2

NHANES Current Structure and Opportunities for Adding Genomics (Session 1)

This chapter summarizes the presentations and discussion in the first session of the workshop, which focused on the current structure of the National Health and Nutrition Examination Survey (NHANES) and opportunities for adding genomics. Leslie G. Biesecker, workshop planning committee member and distinguished investigator and director of the Center for Precision Health Research at the National Human Genome Research Institute at the National Institutes of Health, moderated the first session.¹

OVERVIEW OF THE NHANES RESOURCE

Alan Simon, director of the Division of Health and Nutrition Examination Surveys at the Centers for Disease Control and Prevention’s (CDC’s) National Center for Health Statistics (NCHS), introduced the NHANES resource. He began by sharing information about the survey, its history, and accomplishments. NHANES started in 1959 as the National Health Examination Survey. In 1971 the survey added a dietary component, and the name changed to the National Health and Nutrition Examination Survey. Finally, in 1999 the survey became continuous, meaning that data were collected every year. Simon noted that the ages of participants and the goals of the

___________________

¹ Video recordings of the presentations and discussions, along with copies of the presenters’ slides for Session 1, are available online at https://www.nationalacademies.org/event/12-02-2022/workshop-on-considerations-for-returning-individual-genomic-results-from-population-based-surveys-focus-on-the-national-health-and-nutrition-examination-survey-day-1-virtual

survey have changed over time. Today, the overall goals of the survey are to produce population-based prevalence estimates and trends on health conditions and risk factors, nutrition status and diet behavior, prescription medications and dietary supplement use, and environmental exposures. Another goal is to establish and maintain a biospecimen program, which is accomplished by collecting biospecimens and preserving them in biospecimen repositories (also called biorepositories).²

Simon highlighted the unique nature of NHANES as a nationally representative survey because it not only asks people questions but “unlike almost any other survey” it also performs medical examinations and tests on this nationally representative sample. Simon explained that conducting the survey in this way allows NHANES to collect high-quality data using standardized procedures. NHANES’s data have been used to identify and solve problems in academia, the private sector, and at all levels of government. For example, NHANES data are used to produce the CDC growth charts, which are used by pediatricians and other clinicians to monitor the growth of children. In 1976, NHANES recorded high blood lead levels in the U.S. population, providing the evidence that pushed Congress and the Environmental Protection Agency to take the lead out of gasoline. After lead was removed from gasoline, NHANES observed a precipitous decline in U.S. blood lead levels. Later, NHANES data identified low iron levels as a serious health problem, especially for women of childbearing age, preschool children, and the elderly, which resulted in the government fortification of grain and cereal. In 2016, NHANES data informed the Food and Drug Administration’s decision to permit fortification of corn masa flour with folic acid to address neural tube defects that were prevalent among the nation’s Hispanic population.

Simon next walked the audience through the operations of NHANES. As a cross-sectional survey of the noninstitutionalized civilian resident population, NHANES collects information each year through in-home, in-person interviews, and through health examinations conducted in Mobile Examination Centers (MECs). NHANES uses a multistage probability sampling design³ to collect nationally representative data from about 5,000 individuals each year, and at times has oversampled groups to get better estimates of health and health behaviors for various subpopulations. The

___________________

² For more information about the NHANES Biospecimen Program, see https://www.cdc.gov/nchs/nhanes/biospecimens/biospecimens.htm

³ Simon explained the process of multistage probability sampling: In Stage 1, counties or sometimes groups of contiguous counties are sampled, with about 15 being selected each year. Stage 2 involves sampling segments of counties. In Stage 3, households within the selected county segments are selected. Finally, Stage 4 involves sampling people within the selected households. NHANES uses the inverse of the probability of selection to create sampling weights that are used to yield a sample that is representative of the U.S. population.

data are released in 2-year cycles. Data collection starts with screening for eligibility and obtaining consent to do the household interview at the household doorstep. During the in-person household interview, interviewers also obtain consent for (a) a household interview and examination at the MEC and (b) researchers to conduct future studies using the data collected. In years when NHANES has collected DNA, participants were asked for consent to use their DNA in future studies.

Simon explained that, during the household interview, NHANES collects demographic information, health conditions, health insurance and health care use, and prescription drugs and dietary supplements. Selected participants are then asked to visit the MEC for their exam, which varies slightly from year to year but generally includes activities such as height and weight collection, hearing testing, in-person dietary recall, and a dental exam. During the exam, samples are collected from participants to look for environmental exposures (e.g., lead, mercury), infectious diseases, (e.g., hepatitis B or C, HIV, chlamydia), nutritional biomarkers (e.g., vitamin D and folate), and chronic-disease indicators (e.g., glycated hemoglobin test or Hemoglobin A1c, lipid profiles). In a 2-year cycle, NHANES runs about 500 laboratory tests, or assays, in more than 22 laboratories. Simon emphasized the advantage of NHANES as a well-standardized survey by sharing that each assay is run at only one lab to minimize method effects.

Simon next discussed NHANES’s current practices around returning individual results. Some initial findings are immediately shared with participants. For example, individuals will be told during the exam if they have high blood pressure. However, most abnormal values are returned to the participant after 3–4 weeks. After 4–6 weeks, participants are instructed to call using a preset password to get results about sexually transmitted diseases. Finally, 12–16 weeks postexam, NHANES sends participants a final report that includes their results. For all tests, only valid and clinically actionable findings are reported to the participant. To be considered valid, lab tests must be performed in a laboratory certified according to Clinical Laboratory Improvement Amendments (CLIA) regulations established by the Center for Medicare & Medicaid Services (CMS).⁴ In order to be considered clinically actionable, a finding must also have significant implications for subjects’ health concerns, and a possible course of action for addressing the concern.

Simon concluded this presentation by offering some additional notes about NHANES that impact the considerations around the return of results. First, NHANES has a very limited duration of contact with participants; the last contact with participants is at the 12–16-week follow-up, when individuals are sent their results. Second, the MEC is not a treatment

___________________

⁴ For more information about CLIA, see https://www.cdc.gov/clia/index.html

facility and does not provide health care. Third, NHANES has a list of local clinics at each site and can provide recommendations for referrals if needed. However, NHANES does not pay for any follow-up health care, which Simon noted “is challenging for a lot of people in the population.”

Simon went on to provide additional context for the workshop by explaining that NHANES is coming to the end of its 10-year contract cycle. NHANES expects to come out of the field in August 2023 and resume operations in 2025. Simon shared that NHANES is considering collecting genomics information when it returns and explained that this workshop will inform incorporating genomics into NHANES’s future data collection cycles. Simon then transitioned to a separate but highly related and relevant conversation about the NHANES Biospecimen Program, which was intentionally given its own discussion on the agenda to reiterate the distinction between the NHANES resource generally and the NHANES Biospecimen Program in particular.

NHANES Biospecimen Program

Simon explained that NHANES’s Biospecimen Program⁵ improves the utility of the data and “extends the life of the survey” because it allows people in the future to look at the biospecimens collected in the past, while also providing a baseline understanding of such topics as emerging infections and chemical exposures. The NHANES Biospecimen Program has over a million samples of sera, plasma, and urine for participants ages 3 and up. The program includes DNA samples from more than 20,000 participants aged 20 and up, which were collected in NHANES III (1988–1994), 1999–2002 and 2007–2012, at which point NHANES stopped collecting DNA samples.

To the extent possible, NHANES makes its data publicly available. This means that data generated using NHANES samples are available on the NHANES website, along with data documentation in order to make the data as useful to as many people as possible. However, some data cannot be released publicly; this includes personally identifiable information (PII), indirect PII, and other sensitive data, as well as genetic or geocoded data.

___________________

⁵ The NHANES stored Biospecimen Program is related to, but distinct from the NHANES “active survey.” The NHANES Biospecimen Program oversees the approval of proposals to use stored biospecimens; ensures appropriate long-term storage of biospecimens at the biorepositories; and arranges for transfer of biospecimens from the biorepositories to the investigators with approved proposals. In the context of this workshop, the term active survey was used to distinguish between data from the primary interaction with the participant versus secondary research using existing data and biospecimens (e.g., from the NHANES Biospecimens Program). For more information, see https://www.cdc.gov/nchs/nhanes/biospecimens/biospecimens.htm. Also see Appendices C (glossary of terms) and D (primer on NHANES).

Those data are available only in the NCHS Research Data Center, where additional strict controls and an additional approval process are in place. Furthermore, researchers interested in obtaining access to NHANES biospecimens must go through a proposal process.⁶ After an investigator submits a proposal, it goes through scientific review with the NHANES project officer and a technical panel. A proposal then goes through institutional review with the NCHS human subjects contact, the NCHS confidentiality officer, and the NCHS Research Ethics Review Board. Specimens are then distributed to investigators with approval. Importantly, Simon highlighted that “at this time only proposals with test results that are determined not to have clinical significance for participants are accepted.” Simon explained that for DNA, clinical significance is defined as being part of the list of the American College of Medical Genetics and Genomics (ACMG) Recommendations for Reporting Secondary Findings.⁷

Simon explained that in 2014, NCHS sponsored a National Academies workshop called Issues in Returning Individual Results from Genome Research Using Population-Based Banked Specimens, with a Focus on the National Health and Nutrition Examination Survey.⁸ He explained that this previous workshop specifically focused on how population surveys should address implementation of reporting results from genomic research using specimens collected previously and stored, or banked, in the NHANES DNA biorepository. In 2016, NCHS used the results of that workshop to update its requirements for proposals using the NHANES DNA repository, specifying that new DNA research being proposed could not generate clinically significant results based on the ACMG list or recommendations. The 2014 workshop explicitly did not focus on issues related to informed consent and returning results for DNA collected in a future data collection cycle.

In conclusion, Simon offered that now is a good time for NCHS to reconsider these issues and think about how NHANES might be able to collect genomic data in the next 10-year contract cycle. He shared that NHANES is seeking to use information provided in this workshop to (a) understand how to approach reporting of genomic results to participants and (b) better understand how to address consent and return of results for DNA that are stored for future research.

___________________

⁶ Since 1993, NHANES has had over 135 proposals for serum, plasma, and urine specimens, and 28 proposals since 2000 for the use of DNA specimens. For more information, see https://www.federalregister.gov/d/2022-14702 and https://www.federalregister.gov/d/2021-17265

⁷ For the complete list and history, see https://www.ncbi.nlm.nih.gov/clinvar/docs/acmg/

⁸ National Research Council. (2014). Issues in returning individual results from genome research using population-based banked specimens, with a focus on the National Health and Nutrition Examination Survey: Workshop summary. The National Academies Press. https://doi.org/10.17226/18829

TYPES OF RESEARCH SUPPORTED BY THE CURRENT SURVEY DATA

The next presentation was provided by Dana C. Crawford, professor of population and quantitative health sciences at Case Western Reserve University in Cleveland, Ohio, and associate director for population and diversity research at the Case Cleveland Institute for Computational Biology, with a secondary appointment in the Case Western Reserve University Department of Genetics and Genome Sciences, and planning committee member for this workshop. Setting the stage, Crawford explained that her presentation would explore some popular research questions and study designs that have been applied to the data files available through NHANES, most of which are public use, with a focus on the current surveys as well as future data collection cycles. She stated, “For each of the study questions or scientific questions and study designs I’ll present, I’m also going to discuss the strengths and weaknesses or limitations that you should keep in mind when accessing or applying the NHANES data resource to these questions.”

Research Using NHANES: Prevalence

A major strength of NHANES, according to Crawford, is that it is a nationally representative sample. The United States has neither a nationally representative cohort nor a national health care system. Investigators asking a very simple question about how common an outcome of interest is in the United States are hard pressed to find a resource other than NHANES to help answer that question. To illustrate why she considers this representativeness to be a major strength, Crawford stepped through an example using the outcome familial hypercholesterolemia (FH), a genetic disorder characterized by lifetime very high levels of low-density lipoprotein (LDL) cholesterol levels in an individual. Untreated FH is a risk factor for cardiovascular disease.

Crawford explained that FH is often described as somewhat of a rare or less common condition, with prevalence estimates of about 1 in 500 adults. She said that the population-based prevalence estimates in the literature come from other countries, such as the Netherlands or Denmark, and do not reflect the demographics of the United States. By accessing NHANES’s public files, investigators were able to define definitive and probable cases of FH; they estimated the prevalence to be 1 in 250—more prevalent in the U.S. population than previously assumed. Subgroup analyses demonstrated or suggested that there might be some differences by race and ethnicity.

Crawford noted that NHANES does have some limitations, such as sample size, that depend on the outcome of interest. For FH, even though it is more common than previously estimated, it is still not very frequent.

NHANES’s 1-year samples contain approximately 5,000 U.S. residents. Each 2-year cycle and any combination of 2-year cycles is a nationally representative sample, explained Crawford. Therefore, investigators must use and access several years of NHANES data to ensure the sample size is sufficient for estimating the prevalence.⁹

A second limitation is that investigators are limited to the variables that were collected by NHANES in a given year. In this case, the investigators applied the Dutch Lipid Clinics Criteria to define probable and definitive cases of FH using NHANES data because they knew NHANES collected sufficient kinds of variables to apply these criteria. More specifically, the investigators had access to LDL cholesterol levels, which have been measured by NHANES for decades. NHANES also has self-reported information about fasting status and lipid-lowering medications, and the investigators opted to use LDL cholesterol levels for the blood samples that were taken in the morning to further ensure fasting status.

The investigators also have access to history of premature coronary artery disease, stroke, and peripheral vascular disease, as well as family history. However, the investigators did not have any data related to physical exams that are specific to FH presentation because NHANES physical exams are more general.

Crawford concluded her points on the prevalence study by reminding audience members that NHANES did not do any genetic testing across the years included in this study, and therefore data were not available to use toward FH genetic testing to define definitive cases. While NHANES offered sufficient data for calculating prevalence in spite of lacking genetic data, Crawford noted that the prevalence estimate is probably an underestimate.

Research Using NHANES: Trends

According to Crawford, NHANES data were designed and can be used to observe trends for an outcome of interest or exposure or risk factor. Since 1999, NHANES has been continuously sampling 5,000 residents per year in 2-year cycles. Having an outcome or exposure of interest that has been measured each year—for decades, in the case of LDL cholesterol levels—can provide a very nice snapshot of the trends over time, Crawford said. Using an example which uses NHANES data from 1999–2014 to show LDL cholesterol levels for approximately 17,000 adult participants in the United States, she highlighted that the NHANES data show two downward trends toward favorable lipid profiles that took place during that time period.

Crawford noted that although describing trends over time is a major strength of NHANES, identifying the variables that cause changes in

___________________

⁹ In this case, the investigators accessed data from NHANES spanning 1999 through 2012.

trends is not. NHANES uses a cross-sectional design, which means that the NHANES program has data that can be used to identify or examine correlations that can then be used to generate new hypotheses about how variables are related or impact one another. Using the changes seen in LDL cholesterol levels over time, she explained, the investigators were able to use NHANES questionnaire data to rule out or deprioritize (find limited if any support for) some hypotheses that they had postulated could account for the inflection points: change in statin use over time, changes in obesity or physical activity, and changes in dietary patterns or dietary habits.

Crawford related that another set of investigators used NHANES to support their hypothesis that changes in dietary levels of trans-fatty acid coincided with changes in LDL cholesterol-level trends. These investigators accessed plasma samples through the NHANES Biospecimen Program in order to directly measure trans-fatty acid levels. While NHANES does not directly measure trans-fatty acid consumption in each cycle, the program does periodically collect the plasma specimens needed to measure trans-fatty acid consumption. Using available stored biospecimens from the NHANES biorepository, the investigators measured trans-fatty acid levels on a subset of NHANES plasma samples collected during NHANES 2000 and NHANES 2009. The data suggested that there was a shift downward in the trend for trans-fatty acid levels between NHANES 2000 and NHANES 2009. Unfortunately, because of the cross-sectional nature of the data, these results cannot be used support a causal relationship between LDL cholesterol levels and trans-fatty acid levels, said Crawford. Also, the sample sizes included in this study were too small for important subgroup analyses.

Crawford continued by raising another natural question that stems from a downward trend in LDL cholesterol levels: given that LDL cholesterol levels are associated with cardiovascular disease risk, has this downward trend resulted in healthier outcomes and decreased mortality? Crawford said that NHANES can be used to explore but not answer this question. NHANES can be linked to public-use mortality files mandated as part of the National Death Index, which are created from death certificate data. Displaying a graphic that highlights the LDL cholesterol-level categories associated with all-cause mortality and mortality from cardiovascular disease from one study, Crawford highlighted that, as expected, higher levels of LDL cholesterol levels were associated with all-cause mortality and cardiovascular disease and vice versa. She noted that the linked dataset contained about 4,500 deaths total in the roughly 23 years since NHANES participants have last contact through the survey. She next reiterated that a major limitation here is that NHANES is a cross-sectional rather than a cohort program, meaning in this example that there is only one LDL cholesterol measurement to combine with the event of death. Crawford provided some additional examples of similar studies and noted that NHANES will

be a better resource for mortality over time. She went on to highlight other public-use data files that can be linked to NHANES data, including CMS Medicare and Medicaid files, Social Security Benefit files, and files from the U.S. Department of Housing and Urban Development.¹⁰

Crawford concluded her talk by highlighting some exciting changes that may come out of the NHANES planning for 2025. New variables and measurements are being considered, including (as evidenced by this workshop) a possibility for collecting genetic data. She also shared that in the future, NHANES is expected to produce faster data (e.g., by using 1 year’s worth of data to generate national estimates rather than waiting on 2 years’ worth of data) and transition to smaller mobile units to facilitate nimbler field operations.

GENETICS 101/201: UNDERSTANDING THE CURRENT GENETIC TESTING LANDSCAPE

Ingrid A. Holm, professor of pediatrics at Harvard Medical School, faculty in the Division of Genetics and Genomics and the Division of Endocrinology at Boston Children’s Hospital, faculty at Harvard Medical School Center for Bioethics, and member of the workshop planning committee, gave the final presentation in this session, laying out the current genetic landscape, with a focus on the presymptomatic predictive testing that is most relevant to the NHANES context. She explained that much of what is known about, and has shaped current thinking about, genetic testing comes from postsymptomatic diagnostic testing, which will also be discussed in this talk.

Genetic Testing in a Direct-to-Consumer (DTC) Context

Holm opened by focusing on the topic most familiar to a public audience: genetic testing in the DTC context. This includes ancestry testing; testing for interesting traits, such as bitter taste perception; and testing for disease predispositions, carrier status, and pharmacogenomics. Holm noted that ancestry testing in the DTC context can connect related individuals. For example, the forensics community has recently used DNA data made available from DTC products to find criminal suspects. In 2018, the “Golden State Killer” was famously identified using this method of forensic DNA analysis.

___________________

¹⁰ See https://www.cdc.gov/nchs/data-linkage/index.htm for a complete list.

Genetic Testing in Health Care

Holm stated that the purpose for conducting genetic testing determines which technologies will be used and impacts the information generated. One of the primary ways genomic or genetic testing has been used over the years is for postsymptomatic diagnostic testing, also known as indication-based testing. Indication-based testing is when a patient has a condition that is thought to be genetic (and usually is rare), and genetic testing is performed to identify the cause. Holm noted that although rare diseases are very uncommon individually, when looked at collectively, the incidence of rare disease contributes heavily to health care costs. Holm described standard techniques used in health care for indication-based testing, including chromosomal testing and molecular testing.

Chromosomal Testing

Holm explained that chromosomal testing is one of the field’s earliest genetic testing methods. Early techniques for chromosomal testing included karyotype¹¹ and fluorescent in situ hybridizations testing.¹² Holm noted that, today, chromosomal microarrays are the most common method of chromosomal testing. Microarrays can provide information about structural rearrangements, aneuploidy, and copy number variations (CNVs), which are deletions or duplications of large pieces of DNA that are usually too small to visualize on karyotype analysis.

Molecular Testing—Focus on Genomic Sequencing

Molecular testing is another indication-based genetic technology that includes genomic sequencing, which is perhaps the most relevant to the NHANES study, said Holm.¹³ Genomic sequencing is performed to determine whether there is change in a single gene related to the patient’s phenotype, which is the indication for the testing. Genetic diseases can be dominant or recessive, autosomal or X-linked, and inherited or de novo. Holm expanded on three methods of genomic sequencing: targeted gene panels, whole-exome sequencing, and whole-genome sequencing.

___________________

¹¹ Medline Plus (2022) defines karotype as “a test to examine chromosomes in a sample of cells. This test can help identify genetic problems as the cause of a disorder or disease.” Medline Plus. (2022). Karotype. In Medical Encyclopedia. National Library of Medicine. https://medlineplus.gov/ency/article/003935.htm

¹² For additional information, see https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7122835/

¹³ Other types of testing that Holm stated seem to be less relevant for NHANES’s current research interests include gene expression, RNA sequencing, and epigenetics.

Targeted gene panels are often used to diagnose genetic disorders in which there are a limited number of genes that are generally responsible for the condition. The multiple genes associated with the indicated disease are sequenced. For example, if a patient has features of osteogenesis imperfecta (brittle bone disease) the genes responsible for the disease are sequenced. A capture kit is used to sequence just those genes, which means these genes are sequenced with very high coverage and depth, so the sequencing results are very accurate. Targeted panels are relatively inexpensive and highly accurate but look at a limited number of genes.

Whole-exome sequencing uses a capture kit to capture the entire exome, which is the portion of the genome that has the genes that code proteins. “The exome is only 1.5 percent or so of the entire genome, but it has about 85 percent of the disease-causing variants,” Holm stated. Because the genomic material sequenced is much larger than in a targeted gene panel, whole-exome sequencing has lower coverage and depth and is more expensive than a targeted gene panel. Still, whole-exome sequencing has good accuracy and is very clinically useful.

Finally, whole-genome sequencing interrogates the entire genome, both the genes and noncoding DNA. The depth and coverage of whole-genome sequencing is the lowest, and the cost is the highest relative to the previously described genomic sequencing methods. Whole-genome sequencing does not use a capture kit; instead, it captures the entire genome fairly equally. The lower coverage of whole-genome sequencing reduces its accuracy. However, unlike the other genomic sequencing methods, whole-genome sequencing can detect CNVs and other structural rearrangements.

Gaps in Understanding Genomic Sequencing Data

Holm explained that different approaches to genetic testing produce different kinds and amounts of data. For example, the human exome contains about 30 million bases and has about 35,000 variants, whereas the genome contains about 3 billion bases and has more than 3 million variants. Holm said that despite the ability to collect this vast amount of information, “the reality is there is much of it that we really don’t understand.” For example, although scientists have identified some genes that cause disease, many more genes have not been associated with disease and still others have yet to be discovered.

Holm pointed out that the clinical implications of many gene variants are not yet understood and are complicated by concepts such as incomplete penetrance, variable expressivity, and gene–environment interactions. Holm describes incomplete penetrance by saying, “just because you have what looks to be a variant that disrupts a gene, it doesn’t always affect the phenotype, and so there may be no phenotype,” which is called incomplete

penetrance. Variable expressivity is “where a change in a gene is known to cause a certain disease, but then it may also cause other conditions, and so the expression of it is variable.” Finally, gene–environment interactions also can complicate the understanding of what genes cause disease.

Holm next distinguished between the understanding of genes and the understanding of specific variants. Even when a gene is known to be associated with a disease, the impact of any given variant within that gene on its function is not always known. Since the clinical importance of specific variants is often unknown, scientists and clinicians need a method for classifying variants to determine the likelihood that a change in a gene affects the protein function. To that end, the ACMG has developed a rigorous variant classification framework to categorize the likelihood that a variant affects protein function and causes disease. Variants are classified as benign, likely benign, uncertain, likely pathogenic, or pathogenic. Variants classified as likely pathogenic or pathogenic are very likely to affect gene function and cause disease. Holm ended this section by reminding the audience that “although there is much we don’t understand about the genome, there are some genes associated with disease that we really know a fair amount about.”

Current Variant Classification and Reporting Practices for Indication-Based Genetic Testing

Holm explained how genomic findings from indication-based genomic sequencing are classified based on whether the finding was in a gene that is related to the reason for testing (the phenotype) or was in a gene unrelated to the purpose of the testing. Holm started by referencing Isaac Kohane’s commentary titled “The Incidentalome,” which discusses the concerns that genome-scale screening tests can have “disastrous consequences” for increasing health care costs “with little benefit to patients or physicians” (p. 212).¹⁴

Holm stated that primary findings are those findings that are related to the clinical indication for sequencing. An unexpected finding is unrelated to the indication for sequencing and can either be secondary or incidental. A secondary finding is a genetic finding that was purposely analyzed as part of the test but is unrelated to the primary testing indications. An incidental finding is a genetic finding that is unrelated to the reason for sequencing but was not sought after by the laboratory; instead, it was found incidentally during the course of analysis.

___________________

¹⁴ Kohane, I. S., Masys, D. R., and Altman, R. B. (2006). The incidentalome: A threat to genomic medicine. Journal of the American Medical Association, 296(2), 212–215. https://jamanetwork.com/journals/jama/fullarticle/211038

Holm explained that all genetic findings are classified using the ACMG variant classification framework. However, whether the variant is reported differs depending on whether it is a primary or unexpected (secondary or incidental) finding. Since the prior probability that a change in a gene that is associated with the reason for sequencing is relatively high, laboratories will report variants of uncertain significance, as well as pathogenic or likely pathogenic variants. In contrast, when an unexpected finding is discovered (a secondary or incidental finding) in a gene unrelated to the phenotype, variant reporting is limited to those classified as “pathogenic” or “likely pathogenic.”

Decisions about which unexpected (secondary or incidental) findings are reported to patients by laboratories is based on the ACMG Recommendations for Reporting of Secondary Findings in Clinical Exome and Genome Sequencing list.¹⁵ The ACMG recommends pathogenic or likely pathogenic variants in genes that are “highly actionable” are sought after and returned in all individuals undergoing indication-based genomic sequencing. Highly actionable genetic findings have (a) high clinical utility, meaning that the test and subsequent interventions are known to improve health outcomes, and the risk of the test is low; (b) high clinical validity, meaning that the test accurately identifies a patient’s clinical status; and (c) high penetrance, meaning that degree of risk conferred by the finding is high. Holm provided some examples of diseases that if linked to a genetic finding would be considered highly actionable, such as some hereditary cancer syndromes, arrhythmias, and metabolic disorders.

Presymptomatic Predictive Genetic Testing

Up to this point, Holm focused on describing what is known about the more established practice of indication-based genetic testing. However, at this point in her presentation, she changed focus to discuss presymptomatic predictive genetic testing, which is more relevant to potential testing conducted by NHANES. Holm explained two types of predictive genomic testing to determine disease risk. First, monogenic disease risk describes genomic sequencing, in which the genome is evaluated for variants in single genes to identify rare diseases. The second method is complex disease risk stratification, in which polygenic risk scores (PRSs) are used to look at risk for more common diseases. Finally, Holm highlighted some ongoing projects that have proven the utility of predictive genomic testing in medicine, including the Implementing Genomics in Practice Consortium,¹⁶ the

___________________

¹⁵ For the complete list and history, see https://www.ncbi.nlm.nih.gov/clinvar/docs/acmg/

¹⁶ For more information, see https://www.genome.gov/Funded-Programs-Projects/Implementing-Genomics-in-Practice-IGNITE

Electronic Medical Records and Genomics (eMERGE) network,¹⁷ and the All of Us study.¹⁸

PRSs and Genome-Wide Association Studies (GWAS)

Holm stated that PRSs are “a way of assessing common genetic predispositions to common complex traits, like diabetes, heart disease, and cancers, where there are many genomic variants plus environmental and lifestyle factors impact the risk of disease.” PRSs are used in precision medicine because they can allow genomic information to be considered in combination with lifestyle, medical history, and environmental exposures to determine an individual’s overall disease risk and treatment plan.

A PRS for any given common disease is calculated using information from genome-wide association studies (GWAS). Genetic testing in GWAS is typically performed using single nucleotide polymorphism (SNP) arrays; more recently, however, genome sequencing has also been used. Holm emphasized that whole-exome sequencing would be insufficient for identifying SNPs because many SNPs are in noncoding regions of the genome. GWAS identify SNPs or single-base DNA modifications associated with a complex disease. Each SNP is associated with a different risk level or effect size. PRSs are calculated by summing the effect size of a number of GWAS risk alleles. Holm pointed out that SNPs are common polymorphisms, each occurring in more than 1 percent of the population. They do not necessarily cause disease but are associated with it. PRSs for any given disease will fall into a normal distribution in both control and diseased populations. “If you have a high polygenic risk score, your risk of disease is significantly higher than it is in those that have a polygenic risk score that indicates low risk for disease.”

Although PRSs provide modest predictions for common diseases, several limitations should be considered. First, common diseases are multifactorial conditions and are affected by factors beyond genetics, including lifestyle and environmental exposures. Furthermore, the measurement of the full genetic signal is imperfect. SNPs are associated with a disease but do not directly impact gene structure or function. Finally, Holm pointed out that their dependence on genetic ancestry is a significant issue in PRS calculation and accuracy. She noted that most PRSs are calculated using reference data generated from mostly White European populations, limiting their use in other ancestries. Because of this imbalance, PRSs have a higher

___________________

¹⁷ For more information on the eMERGE Network project, see https://www.genome.gov/Funded-Programs-Projects/Electronic-Medical-Records-and-Genomics-Network-eMERGE

¹⁸ The All of Us Research Program was discussed at multiple points throughout the workshop. For more information on the All of Us Research Program, see https://allofus.nih.gov/

predictive value in Europeans than in other ancestries. Holm highlighted the need for PRS reference data that are developed specifically in the ancestry of the group or individual being studied to provide an accurate disease risk assessment. She also emphasized that for all these reasons, PRSs should be used in clinically based risk prediction models that consider other clinical risk factors, such as behavioral and environmental factors.

Current Variant Classification and Reporting Practices for Presymptomatic Predictive Genetic Testing

Holm first pointed out that, unlike indication-based testing, presymptomatic predictive genetic testing will not result in any primary findings. Sequencing is performed without a clinical indication; therefore, all findings resulting from predictive genetic testing are secondary. Importantly, Holm highlighted a concern that the clinical utility for returning highly actionable findings is better established for rare monogenic disease findings than for PRSs. When looking at rare monogenic diseases, the practices for classification and reporting of monogenic results are treated like secondary results that result from indication-based testing. In other words, they must meet all the criteria of a highly actionable result. However, ACMG does not provide guidance highly actionable PRSs that indicate risk for common diseases. Therefore, more guidance is needed to determine if and when PRSs should be reported to a research participant.

QUESTIONS AND REFLECTIONS FOR SESSION 1

Biesecker led the discussion for this session. He started the discussion period by offering his opinion that NHANES has some structural attributes that may not be ideal for genetics research, and more specifically, for return of individual genetic results. With his first question, Biesecker asked Simon about his vision for the future evolution of NHANES and whether he has begun to think about some of the attributes that might be more compatible with genetics and genomics research. Simon explained that there were limits on what he was willing or able to say at this stage about the future because the 10-year NHANES contract has yet to be awarded. Instead, Simon encouraged audience members to think broadly about what could be if NHANES could change in small, incremental ways. He reminded the audience that NHANES is a population-based survey and therefore must be representative of the U.S. population. However, unlike any other population-based survey, NHANES can do tasks that can only be done in person, such as drawing labs and conducting physical examinations. Simon also clarified that NHANES is currently focused on its next 10-year cycle, so any thoughts about what future NHANES

could look like would happen in the next 10-year contract cycle spanning 2023 to 2033.

Turning next to Crawford, Biesecker asked her to reflect on three attributes of NHANES that she would change to make it “much more useful for genetics and genomics.” Crawford offered that as an outside investigator interested in genetic-association studies, or how genetic changes are related to an outcome or a phenotype of interest, “it would be useful to have a U.S. resource that’s not biased, much like NHANES, that has genetic data genotypes available to ask how they’re related to the phenotype.” Crawford pointed out that one constraint is that NHANES prohibits genotyping that could be considered clinically actionable, offering, “Even within that constraint, I think you can generate much data on common genetic variants that would not be considered returnable or actionable, and then look at the relationship with your outcome of interest.” Crawford’s second and third wishes for NHANES focused on data access and computational resources. She reminded the audience that investigators must go through a formal proposal process to get access to NHANES’s non-public-use data files, and that if granted access, investigators and their teams must analyze the linked data in the research data center. She explained, “That is different compared with some other data resources that we have available to us as outside investigators for these kinds of analyses.” Finally, Crawford implored that, if NHANES was required to keep the research data center model, it ensure that the analytical environment includes computational resources that support genetic analyses. She explained that the current environment uses SAS^® statistical software, which is not amenable to genetic analyses, and that other computational resources would need to be made available within that environment for investigators to fully leverage the data. Crawford did not offer the exact computational resources she would want to see.

Biesecker next asked if Simon could give the audience a sense of NHANES recruitment over the decades, inquiring about NHANES’s use of incentives. Simon responded that, unfortunately, participation rates have been declining over time, and “since the pandemic, they’ve dropped further.” But, he added, “we’re constantly trying to figure out both how to account for potential biases that we see and also how to improve response rates.” Simon explained that NHANES does provide incentives for people who participate and that the exact amount is an ongoing discussion. When asked by Biesecker whether NHANES is doing follow-up studies to determine how the program values and utilizes the nongenetic results that NHANES has provided, Simon said he was unaware of any during his 2-month tenure and acknowledged the value of such information.

The next question was raised by an audience member asking for clarification or expansion on the “blanket restriction on returning clinically relevant genetic results.” Simon explained that the criteria are based on a

decision from the 2014 National Academies workshop for the banked specimens.¹⁹ Jeffrey R. Botkin, University of Utah, planning committee chair for the 2022 workshop and former member of the 2014 workshop steering committee, explained the rationale: the consent form that NHANES was using at the time stated clearly that participants would not be getting any results back. The Institutional Review Board questioned whether it was ethical for NHANES to produce clinically relevant results and then not return them to the participants (per the consent form). “So it sounds like the response from NHANES was to say we’re just not going to generate clinically relevant information and put ourselves in an ethical quandary with not being able to return that because of the nature of the consent form,” said Botkin.

Biesecker next inquired about NHANES’s current access to staff or other individuals who have the expertise necessary to support the return of genetic results and the current thinking about how to address this moving forward. Simon responded that while they have one or two people on staff who could do something like that, “that’s really different than being able to assure that at any time when you have a result that you’re going to be able to follow up with a participant and tell them what it means. You really need to be staffed up for something like that appropriately, and we are not staffed up for that now.” Simon went on to say that he would not rule out the idea of having the necessary staffing resources in the future based on the current staffing.

Ingrid A. Holm, Harvard Medical School and Boston Children’s Hospital, pointed out that, as reflected in her presentation, exome or genome sequencing can generate incidental findings that may be actionable. Simon clarified that, at present, only targeted sequencing of banked NHANES genetic specimens is allowed. In the discussion that ensued, he also noted that current policies would prohibit whole-exome or whole-genome sequencing of any future genetic samples if NHANES were to collect those. Crawford added that in 2003, when NHANES’s genotyping data were first made available and before genome-wide sequencing was available in general as a technology, she accessed the NHANES dataset to do candidate gene or gene-region genotyping for individual variants. She explained that investigators had to submit a list of the specific reference SNP database ID numbers, or rsIDs, and the rationale for each to the Ethics Review Board (ERB). In addition, investigators had the duty to tell

___________________

¹⁹ National Research Council. (2014). Issues in returning individual results from genome research using population-based banked specimens, with a focus on the National Health and Nutrition Examination Survey: Workshop summary. The National Academies Press. https://doi.org/10.17226/18829. Also see the Introduction in Chapter 1 of this proceedings for additional context and background for the 2022 and 2014 workshops.

the ERB whether or not the data could potentially be clinically actionable. She explained that the 2014 workshop led to the requirement that results be generated in a CLIA-certified environment and offered the perspective that NHANES has not kept pace with the technological developments because NHANES is not the one generating the data. Instead, individual investigators with access to the DNA samples disparately generated data using an approved protocol. She added that NHANES investigators are still limited to targeted genotyping, even though chips may be more economical, because a chip may contain a gene that is listed by ACMG as actionable. “I think that’s the challenge for any lab who would be tasked with future genotyping or sequencing, is they’re going to do these genome-wide surveys because that’s the economical thing to do, but then you have to deal with the data that you generated, whether or not it’s responsible to return those results,” said Crawford.

Biesecker added his opinion that “this notion of farming out samples for ‘bespoke’ phenotyping, in this age of genomics, is a pretty stupendously inefficient and expensive way to acquire a genotype.” He offered that the cost per genotype is significantly lower when performing genome-wide analysis once and then distributing the data (rather than distributing samples to individual investigators to look at specific genes independently). Biesecker asked whether the program could ever consider a model in which the data are generated then shared with investigators who ask the questions and analyze the data. Simon responded that, while he would never rule anything out, the current funding could not cover those activities. Simon also reminded the audience that the survey, as it stands, has very limited contact with people, and it is incredibly hard for NHANES to get back in touch with participants 1–2 years later. Biesecker added that recontacting individuals is a challenge that a lot of genomics researchers have had to deal with over the past 10 years and that this workshop hopes to highlight some lessons learned from current practices in the genomics field that could be applicable here. Crawford reminded the audience of the distinction between the NHANES Biospecimen Program and the NHANES active data collection program, noting that her examples for accessing the DNA samples and generating genotypes from the biorepository as part of the NHANES Biospecimen Program. Biesecker acknowledged that this distinction is important because it impacts an important structural question about whether future NHANES would need to use distributed genotyping versus a single or central model.

Biesecker returned to Holm on the accuracy of exome versus panel testing, which was raised in Holm’s talk. “It’s probably worth clarifying that that is almost entirely due to sensitivity, and not due to lower positive predictive value. That is, what you do find in exomes and genomes is very highly accurate; the concern is that you don’t necessarily find everything

as you would find if you had an optimized single gene or panel test,” said Biesecker. Holm agreed with Biesecker’s clarifying comment and restated that a panel sequences only a limited number of genes with great depth of coverage, while with the entire exome or genome the depth of coverage is much more limited. Using an example of an error in one of the base pairs that has an A instead of a T, Holm explained that such an error can throw the testing off because “you only have a few to compare it to … you don’t have as many reads in any particular region.” The issue in that scenario would be related to the depth of coverage and not related to anything about the variant itself, Holm explained.

This page intentionally left blank.