van de Brandt (5) Li (15) Hartman (16) Helzbouer (17) Goodman (18) Knekt (19) Nomora (20) Coates (21) Yoshizawa (22) Pooled RR Case-Control Studies Brooks (23) Allen (24) Vogt (25) Ghadirian (26) West (6) Pooled RR 0.5 1 2 FIGURE 3-4  A meta-analysis of observational studies of selenium intake and pros- tate cancer risk. 3-4.eps NOTE: RR = Relative Risks. SOURCE: Etminan et al. (2005). Reprinted from Cancer Causes and Control 16:1125–1131, figure 1, with kind permission from Springer Science and Business Media. Copyright © Springer 2005. The Biological Approach The biological approach uses the mode of action framework. The idea is that in order to approximate a dose–response, we need to understand the mode of action of nutrients. This has a straightforward application to ULs, but it can apply equally to nutrient deficiency. Key molecular and biological systems and pathways that are modulated by nutrients need to be identified. The background paper on the biological approach (“Approximating Dose–Response in the Face of Limited Data,” posted on the IOM website [www.iom.edu/driworkshop2007]) describes the tools and technologies that are in use in other fields that may be helpful in establishing dose–response

CRITERIA FOR SCIENTIFIC DECISION MAKING 75 2.0 1.5 Relative Risk 1.0 0.5 0.0 <20 20–<25 25–<30 30–<35 35–<40 40–<45 ≥45 % of Energy from Fat FIGURE 3-5  Fat and breast cancer: pooled analysis. SOURCE: Hunter et al. (1996). Copyright © 1996. Massachusetts Medical Society. All rights reserved. 3-5.eps in the nutrition literature. Mapping pathways could be helpful. However, we often know the pathways that cause deficiency, so mapping pathways may not necessarily get us closer to dose–response. In vitro tests in human cell lines are being widely used in the pharmaceutical industry, for example. However, human dose data supersede in vitro tests in human cell lines. High- throughput methods are another technique where, for nutrients, different doses could be used to see if any inference or input about dose–response can be obtained, but they will not necessarily solve any problems. Micro- arrays, computational biology, and physiologically based pharmacokinetic and pharmacodynamic models are all reasonable approaches where animal models of nutrient toxicity and deficiency are available and are particularly helpful for different life stages. Not mentioned in the background paper are metabolomics and translational biology, fields where there is much research progress that may be helpful in terms of informing us about dose–response in the future. Conclusion There is a real trade-off between the statistical and biological ap- proaches. The statistical approach targets the right population and the right nutrients, but there is limited causal inference when dealing with chronic disease endpoints. The biological approach is mechanistically driven, but there is a tenuous link to the human dose data.

76 THE DEVELOPMENT OF DRIs 1994–2004 In conclusion, newer options exist for examining dose–response in the setting of DRIs, but none are yet mature or ideal. To move forward, we must have a multidisciplinary, integrated approach involving biostatisti- cians, toxicologists, and nutrition scientists. There is no obvious advantage of one approach over another at this time. An approach of data conver- gence, where we look at all of the evidence to determine if we can charac- terize dose–response, may be most useful. General Discussion One audience member raised the issue of meta-analysis studies and sensitivity testing, and asked whether Dr. Mayne saw any role for sensitiv- ity testing and, if so, what sort of new data would be needed to determine the quality of the current body of data or to change overall findings. Dr. Mayne responded that sensitivity analysis usually examines whether a par- ticular study has undue influence on the results and whether its inclusion or exclusion changes conclusions about the existence of an effect. Regard- ing her focus on dose–response data, Dr. Mayne indicated that it would not necessarily be informative to do a meta-analysis based on excluding or including specific trials to examine dose–response relationships among the studies. The audience member further speculated on ways to determine if small studies or a large study is needed. In response to a request for clarification on the New Zealand and Chi- nese selenium studies, Dr. Mayne said she was not familiar with all of the details of the studies. While both studies were published, they were not in a readily accessible journal at the time the committees initiated their work. She noted that the Chinese data were published in a book, and the New Zealand data were analyzed for the DRI report and subsequently published in 1999. She commented that nutrient deficiency data tend to be limited, often from other countries, and frequently outdated. However, she noted in contrast that an intervention study on selenium involving 35,000 people is currently underway; it is a 13-year clinical trial that completed recruitment about 2 years ago. Another question was related to the earlier comments about the chal- lenges associated with chronic disease endpoints for DRIs because the avail- able studies often lack dose–response data. The question was raised, given Dr. Mayne’s presentation on tools to approximate dose-response relation- ships, as to whether it was premature to suggest eliminating chronic disease endpoints because they lack dose–response data. Dr. Mayne responded that some tools are available for this purpose, but the confounding issue remains for certain chronic diseases. For example, nutrients are only one of numer- ous factors that determine cancer risk. However, for some chronic diseases there may be a nutritional role more proximal to the disease endpoint, such as lutein and zeaxanthin for macular degeneration. Therefore it is unwise

CRITERIA FOR SCIENTIFIC DECISION MAKING 77 to “throw out” all chronic diseases because some are multifactorial and confounded. In short, chronic diseases should not be lumped together in considering their utility in the DRI process; as endpoints they likely need individual consideration. WHAT ARE THE CHALLENGES IN ADDRESSING EXTRAPOLATION/INTERPOLATION FOR UNSTUDIED GROUPS? Presenter: Stephanie A. Atkinson The paucity of data for certain subpopulations resulted in extensive use of extrapolation models during the DRI development process. In fact, about 60 percent of the DRIs were derived by extrapolation for 1- to 18-year-olds. The paucity of specific data available based on research in infants and chil- dren is concerning. For this reason, careful consideration of extrapolation methods is needed to ensure that we are doing the best we can until that point when data are available and DRI reference values can be set without the need for extrapolation. Our experience suggests that various approaches to extrapolation have been used, which has led to inconsistencies in reference values among age groups. For example, for the 6- to 12-month age group, extrapolation up from the AI for 0- to 6-month-olds was done for niacin, choline, biotin, and vitamins B12, A, and K. At the same time, extrapolation down from the adult EAR or AI was done for vitamins B1, B2, and B6; folate; and pantothenic acid. Furthermore, extrapolating down from adults, with inappropriate mod- els in particular, leads to DRIs that do not make much sense. For vitamin A, the AI is 500 µg retinol activity equivalent (RAE)/day for 6- to 12-month- olds compared with an RDA of 300 µg RAE/day for 1- to 3-year-olds, one being extrapolated up from the 0- to 6-month group and the other being extrapolated down from adults. For vitamin C, the AI is 50 mg/day for the 6- to 12-month group based on the composition of human milk and intake from food, but the RDA is only 15 mg/day for the next age group, being extrapolated down from adults. An effect in the opposite direction is observed for the derivation of DRIs for molybdenum: an AI of 3 µg/day for 6- to 12-month-olds based on human milk and food and an RDA of 17 µg/day for 1- to 3-year-olds extrapolated down from adults. In the case of fiber, there were no data for young children, so the AI was extrapolated down from an adult AI of 14 g/1,000 kcal based on the reduction in CVD risk. The AIs for children 1–13 years of age range from 19 to 31 g/day, whereas the intakes obtained from diet surveys range from 5 to 18 g/day (Suitor and Gleason, 2002; Devaney et al., 2004). Clearly young children and adolescents are not consuming anywhere near the

78 THE DEVELOPMENT OF DRIs 1994–2004 amount of fiber predicted by extrapolating down from adults. Thus, the AI may be impossible to achieve; more importantly, it may not be physiologi- cally appropriate. Overview of Available Approaches In North America, extrapolation is used in DRIs to adjust for physi- ological differences between groups of varying body size or age to establish a reference value for an unstudied age/gender group. In Europe, scaling has been used since the early 1800s with regard to expressions of body weight or compartments scaled to height (e.g., the body mass index, or BMI). “Scaling” may be the more appropriate term to use for the purposes of DRI development. Regardless of terminology, the types of extrapolation/scaling models used are fairly similar. There are linear models, where body size (mass) can be used with a reference body weight for age and gender (e.g., the AI for fiber for children) or as a function of energy, where the median reference energy intake for age and gender is used (e.g., the AI for water, sodium, and potassium for children). The problem with the linear model is that there is no accounting for age variations in intermediary metabolic rates, energy intake, or basal metabolic rate (BMR). The exponential model, on the other hand, tries to adjust for metabolic differences related to body weight (BW) (e.g., the UL extrapola- tion from adults to children uses BW0.75). The issue is that this assumes that maintenance needs of nutrients as a function of metabolic weight are similar for adults and children and similar across genders. It also assumes that absorption, digestion, and excretion are similar across age groups. Apparently there is a lack of consensus on which adjustment factor best reflects BMR (e.g., adjustment factors in the range 0.6–0.8 have been used). The values produced by the exponential model are always higher than those produced by the linear model. The other model for scaling is relative to body surface area, and this adjusts for metabolic differences between ages related to body surface area based on its relation to BMR. This will always result in higher nutrient reference values than those based on body weight. A study by Przyrembel (2006) shows a nearly twofold difference between relative nutrient intakes using body weight and those derived using body surface area for children up to 1 year of age (Table 3-1). Some reports, such as that of the Scientific Committee for Food of the European Union (SCF, 1993), use interpolation, which is different from scaling or extrapolation, in that the value is interpolated for an age group between known values of age groups older and younger. Which of these models is most accurate is open to interpretation.

CRITERIA FOR SCIENTIFIC DECISION MAKING 79 In all of these models, few other factors can be applied differently across reference intake standards of various agencies. One of those is the values used for growth in extrapolating from adults to children. In the DRI reports, we used approximate proportional increase in protein requirements for growth as established by the World Health Organization (WHO, 1985); the growth factors were 0.30 for 7 months to 3 years and 0.15 for older age groups. These were applied for every nutrient by assuming that the growth factor was the same as that established for protein. The other variable is reference body weights, which should reflect those of the country or countries to which they are being applied. The reference body weights changed during the DRI process. At first, National Health and Nutrition Examination Survey (NHANES) III values were used (IOM, 1997); later, when the new Centers for Disease Control and Prevention (CDC)/National Center for Health Statistics (Kuczmarski et al., 2000) val- ues became available, median heights and weights computed from median BMI were used. Infants were not based on the breastfed infant population because the data were not available (the CDC data mostly reflect for- mula-fed infants). For the next round of DRIs, we could use the recently published WHO reference growth standards for breastfed infants (Garza, 2006; de Onis et al., 2006, 2007) for 0- to 5-year-olds or perhaps Canadian reference growth data, which may be available soon. Challenges Extrapolation is a proxy; thus, it yields a risk of error, especially when values are extrapolated from adults to children. The evaluation of physiol- ogy requires scientific judgment. We need knowledge of substrate absorp- tion, metabolism, and deposition in tissues during growth phases and renal and other excretion that may affect the EAR or UL, as these may not be related in a simple fashion to body size, even in the exponential model. For some nutrients, especially for the UL, extrapolation on the basis of body weight or body surface area yields a UL for children that is incompatible with known nutrient intakes. Perhaps ULs should not be set for children until we have direct experimental evidence. Children are not just little adults. They need their own evidence-based DRIs. If DRIs are inappropriately set, we might adversely affect the health of children. We could identify the wrong nutrient intake problems (either inadequacy or excess), which could lead to inappropriate recommendations for child health feeding programs (e.g., Special Supplementation Nutri- tion Program for Women, Infants and Children) and have a public health impact.

80 THE DEVELOPMENT OF DRIs 1994–2004 TABLE 3-1  Relative Nutrient Intake (RNI) Reference Values by Extrapolation from Adults: Body Weight Versus Surface Area RNI Based on Child/ RNI Based on Child/ Age Adult Body Weight Adult Body Surface Area Newborn 0.05 0.11 0.5 years 0.10 0.19 1 year 0.14 0.23 10 years 0.46 0.59 SOURCE: Przyrembel (2006). Conclusion In an ideal world, the use of scaling and extrapolation models in setting DRIs should be unnecessary. However, the reality is that they must be used. In such cases, we need to ensure that we are using biologically plausible models and recognize the role of well-reasoned and transparent scientific judgment. Today there are opportunities that did not exist 15 years ago to conduct research in children, including the use of stable isotopes to measure energy requirements, amino acid oxidation, and amino acid requirements, as well as for trace element turnover. The pursuit of these appropriately designed studies is critical. General Discussion One person pointed out that obtaining data for currently unstudied groups will take a long time, even with new methods. Given that, he asked how we can use data from animal models, which are more readily available and can be obtained in a shorter time. Dr. Atkinson responded that animal models can be helpful, but need to be closely aligned to the human infant or young child, such as monkeys or piglets. These are expensive research models. An audience member then questioned whether, given the difficulties and lack of data, ULs for children should not be developed. Dr. Atkinson suggested it would be important to pursue appropriate animal models to study adverse effects as a preliminary and hypothesis-generating step for this purpose. The audience member suggested that at least two or three species should be used to reduce uncertainty. A participant remarked that stable isotopes are an excellent approach for children. However, noting they are also expensive, he asked whether marker nutrients might be translated to other nutrients in the same category to lower costs. Dr. Atkinson agreed it was possible. She also noted that a practical barrier in doing research in normal children is the ability to draw blood, and that obtaining urine is less challenging.

CRITERIA FOR SCIENTIFIC DECISION MAKING 81 The comment was made that Dr. Atkinson did not address pregnancy and lactation as extrapolation concerns in her presentation. She responded that extrapolation for pregnancy and lactation requires a different focus from her main topics for this presentation. She noted a reference that would be helpful regarding pregnancy and lactation (Atkinson and Koletzko, 2007). WHAT ARE THE CHALLENGES IN ADDRESSING ADJUSTMENT FOR DATA UNCERTAINTY? Presenter: Hildegard Przyrembel The root of the problem we face is a lack of data, which results in uncertainty. Uncertainty can be reduced only by the acquisition of more data. Alternatively, analysis of the impact of sources of uncertainty can be used to understand and thereby help to address uncertainty. The analysis can be qualitative (descriptive) or, preferably, quantitative (mathematical modeling). This presentation will focus on adjusting for data uncertainty from the perspective of establishing ULs, but many of the principles may also apply to establishing reference values. For the purposes of establishing ULs, the mode of action (i.e., endpoint of interest) and the related dose–response relationship are critical pieces of information, as shown in Figure 3-6. A no-observed-adverse-effect level (NOAEL) or lowest-observed-adverse-effect level (LOAEL) is identified (or, alternatively, a benchmark dose is calculated), then modified by the use of uncertainty or adjustment factors in order to derive the UL. Uncer- tainty factors refer to default values with no or little factual basis, whereas adjustment factors are values supported by actual toxicodynamic and/or toxicokinetic data. On the other hand, establishing requirements and avoiding deficiencies also require an understanding of the mode or mechanism of action as well as information on the dose–response relationship (see Figure 3-7). However, the identification of the mirror image of the NOAEL or LOAEL (i.e., a critical dose) would be problematic. Assuming it could be determined, then it would need to be multiplied by an adjustment or uncertainty factor in order to obtain the “lowest threshold value of intake.” A question would remain, however, as to how to convert this value into an average estimated requirement, the most obvious but perhaps unsuitable suggestion being that   Data uncertainty must be differentiated from data variability, which is due to the het- erogeneity of a quantity over time, space, or members of a population. It can be reduced by selection of the sample, not by the provision of more data. However, probabilistic assessment methodology is now used widely for assessing data variability.

82 THE DEVELOPMENT OF DRIs 1994–2004 A: Deficiency B: Excess ∗ mode (mechanism) ∗ mode (mechanism) of action of action ∗ dose–response ∗ dose–response relationship relationship Critical dose NOAEL or LOAEL multiplied by ? divided by UF Lowest threshold value of Tolerable upper intake intake (= EAR – 2SD) Level (UL) Targeted result: RDA < UL 3-6.eps FIGURE 3-6  Risk assessment of essential nutrient and adverse health effect. NOTE: EAR = Estimated Average Requirement; SD = standard deviation; NOAEL = no-observed-adverse-effect level; LOAEL = lowest-observed-adverse-effect level; UF = uncertainty factor; RDA = Recommended Dietary Allowance. External Dose External Dose Toxic Response Absorbed Dose Absorbed Dose Clearance Intracellular Intracellular Clearance Pathological Changes Concentration in Concentration in Pathological Changes General Circulation General Circulation Distribution to Distribution to Nontarget Tissues Nontarget Tissues Concentration in Concentration in Interaction with Interaction with Target Tissues Target Tissues Intracellular Intracellular Targets Targets Local Bio- Local Bio- activation activation Physiological Physiological Responses FIGURE 3-7  The multiple steps between intake of a nutrient and either physiologi- cal or toxic responses, depending on3-7.eps dose. SOURCE: Modified from IPCS (2005).

CRITERIA FOR SCIENTIFIC DECISION MAKING 83 this lowest threshold value reflects a value two standard deviations below the median requirement. Considering the Points of Impact for Data Uncertainty In general, from the metabolic and physiological perspectives, the points at which uncertainty can have an impact and therefore be assessed are divided into multiple steps between intake of a nutrient and either physiological or toxic responses, depending on the dose (Figure 3-7). Most of the literature and research on uncertainty analysis have been done for chemicals, rather than nutrients. This rendition is derived from those fields of study. The pathway moves from the “effect” of the dose and separates at the interaction of the nutrient or metabolite of that nutrient with intra- cellular targets, to go toward either the physiological response or a patho- logical or toxic response. All the different steps can be characterized either in animals or in different age or gender groups of humans, which helps to modify or quantify the necessary adjustment or uncertainty factors. How- ever, few data are available on these different steps, making it difficult to develop reasonable adjustment factors. Another illustrative example from the field of chemical study that is inexplicably missing from work in the nutrition area is a theoretical dose–response curve for various effects occurring in the population. This type of mapping greatly assists efforts to study and address uncertainty. For instance, Figure 3-8 plots the percentage of the population with an effect against the range of acceptable daily oral intake of a nutrient; it shows that different endpoints can be identified, and it suggests that these dose–response curves should be parallel (although there is no reason why they should be parallel). Further, the same stepwise procedure for increase of effects could apply for both nutrient intakes higher than the requirement (excessive) and in- takes lower than the requirement (deficient). Figure 3-9 shows a combined curve of dose–response relationship for the risks due to deficiency (absence of benefit) and toxicity. What is a benefit? There is no assurance that a ben- efit will result from higher intakes or that the benefit always needs higher intakes and requirements. What is needed for this kind of parallel assess- ment of excess and benefit is the intake that gives a 50 percent incidence response, the ED50, and the coefficient of variation (CV) of response. In the graph, different CVs have been assumed (as they often are). The CV influences the point where the two curves for toxic response and benefit response intersect.

84 THE DEVELOPMENT OF DRIs 1994–2004 Deficiency Toxicity 1 2 3 4 5 6 7 8 Percentage of population with effect 100 Bioclinical marker without Bioclinical marker without functional significance functional significance Subclinical biomarker functional impairment Subclinical biomarker functional impairment Clinical effect Clinical effect of effect with of effect with Death Death 0 A R O I Range of acceptable daily oral intake FIGURE 3-8  Theoretical dose–response curves for various effects occurring in a population. 3-8.eps NOTE: AROI = Acceptable Range of Oral Intake. SOURCE: IPCS (2002). Sources of Uncertainty The main sources of uncertainty include the model used, the quality of the available data, and the scaling algorithm. The model used is determined by both structure and parameters. Sensitivity analysis can be performed to identify those parameters with significant impact on model output. Regard- ing available (or input) data, they are often lacking and need to be evalu- ated in terms of their quality, variability, and measurement and database errors. Quantification of missing data is impossible. The only solution is to input fictive or virtual data into the models and assess the difference in outcome. For scaling algorithms, it is always difficult to be certain that the algorithm chosen is appropriate. One problem of scaling is that it propa- gates errors made earlier in the process. Turning more specifically to the DRI development process, uncertain- ties in available data include methodology of balance studies (e.g., calcium), lack of data (e.g., pantothenic acid), physiological significance (e.g., vita- min K), and lack of identification of an adverse effect (e.g., vitamin B1). Major sources and types of uncertainties in dietary exposure assessment include food consumption, body weight, and content in food. Uncertainty in relation to the food composition database is large and relates to fac-

CRITERIA FOR SCIENTIFIC DECISION MAKING 85 FIGURE 3-9  Dose–response relationships for the risks due to the absence of benefit or the presence of toxicity. The data were plotted assuming a log-normal distribu- tion. The absence of benefit (equivalent to a deficiency condition) has been plotted assuming coefficients of variation of3-9.eps line) and 15% (narrow line), and 10% (thick low-res bitmap image the toxicity line has been plotted assuming a coefficient of variation of 45%. The intersection of the two lines is the optimum intake, provided that the nature of the deficiency and the toxicity are of equivalent adversities. ED50, the dose that gives a 50% incidence; CV, coefficient of variation. SOURCE: Renwick (2006). J. Nutr. (2006; 136; 4935–5015), American Society for Nutrition. tors such as differences in bioavailability of nutrients from different foods and variability in composition with, for example, storage, processing, and preparation. The uncertainty factors that have been used traditionally in chemical toxicology are default values—for example, 10 for extrapolation from animals to humans (interspecies) and 10 for coverage of human variability (interindividual). There are other uncertainty factors for use of subchronic rather than chronic studies, use of a LOAEL instead of a NOAEL, and defi- ciencies in the database. The applicability of these to nutrient considerations

86 THE DEVELOPMENT OF DRIs 1994–2004 is questionable (especially given the dual risk of essential nutrients—risk from too little and risk from too much), and they need considerable modi- fication for use with nutrients, assuming they are useful at all. In the DRI reports, where the endpoint was a human NOAEL, an adjustment factor of 1.0 was chosen in half the cases. The justification for using an uncertainty factor above 1.0 was mostly variability in the popula- tion or insufficient data. Where a human LOAEL was used, the uncertainty factors were higher, except for magnesium, fluoride, and sodium, and the justification for the larger uncertainty factor was often the use of the LOAEL. Where animal LOAELs were used, the variation in the selection of adjustment factors was especially great. Conclusion Uncertainty analysis can help those responsible for developing DRIs. It is intended to systematically examine the adequacy of the selected model (e.g., to see if predictions agree with observations), the uncertainties in model parameters and input data (using mathematical methodologies), and the presentation of the results (e.g., probability distribution). Com- munication of the uncertainties and how they have been compensated for are very important. Of course, uncertainty analysis does not preclude the need for appropriate exposure data, relevant endpoints, and trustworthy dose–response data. In conclusion, uncertainty in nutrient risk assessment and in the defini- tion of requirements of nutrients is unavoidable. It should be characterized with respect to its nature and magnitude, and different types of uncertainty should be ranked according to their impact on the results of the procedure. For nutrients, the default uncertainty or adjustment factors convention- ally used in the risk assessment of chemicals have to be modified, ideally by either chemical-specific kinetic or process-specific dynamic data. We should try to obtain these data, but we will have to do this for each nutrient individually because of the different and multiple physiological functions of different nutrients in the human body. Uncertainty due to variability in both kinetics and dynamics can be dealt with by mathematical procedures. Uncertainty due to gaps in data, however, can be effectively relieved only by the acquisition of more data. In the meantime, assumptions about data used in the assessment and the impact of their intentional variation in the calculation need to be identi- fied and communicated both qualitatively and quantitatively. However, how this communication is understood by the users is uncertain.

CRITERIA FOR SCIENTIFIC DECISION MAKING 87 General Discussion A participant noted that one of the slides showed that the CV for ben- efit was assumed to be either 10 or 15 percent, whereas the CV for toxicity was apparently assumed to be 45 percent. He expressed the view that none of the ULs are set on the basis of a measure of central tendency; instead, the ULs are based more on a threshold concept. He queried whether Dr. Przyrembel concurred with the validity of the graph. Dr. Przyrembel re- sponded that the author of the slide justified his selection in his publication, giving data to support his assumption that variability in sensitivity to toxic effects differs from variability in requirement. The participant then outlined a different approach (when there are sufficient data) based on rank ordering the clinical trials and omitting the use of numerical adjustments. An audience member asked how to deal with the uncertainty associ- ated with studies that administer similar doses, but demonstrate different responses. Dr. Przyrembel responded that the only solution is to carefully examine the study design for an explanation of the differences. Another participant raised a question about using clinical studies that were designed for efficacy or benefit trials to ascertain adverse event infor- mation. Evaluating the equivalency of studies is very difficult if they have been conducted for a different purpose. ESTIMATING DIETARY INTAKE: WHAT ARE THE IMPLICATIONS FOR DRI DEVELOPMENT? Presenter: Amy Subar This presentation addresses the use of dietary intake estimates in DRI development, notably as it relates to the step focused on dietary exposure (or intake) assessment. These estimates of current intake in the United States and Canada allow study committees to place DRI values once they are developed within the context of the population’s current estimated con- sumption and, in turn, to characterize the risk of inadequate or excessive intake. An understanding of the strengths and weaknesses of the various dietary assessment methods for estimating current population intakes is important in ensuring the proper use and interpretation of these dietary estimations. Other types of data on intake relevant to DRI development include studying dose–response relationships in clinical feeding studies, evaluating DRIs in population-based epidemiological or clinical studies, and develop- ing AIs from national dietary surveys, such as NHANES. These types of data and the use of estimated intakes to examine the relationship between intake and health outcome will not be specifically discussed.

88 THE DEVELOPMENT OF DRIs 1994–2004 Methods for Estimating Intake: Self-Report Instruments It should be noted that the goal for all applications of dietary intake estimation is an estimate of usual intake, which is the theoretical long-run average daily intake of a dietary component. Three main types of self-report instruments are used to collect such data: 24-hour recalls, food diaries or records, and food frequency questionnaires (FFQs). Twenty-Four-Hour Recalls Twenty-four-hour recalls can vary in many ways. Training of the in- terviewers and standardization of probing questions (i.e., questions that follow after someone reports eating a particular food, such as what kinds of fats were added to foods) can vary from study to study. Most 24-hour recalls are collected by some sort of standardized computerized approach, but some studies use pencil-and-paper administration with later coding of the data. Some recalls are done in person, others by telephone. Different kinds of portion size models or measurement aids are used to estimate por- tion size. The 24-hour recall has various strengths. The intake data can be quan- tified in detail. In theory, it should not affect human eating behavior be- cause the respondents are asked to report what they ate yesterday, intake that would have occurred before they knew they would have to report such intake. There is lower sample selection bias than for other methods because the recall does not require literacy and the respondent burden is low. It is generally agreed that this is the most reliable method for dietary assessment. Furthermore, usual intake distributions can be estimated from as few as two dietary recalls. One weakness is that recalls rely on memory. Also, 24-hour recalls are costly to develop and administer because highly trained interviewers are needed. In addition, because recipes and preparation methods vary for many foods, default recipes and hence nutrient values are used, and these may not accurately capture the level of nutrients consumed. Underreporting of foods and amounts eaten is also common, especially among those who are overweight or obese. Finally, at least 2 days and statistical modeling are required to obtain usual intake estimates. Food Diaries or Records Food diaries or records are, in general, less standardized than dietary recalls. Respondents do not have to be trained, but the diaries may or may not obtain comprehensive data, and the coding of those data is highly vari- able from study to study. The use of technology to collect real-time dietary

CRITERIA FOR SCIENTIFIC DECISION MAKING 89 data has been a research topic of great interest, with technology such as personal digital assistants, cell phones to take pictures, and voice recogni- tion being explored. If done correctly, a food diary or record can provide quantified and detailed intake information. It can be relatively accurate, and it is done in real time so in theory should not rely on memory. The biggest weakness of a food record is that it is reactive and hence biased. Because respon- dents know they have to record, they may change what they eat because it is difficult to record, or they may undereat. The food record requires literacy, and it has a high respondent and investigator burden. There is a high sample selection bias because only certain people are willing to keep records. The longer people keep records, the worse the data quality is. Al- though it should be real time, people often record the data at the end of the day. Underreporting is typical, and worse with those who are overweight or obese. Food Frequency Questionnaires In the often self-administered FFQs, people are asked a series of questions—usually hundreds—about how often they usually consumed a particular food in a given time period; what preparation methods were used; and what the typical portion size was. These components vary among FFQs, as do procedures to determine the food list and the nutrient composi- tion assigned to each food. One strength of the FFQ is that the respondent burden is relatively low because the questionnaire is filled out only once. The focus is generally usual intake and the total diet. An FFQ should not be biased by changes in eating behavior because intake in the past is queried. Another benefit of the FFQ is the low cost associated with administering the instrument and processing the data. One weakness of the FFQ is that it lacks detail because it contains a finite list of foods and details are not generally collected. It is cognitively complex for respondents to report what they ate over the past year, for example. It requires literacy. Different FFQs can produce different results in the same population, whereas the same FFQ can produce different results in different populations. There is severe measurement error when looking at absolute intakes. To reduce this bias, epidemiologists rank individuals and adjust the models for energy intake. In general, outcome findings are attenuated by the amount of error in the FFQs. Methods for Estimating Intake: Biomarkers Certain so-called “biomarkers of intake” may be used to assess dietary intake. A recovery biomarker is one in which there is a 1:1 relationship

90 THE DEVELOPMENT OF DRIs 1994–2004 between what is consumed and the biomarker value. Such biomarkers provide very accurate data on what individuals are consuming, but few of these can be used: doubly labeled water, urinary nitrogen, and possibly urinary potassium. Concentration biomarkers reflect a direct biological response to what someone consumes. It is more of a correlated response, and it is affected by other characteristics of the individuals (e.g., whether they smoke or pos- sibly their body weight). Therefore, it cannot be used to assess the amount consumed, and it may reflect short- or long-term intakes. In general, it is difficult to use such biomarkers to evaluate direct dietary intake for pur- poses of DRIs. There are also homeostatically controlled biomarkers, which have no direct relationship to intake. Challenges and Sources of Error/Bias First, underreporting occurs in all of the self-report dietary assessment methods described above. The percentage of energy underreported based on a review of doubly labeled water studies was up to 58 percent for food records, 38 percent for FFQs, and 26 percent for 24-hour recalls (Trabulsi and Schoeller, 2001). Underreporting can vary by gender, age, and BMI. In general, underreporting tends to increase as body weight increases. For example, results from NCI’s Observing Protein and Energy Nutrition (OPEN) study, conducted with about 500 men and women using doubly labeled water and urinary nitrogen, show that energy underreporting oc- curs for both 24-hour recall and FFQ, and is greater for FFQ (Figure 3-10). The results also show that underreporting varies by BMI for the FFQ and 24-hour recall (not shown). Second, data on dietary supplements may not be collected in many studies. We have to assume that measurement error is present in assess- ing self-reported dietary supplement intake. However, not accounting for supplement intake leads to substantial underestimation of total nutrient intake. When supplement intake is included, this results in highly skewed intake distributions, which present challenges for describing usual intake distributions. Another source of error in all self-report dietary data relates to the nutrient database. Analytical methods for nutrient composition change and improve, and, just as importantly, the composition of finished food products is constantly changing. Therefore, the database that we use needs to be updated and to match the time period of the study. Obviously it is impossible to observe long-term or usual intakes. Rather, the approach is to acquire estimates based on statistical modeling using

CRITERIA FOR SCIENTIFIC DECISION MAKING 91 - 3-10.eps FIGURE 3-10  Results from the Observing Protein and Energy Nutrition (OPEN) Study: Energy intake underestimation bitmap image low-res by 24-hour recall and food frequency ques- tionnaire compared with total energy expenditure. SOURCE: Subar et al. (2003). short-term, self-reported data. Early in the evolution of dietary intake esti- mation, we used a single day of intake and called it usual intake based on recalls from national surveillance studies. Then we realized we needed at least the average of a few single-day measurements to improve estimates. Next, we became more sophisticated and used statistical modeling: first the NRC method, then the Iowa State University (ISU) method, and more recently the NCI method. Given that the assumptions involved are taken into account, these statistical models remove day-to-day variability from the 24-hour recall so that a better estimate of usual intake is obtained. This is illustrated in Figure 3-11. The probability is plotted against the usual intake of energy; 2,200 calories is the cutpoint. If 1 day of intake is used, the distribution would be long and skewed to the right side. When statistical modeling is applied—removing some of the variability—a more normal distribution of intake is obtained, as would be expected in the population as a whole. This statistical treatment of the data is important, and methods continue to be developed to establish usual intake distributions. The NCI method builds on the NRC/ISU methods to estimate usual nutrient intake distributions. It can also handle episodically consumed dietary constituents, such as vitamin A, and it can be applied to foods and dietary supplements. It also provides greater power to conduct subgroup analyses within the same model.

92 THE DEVELOPMENT OF DRIs 1994–2004 Usual Intakes One-day Intakes 3-11.eps FIGURE 3-11  Probability of consuming above or below cutpoint (dashed line): low-res bitmap image One-day versus usual intake distributions. dotted rule, arrows, & some type vector objects Implications Dietary exposure (or intake) assessment for DRI development ideally would be based on usual intake distributions estimated from some multiple days of intake and statistical modeling. Sometimes there is interest in using intake data from observational studies. However, we have to be careful, given the amount of error that can occur in FFQs and other methods used in such studies. The starting point for DRIs is the available clinical and metabolic data concerning requirements, health outcomes, and adverse events; DRIs are not derived (AIs excepted) from estimates of usual intake. Therefore, it is understandable that DRIs, even when developed using the best available scientific data, might be disparate from estimated intakes from dietary sur- veillance data. A clear understanding of the strengths and limitations of the dietary intake estimates allows those responsible for DRI development to put the scientifically derived DRI values in the context of current estimated intakes and, in turn, advise users of DRI values about differences between values and estimated intakes and possible reasons for them; it also identifies avenues for further research.

CRITERIA FOR SCIENTIFIC DECISION MAKING 93 General Discussion An audience member questioned whether progress can be made as long as we rely on people to report dietary intake information. Dr. Subar emphasized that the data are not all poor and that newer advances have shown considerable promise for ensuring good-quality estimates of intake. She pointed out that even though there is some level of underreporting, better ways to adjust the data are likely to be developed. The key point is that existing data need to be used appropriately, with an understanding of their limitations. Dr. Subar commented that biomarkers of intake would be very helpful. An audience member commented that doubly labeled water appears to quantify underreporting. However, she asked about the valida- tion of this technique and expressed concern about whether known dietary intake is actually underreported to the extent currently suggested by doubly labeled water studies. Dr. Subar indicated that the doubly labeled water methodology is well established as a measure of true energy expenditure in individuals, but she did not know if the intake matches the estimation in a steady state. Another participant suggested that statistical modeling depends on the assumptions used. The assumption that a yearly intake reflects usual intake may be appropriate in some cases but not others, specifically in developing countries. Dr. Subar commented that the usual intake distribution is based on usual intake in the population. The participant suggested that in the United States, the intake does not vary much with the seasons, but in other countries seasons have considerable impact. One question was raised about using the usual intake distribution when dealing with ULs. Dr. Subar was unfamiliar with any studies or delibera- tions intended to explore this particular issue. Another question was asked about the trustworthiness of the nutrient values on nutrition labels. Dr. Subar responded that others with expertise in this area would be better suited to answer the question. HIGHLIGHTS OF OTHER IMPORTANT ISSUES: PHYSIOLOGICAL, ENVIRONMENTAL, AND GENOMIC FACTORS Presenter: Cutberto Garza Physiological, environmental, and genomic issues relate to the DRI conceptual framework as well as to the applications of the DRIs. This pre- sentation first outlines some general principles to provide a context, then focuses on examples of challenges that physiological, environmental, and genomic issues present.

94 THE DEVELOPMENT OF DRIs 1994–2004 General Principles The governing principle in any expanded consideration of physio- logical, environmental, and genomic issues is the definition of nutritional health. It is helpful to think about nutritional health in terms of a progres- sive overlapping continuum, moving from the bottom to the top of the trapezoid shown in Figure 3-12. The bottom of this continuum focuses on essential food components that, when lacking, give rise to unambiguous pathology related to a specific deficiency; or, if they are in excess, to an adverse effect. The single-agent, single-outcome paradigm governs this part of the continuum. Moving up along this continuum, there is a greater focus on primary and secondary prevention of nutrition-related chronic diseases. The top of the continuum is increasingly attentive to enhanced performance through improved nutrition. Not surprisingly, uncertainty increases as we progress through this continuum from bottom to top. These uncertainties are due to decreases in basic knowledge (shown to the left of the trapezoid), reflecting the need for more research as we move from basic pathology and specific deficiency to concerns such as enhanced performance. There is also growing complexity of underlying biological mechanisms as we move toward enhanced perfor- mance. All this requires some broadening in the use of our tools. There is LESS MORE MORE Complexity of Underlying Mechanisms Enhanced performance and Environmental Conditions Responsiveness to Behavior Knowledge Primary and secondary prevention of diet-related chronic diseases Avoiding classic deficiency disease MORE LESS LESS FIGURE 3-12  Nutritional health continuum. 3-12.eps

CRITERIA FOR SCIENTIFIC DECISION MAKING 95 also rising sensitivity to a wide range of behaviors and environmental condi- tions as we move from bottom to top. The significance of physiological, ge- nomic, and environmental factors will differ along this continuum in ways likely to be specific to individual nutrients and life stages (Figure 3-12). Two principles will help determine when such expanded considerations are appropriate. The first is that the anticipated benefit of modifying a ref- erence value on the basis of any factor—physiological, environmental, or genomic—must be qualitatively significant to either individual or popula- tion health and well-being, somewhat analogous to, but the mirror image of, hazard characterization. Second, the equivalent of an individual- or population-attributable benefit must be quantitatively significant. These seemingly straightforward statements beg the question of what triggers quantitative and qualitative significance. Criteria for determining qualitative significance are not independent from criteria for quantitative, and neither are likely to be determined purely on an objective basis. Assess- ments of both will be influenced by culturally or socially bound values and the ability to use the information. Physiological Factors Physiological factors include gender, age, reproductive status (including lactation), and body size. Considerations of body size are generally limited to expressions of nutrient needs per kilogram of body weight. Body size also incorporates elements of body composition to the extent that these two variables are related in a given population. Four challenges exist with respect to physiological factors, recognizing that historically nutrient-based dietary recommendations have historically excluded nonhealthy populations: 1. The prevalence of obesity and overweight 2. The aging of the North American population 3. The increasing understanding of long-term risks associated with intrauterine growth retardation (slow-for-gestational-age infants) 4. High rates of prematurity, the health consequences of this condition, and increasing technological capabilities that enable survival at pro- gressively lower gestational ages, which will bring special pressures to the DRI process The IOM undoubtedly will be faced with including one or more of these conditions in the future DRI process; there may be a need to develop an ancillary effort to consider these groups beyond the brief paragraphs that have been included in the sections of the DRI reports labeled “special considerations.” In addition, metabolic and other common morbidities that

96 THE DEVELOPMENT OF DRIs 1994–2004 accompany overweight, obesity, aging, intrauterine growth retardation, and/or prematurity likely will influence recommended intakes, at least for some subgroups. The most salient example is the growing prevalence of Type II diabetes, which will be more difficult to ignore. As challenging as these projections may seem, they pale when compared with the implications of considering environmental and genomic issues. Environmental Factors The framework that we have been using generally ignores environmen- tal influences, with the possible exception of energy requirements. How- ever, on an international level it is not uncommon, for example, to at least consider higher rates of endemic infectious diseases in the determination of nutrient requirements. Such conditions are often environmentally driven. Perhaps it is time that we too consider somewhat analogous environmental issues within our North American context. Two examples illustrate this point. The first is the food environment. The millions of North Americans categorized as overweight or obese did not plan to develop these conditions. For the most part, overweight or obe- sity happens. Although one has to think intentionally about being healthy, consumers do not have to be as intentional about becoming overweight. Are there inherent biological reasons why health could not also happen to people as unintentionally as overweight or obesity appears to occur? Consumers experience free market forces related to food to a much greater degree than we appear to tolerate other areas of public health and safety. For example, given the perils of unsafe highways and cars, we do not rely solely on educating the public so that they can become better drivers: We engineer safer highways and cars. Are there analogous roles the DRIs can or should play to help safeguard nutritional health, such as modify- ing the width of the Acceptable Micronutrient Distribution Range, or do the DRIs make sense at all without greater specificity in terms of the type of fat? A second dimension of the food environment is our increasing ability to manipulate nutrient intake through fortification, genetic engineering, and supplements. The potential for adverse nutrient interaction merits contin- ued close attention. Perhaps the most salient example of the importance of such considerations is higher than initially projected levels of folate intake and their potential adverse impact on individuals and groups with inad- equate vitamin B12 intakes and/or impaired vitamin B12 uptake capabilities or the progression of early cancer. The second example relates to environments that either enable or dis- courage physical activity. Although we think of physical activity primar-

CRITERIA FOR SCIENTIFIC DECISION MAKING 97 ily in terms of weight status, physical activity also influences the risk of other chronic conditions. Heightened consideration probably will focus on whether nutrient needs are modified by diverse levels of physical activity if chronic disease risk reduction is among the desired outcomes. Genomics In terms of issues that fall under the broad category of genomics, cur- rent considerations for DRI development are limited to genomic variability, which specifically takes the form of including body size (to the degree that size is genetically controlled) in estimating requirements. The broad con- siderations of interindividual variability may be considered, and there may be attempts to address a few specific polymorphisms. Until recently, other than for folate, no other adjustments were made for well-known polymorphisms. Generally, nutrient needs modified by groups of specific polymorphisms were viewed as condition requirements. Among the most salient examples of these are vitamin D-dependent rickets, hemochromatosis, and phenylketonuria, conditions that either increase or decrease appropriate levels of nutrient intake. What about the future? For the most part, complex traits that ac- count for diet-related chronic diseases appear to be influenced by multiple polymorphisms that individually have only modest adverse or beneficial effects on risk, but collectively appear to have significant influence. A study reported recently in the New England Journal of Medicine (Rosenzweig, 2007) that used genomic scanning techniques to assess coronary disease risk supports this view. The value of such work in improving the definitions of risk, enhancing mechanistic understanding, and generating potential interventions for future investigation is acknowledged. For the moment, however, results of such studies appear to have limited immediate impact on specific preventive measures. We also have to recognize that work such as that from Waterland and others (e.g., Waterland and Jirtle, 2003; Waterland et al., 2006) points to the complex epigenetic effects of some nutrients in determining phenotype. There is little doubt that greater understanding of environmental–genomic interactions and the influence of genomic context will result in improved definitions of risk and mechanistic underpinnings. Also, based on what we know now, it is likely that improved under- standing of these relationships eventually will result in better individualized care. What is less clear is how this type of information will help in designing strategies that target populations, particularly as North American societies become more ethnically diverse.

98 THE DEVELOPMENT OF DRIs 1994–2004 Implications Future approaches for DRI development are likely to be increasingly more sophisticated in their inclusion of an array of physiological, environ- mental, and genomic characteristics. The interplay of these factors in deter- mining the prevalence of various phenotypes will need to be recognized, and the interpretation of the special nutrient needs imposed by this interplay will require an expanded DRI process. This increased sophistication will impose important challenges to further address knowledge gaps, mechanistic com- plexity, and the present inadequate understanding of interactions among diverse environmental conditions and individual behavioral choices. Finally, an improved understanding of genomic influences on health will cause us to rethink the use of DRIs in designing strategies to promote individual and population health. General Discussion An audience member commented that genomic variability and the presence of polymorphisms will undoubtedly play an increased role in DRI development. However, after describing the example of methylenetetrahy- drofolate reductase polymorphism, he suggested that the changes involved may not be dramatic. Dr. Garza added that we often forget that these poly- morphisms were positively selected. At some point in our evolution, they must have played some beneficial role. In some context, they may increase risk, whereas in other contexts, they may be protective. Another participant addressed the issue of environmental influences, noting that Dr. Garza had mentioned infectious diseases as pertinent to nutrient reference values for persons in developing countries. Given that inflammation is shown to play a role in the pathogenesis of chronic dis- eases and may be relevant to the aging North American population, the participant questioned whether inflammation should be added to the list as either a physiological or environmental factor to be considered. Dr. Garza responded that aging is germane, and the physiological adjustments and metabolic abnormalities that accompany aging, are relevant to the deriva- tion of future DRIs.

Next: 4 General Guidance for Users of DRIs: Session 3 »
The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary Get This Book
×
Buy Paperback | $60.00 Buy Ebook | $47.99
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

In what ways can the process for developing Dietary Reference Intakes (DRIs) be enhanced? The workshop entitled "The Development of DRIs 1994-2004: Lessons Learned and New Challenges" offered a valuable window into the issues and challenges inherent in the development of nutrient reference values. The dialogue—carried out under the auspices of the Institute of Medicine (IOM), Food and Nutrition Board (hereafter referred to jointly as the IOM)—was enriched by the 10 years of experience in deriving the expanded set of values known as the DRIs, plus the decades of experience that grounded the earlier Recommended Dietary Allowances for the United States and the Recommended Nutrient Intakes for Canada. The lessons learned and the knowledge gained will guide decisions about the next phase of the DRIs. To paraphrase one participant, we are now asking better questions.

In 2006, the IOM, with support from the United States and Canadian governments, undertook an effort to synthesize the research needs identified during the 10 years of DRI development. While the workshop summarized here was predicated on the fact that the development of DRIs is improved by better data, its focus was different. Its goals were to examine the framework and conceptual underpinnings for developing DRIs and to identify issues important for enhancing the process of DRI development.
The workshop was designed to use the existing framework for DRI development as a basis for the discussions and to consider the components of the framework in sequence. Consideration of the pros and cons of the current conceptual underpinnings of the framework opened the workshop, followed by the general "road map" for decision making and the needed scientific criteria. Next, the challenges associated with providing guidance for users were explored. The Development of DRIs 1994-2004: Lessons Learned and New Challenges: Workshop Summary explains an array of issues germane to the future process for developing DRIs, including strategies for updating and revising existing DRIs and opportunities for stakeholder input.
  1. ×

    Welcome to OpenBook!

    You're looking at OpenBook, NAP.edu's online reading room since 1999. Based on feedback from you, our users, we've made some improvements that make it easier than ever to read thousands of publications on our website.

    Do you want to take a quick tour of the OpenBook's features?

    No Thanks Take a Tour »
  2. ×

    Show this book's table of contents, where you can jump to any chapter by name.

    « Back Next »
  3. ×

    ...or use these buttons to go back to the previous chapter or skip to the next one.

    « Back Next »
  4. ×

    Jump up to the previous page or down to the next one. Also, you can type in a page number and press Enter to go directly to that page in the book.

    « Back Next »
  5. ×

    To search the entire text of this book, type in your search term here and press Enter.

    « Back Next »
  6. ×

    Share a link to this book page on your preferred social network or via email.

    « Back Next »
  7. ×

    View our suggested citation for this chapter.

    « Back Next »
  8. ×

    Ready to take your reading offline? Click here to buy this book in print or download it as a free PDF, if available.

    « Back Next »
Stay Connected!