Building Evidence and Measuring Clinical Outcomes for Genomic Medicine

Summary

Genomic sequencing of individuals has potential diagnostic, prognostic, and therapeutic value across a wide breadth of clinical disciplines. One barrier to widespread adoption is limited evidence for improved outcomes in patients who do not already have an indication for more focused testing. We review the current and ongoing clinical outcome studies in genomic medicine and discuss the important features and key challenges to building evidence for next generation sequencing in the context of routine patient care.

Introduction

A vision of genomic medicine is the use of newer broad-based genetic testing by individuals and their health practitioners to enhance routine clinical activities including diagnosis, risk assessment, tailored therapy, and more precise prognosis.^1,2 Rapid advances in laboratory technologies, particularly next generation sequencing, have introduced relatively inexpensive approaches to acquiring a large set of genetic data with potential applications across many specialties of medicine.³ Both widespread marketing of genomic medicine services and health system implementations have increased the availability of testing to patients and their clinicians.^4–6 However, investigations of the clinical utility of genetic testing have often not kept pace, leading to uncertainty about the value of returning findings not related to the original indication and concerns about unintended consequences.⁷ A lack of clinical outcome data has been cited as one significant factor to the slow uptake of genetic testing into clinical guidelines and inconsistent payer reimbursement policies.^8–11 As some clinicians await further study of verification of benefits, and others adopt testing more readily, the assessment of outcomes is increasingly critical to the future practice of genomic medicine.

Outcome studies should be planned as part of a pipeline from discovery to implementation. Similar to phased drug studies, outcome studies can prospectively validate a discovery, demonstrate the efficacy of genome-informed strategies, or assess the effectiveness of an implementation. However, clinical studies of genomic medicine face unique challenges. First, many variants discovered with next generation sequencing (NGS) are rare (present in less than 1% of the population) and have uncertain association with clinically important health states. Secondly, while germline genetic risks are static and detectable from birth, the risk may be latent and expressed decades after measurement. Both of these factors indicate very large study populations with long duration of follow-up will be needed to capture all relevant health impacts. The purpose of this paper is to describe the methods and challenges to conducting clinical outcome studies related to a genomic medicine practice or intervention. Two topics covered in other articles in the series, pharmacogenomics and genetic studies of undiagnosed disease, will not be covered in this review. Rather, we will focus on the international efforts to build evidence for the optimal return of disease risk genetic variants and use of this data within routine clinical care.

A Framework for Collecting Outcomes and Building Evidence

Genetic testing is evolving from individual gene or single nucleotide polymorphism (SNP) variant testing to exome or genome testing using NGS. The discovery of both rare and common variants has increased exponentially in the past two decades.¹⁴ While many variants can be confidently assigned as pathogenic (for example, loss of function BRCA1 variants), others such as novel missense variants, or loss of function variants in genes where the disease mechanism does not depend on insufficient protein production (e.g., PCSK9 and familial hypercholesterolemia) often have uncertain pathogenicity. In fact, only a small fraction of novel variants is sufficiently understood, that when incidentally discovered, are considered for reporting to patients. For example, the American College of Medical Genetics and Genomics (ACMG) recommends that pathogenic variants in 59 genes be returned to tested patients regardless of the indication for sequencing.¹⁵ The association between variants within these genes and specific medical conditions in cardiology, oncology, and many other medical specialties are well established and should contribute to individual risk assessments or to justify additional screening (Table 1). However, whether returning these variants to individuals or family members improves health is often uncertain, particularly for individuals of average risk prior to the testing. Fully capturing changes in health delivery and clinical outcomes across a large study population is challenging when considering the breadth of genomic conditions that could be identified and that important health effects could be latent for decades. Potential outcomes of interest span the individual, their family, and the health system that cares for them. Outcomes in each of these three domains can then be captured across three phases: return of genetic results, the application of the data to clinical decision making, or during longitudinal follow-up (Figure 1). Given that much of genomic medicine will be evaluated in the context of large observational studies and implementation strategies as opposed to clinical trials, linking clinical outcomes to the return and application of genetic results will be particularly important to establishing causality. The scope of reportable outcomes will also depend on whether the focus is on a small panel of genes or a larger sequencing effort such as genome or exome, the timeframe over which outcomes are assessed in sequenced patients, the perspective of the study (societal, health system, or patient-centered) and whether clinical data from family members is sought and captured. In the next sections, we discuss specific outcomes within each domain and how they relate to the larger goal of informing genomic medicine.

Table 1:

Examples of process, intermediate, and clinical outcomes potentially resulting from sequencing studies^*

Genetic Syndrome(s)	Associated Genes***	Pathogenic variant rate among unselected population**	Process Outcomes	Intermediate Outcomes	Clinical Outcomes
Hereditary breast and ovarian cancer (HBOC)	BRCA1, BRCA2	0.5%54	Breast cancer screen modality and schedule	Breast biopsy findings	Prophylactic mastectomy or oophorectomy; diagnosis of breast or ovarian cancer and presenting stage
Lynch syndrome	MLH1, MSH2, MSH6, PMS2	0.4%	Colorectal cancer screen modality and schedule	Colonoscopy findings, Polypectomy	Bilateral salpingo-oophorectomy, incidence and presenting stage of colorectal cancer, ovarian cancer, or endometrial cancer
Familial hypercholesterolemia	LDLR, ABOB, PCSK9	0.4%65	Measurement of low-density lipoprotein (LDL) cholesterol	Initiation or intensification of statin or PCSK9 inhibitor therapy	Atherosclerotic disease: myocardial infarction, cerebrovascular accident, or peripheral vascular disease
Familial hypertrophic and dilated cardiomyopathy	TTN, TNNT2, LMNA, MYH7	0.2%66,67	Echocardiogram screening, CK measurement	Left ventricular wall thickness; Implantation of defibrillator or pacemaker	Diagnosis of cardiomyopathy; Incidence and presenting stage of congestive heart failure
Familial arrhythmia	SCN5A, KCNH2, KCNQ1, RYR2	0.03%68,69	Electrocardiogram (ECG) or Electrophysiology (EP) studies	Medical prophylaxis; Defibrillator placement	Incidence of ventricular arrhythmia or sudden death
Hereditary Hemochromatosis	HFE	0.5%	Ferritin, transferrin saturation measurement	Liver biopsy	Diagnosis of iron overload, cirrhosis, diabetes mellitus, or dilated cardiomyopathy

^*Subset of returnable conditions. For selected rows, distinct genes and genomic diagnoses are grouped by related phenotypes.

^**Approximate pathogenic and likely pathogenic rate; variant rates vary by ethnicity

^***Partial list of genes associated with condition

Figure 1:

Potential Outcomes Measured within Individuals, Families, and Health Systems

Individual Outcomes

Return of sequencing results may lead to changes in an individual’s understanding of genetic findings and clinical risks, anxiety or decisional conflict about the results, changes in health behaviors or lifestyle, or increased information seeking and health care utilization.¹⁶ Any of these psychological effects or behaviors could have a substantial impact on downstream clinical outcomes. For example, a woman who learns of increased breast cancer risk due to a BRCA1 gene variant might react by engaging with her health care provider or a genetic counselor and follow through with accelerated breast cancer screening or, alternatively, might avoid additional follow-up due to anxiety or perceived futility of efforts to prevent a poor outcome. Initial reviews of the literature suggested little effect of genetic testing on health behaviors,¹⁷ but later publications suggest the clinical context, target condition, and aggregation of genetic and non-genetic factors may be important to motivating change.^18,19

The application of results to clinical risk prediction, the performance of additional screening tests, and receipt of an individualized intervention are key process outcomes that link return of sequencing results to patients and potential improved health (Table 1). Relatively few guidelines are published to help clinicians manage patients with identified genetic risks, but the ones that are available help define outcomes of interest through recommendations for additional diagnostic testing, accelerated screening or surveillance schedules for cancer risks, or risk reduction with medical or surgical prophylaxis.^20–26 As an example, the National Comprehensive Cancer Network (NCCN) defines a surveillance strategy of colonoscopy starting at age 20–25 with a repeat every 1–2 years for patients with a known Lynch Syndrome related pathogenic variant, and also suggests bilateral salpingo-oophorectomy be considered as a risk-reducing option for women with Lynch Syndrome who have completed childbearing.²⁶ In a cohort study of patients with a variant in one of the Lynch Syndrome associated genes (MLH1, MSH2, MSH6, PMS2) all of the medical and surgical therapies used to reduce risk should be captured, in addition to tracking the incidence of colorectal, ovarian, endometrial, and other cancers associated with the syndrome.

Family Outcomes

One unique feature of genomic medicine studies is the potential for genetic risks identified within an individual (the “proband”) to impact the care of family members through cascade testing of relatives. Ideally, clinical outcome studies would test 1^st and 2^nd degree family members of study participants with a pathogenic variant and track family members (with the consent of the proband and family member) for changes in health care delivery and clinical outcomes. Given that family members are much more likely than average to also have the variant, cascade testing will increase the efficiency of the study and health impact of the original finding. This has been demonstrated in the context of screening patients with colorectal cancer to identify Lynch Syndrome; cascade testing of relatives prior to development of cancer was found to be cost-effective.²⁷ One recent study showed that as many as 47% of first-degree relatives of the proband complete such testing, when low cost and convenient, although other studies show lower uptake.^28,29 In practice, investigators need to plan to overcome significant logistical and health policy hurdles to conduct cascade testing.^29,30 If successful, the number of patients with a variant of interest to contribute to outcome assessment could nearly double, as was found in a biobank study of Estonian patients with familial hypercholesterolemia³¹

Health System Outcomes

One of the most important mediators of individual outcomes is the context where the test results are obtained and applied to clinical care. Testing that is done within an established clinician-patient relationship and in a health care environment where results are interpreted, and clinical decision support is provided to both patients and clinicians, may have substantially different impact than testing done in other contexts such as direct-to-consumer or outside a traditional health system. The availability of services, such as genetic counselors and medical geneticists, could predict improvements in process outcomes such as effective delivery of results and recommendations to patients, but also incur costs to the system in the form of increased healthcare utilization which could reduce overall cost-effectiveness.

Challenges and potential solutions to conducting genomic medicine outcome studies

Building evidence for genomic medicine with outcome-oriented studies involve a host of considerations: the rarity of the returned variants, heterogeneity in minor allele frequencies (MAF) between different ethnicities, incomplete penetrance, the pleiotropy (i.e. heterogeneity) of gene functions, the age of onset of the target conditions, epigenetic effects representing interactions between the environment and gene risks, and differences in disease expression between the sexes among other issues. All of these factors suggest very large and diverse study populations will be needed to comprehensively determine the impact of sequencing on human health. Putative pathogenic variant rates for the more common Mendelian conditions are present in less than 1% of an unselected population (Table 1). Estimates for population-based variant rates across the entire ACMG 59 gene set range from 1–3%.^32–35 A study of 10,000 participants would be expected to yield only 100–300 with any variant and typically less than 100 participants with a variant associated with a specific phenotype. Thus, even a large study with thousands of patients would have difficulty discriminating between outcomes in patients with and without a variant or comparing tested to an untested population. Conducting comparative effectiveness research to test a genome informed strategy would also need to account for incomplete penetrance or penetrance that is strongly dependent on age. For example, among patients with multiple endocrine neoplasia type 1 (MEN-1) age-related penetrance can vary from 7–10% in the young to nearly complete penetrance by age 60.³⁶ Unless cohorts are followed for decades, the ability to detect a phenotype will be strongly affected by having the appropriate distribution of ages in the study cohort.

Potential solutions

Several authors have pointed out that precision medicine in general, and genomic medicine in particular, would benefit from a convergence of implementation science and a learning health system to measure outcomes and generate evidence across a large population.^37,38 Implementation science has been defined as the “the scientific study of methods to promote the systematic uptake of research findings and other evidence-based practices into routine practice, and, hence, to improve the quality and effectiveness of health services and care.”³⁹ To support outcome evaluation for genomic medicine, the implementation must effectively deliver genetic data to patients and clinicians and provide support for clinical decision making, while the learning health system aspect should capture the process and clinical outcomes during routine clinical care. At least one NIH funded network is making progress toward this goal. The IGNITE network (ignite-genomics.org) has developed a model and associated demonstration studies to facilitate implementation studies in genomic medicine.^4,40,41 The model’s implementation outcomes are based on the principles of the RE-AIM framework (www.re-aim.org/): Reach, Effectiveness, Adoption, Implementation, and Maintenance.⁴² Reach describes the target population, including how generalizable it is, and the uptake of the intervention in that population. Effectiveness addresses the impact of an intervention on important outcomes, including potential negative effects, quality of life, and economic outcomes. Adoption describes the target health setting or provider population and the uptake, Implementation describes the cost and the ability to implement the intervention as designed, and Maintenance describes the consistency of intervention use over time. The importance of how different dimensions of implementation might impact the effectiveness of a genomic medicine intervention is shown in the following example: if an intervention is completely effective but only reaches a minority of the population or adopted by a minority of providers and poorly sustained, the impact of the intervention will be minimized and create potentially imperceptible health outcome differences.

Methods for Capturing Outcomes

The need for large studies will necessitate efficient, low-cost strategies for collecting outcomes. Several existing genomic medicine networks have demonstrated the value of electronic health records (EHRs) in aggregating phenotype data across large populations for both discovery and outcomes assessment within a genomic medicine implementation.^5,43 The prospect of using EHRs for population-based outcomes research has improved with broad implementation across many of the countries which are also investing in genomic medicine studies^44,45, and with the development of public resources such as the Phenotype KnowledgeBase (phekb.org) to define phenotypes in terms of EHR data algorithms.⁴⁶ Sharing of data across diverse clinical environments will also address the challenge of conducting large outcome studies. The introduction of a common computable languages to represent coded clinical data and phenotypes – for example, the Observational Medical Outcomes Partnership (OMOP) common data model- is expected to accelerate the trend of merging data across many health systems.^47–51 While these capabilities are important, additional work is needed to ensure the electronic phenotype algorithms applied to the data in large biobanks or other clinical data repositories are sufficiently validated, reproducible, and specific to the outcomes of interest for genomic medicine.

The State of Outcome Studies in Genomic Medicine

The effect of sequencing on clinical outcomes is a subject of active investigation and has led to the establishment of dozens of large cohorts internationally (Table 2). These are primarily organized as a prospective cohort or biobank study with the added dimension of return of results planned around the ACMG 59 set of genetic criteria or analogous internally developed criteria. Federally funded consortia (eMERGE III, All of Us, Million Veterans Program) within the US, academic-industry partnerships (Geisinger MyCode Community Health Initiative), and national (UK 100,000 genomes project, Estonian Genome, Genome Canada) are expected to produce critical information about process and clinical outcomes over time.^31,52,53 While ongoing, many of these have already begun reporting process outcomes. In a study of 50,000 women participating in the MyCode Community Health Initiative who were evaluated for BRCA status, 75% of carriers of a pathogenic or likely pathogenic BRCA variant were not identified as carriers as they had not had clinical testing, thus were not receiving recommended care.⁵⁴

Table 2:

Selected large cohort studies which return results to participants and conduct longitudinal follow-up.

Study Name	Type of Genetic Data	Source population	Planned enrollment	Enrollment (as of 11/2018)	Genetic and clinical focus of program
All of Us70	Sequencing	United States (US)	1 million	76,000	Clinical conditions associated with the ACMG 59 and drug response related to pharmacogenes.
Genome Canada71	Sequencing	Canadian	30,000	0	Rare genetic disease
eMERGE Network (3rd round)5	Targeted Sequencing	US based healthcare network	25,000	25,000	Clinical conditions associated with the ACMG 59*
Estonian Genome Project31,72	Genotyping	Estonian	150,000	52,000	Rare genetic disease and familial hyperlipidemia
Geisinger MyCode Community Health Initiative52,73	Exome sequencing	US based integrated health system	500,000	225,000	Clinical conditions associated with a Geisinger defined gene list
UK 100,000 genomes project74**	Whole genome sequencing	United Kingdom	100,000	87,231	Rare genetic disease and cancer

^*ACMG 59 is a list of genes curated by the American College of Medical Genetics that are returnable in the context of sequencing regardless of the indication for testing.

^**Expansion announced

Relatively few studies are in the form of a clinical trial. The pilot MedSeq project randomized healthy primary care patients to whole genome sequencing in a primary care setting and found that primary care providers took clinical actions in 1/3 of patients with a medically actionable secondary finding, and that downstream costs did not rise in response to the return.⁵⁵ MedSeq found 2% of tested participants had a Mendelian trait linked to a phenotype but was underpowered to determine the penetrance of the variant or changes in clinical outcomes related to the genetic testing.

Implications of Outcome Research

As the practice of genomic medicine expands using next generation sequencing technologies, patients in many different clinical contexts are likely to be recipients of genetic findings that were not related to the indication for testing. Conducting clinical outcome studies in this population is critically important to developing evidence-based policies around the return of secondary findings and guidelines for genome-informed care. While randomized clinical trials would provide the strongest evidence for clinical benefits or harms for returning specific types of variants, the uncommon frequency and latency of genetic variation strongly associated with disease make such trials expensive, impractical, or pre-empted by compelling observational data. We anticipate that large cohort studies which return results to participants and follow participants over time will gradually inform the use of sequencing results in clinical practice, but this process may take decades to complete for conditions associated with rare variation. Several steps can be taken to accelerate the conduct of such studies and dissemination of the findings. As yet, there are no standard outcome measures for cross-study use, but collaboration and interchange between National Institute of Health (NIH) consortia are beginning to define common methods.⁵⁶ Standard outcome approaches will enable the aggregation of outcome data across different study populations, a feature that could overcome the inherent need for ever larger study populations to assess penetrance in rare variants and the associated change in clinical outcomes. Development of publicly available standardized outcome measures, can rapidly expand knowledge as demonstrated by the Patient-Reported Outcome (PROMIS) measures, initially developed with NIH funding and now in broad use for research and increasingly in clinical care.⁵⁷ Secondly, offering cascade testing to families of proband study participants will increase the efficiency of identifying carriers; if these family members are also followed clinically, the pace of determining outcomes will be significantly amplified. Finally, national and international consortia, similar to the ClinVar and Clinical Genome Resource (ClinGen),^58–61 could accumulate evidence based algorithms for managing secondary findings, just as existing resources catalog the clinical relevance of genes and variants.

Reporting how individual and family health is improved by return of sequencing results will help address several of the barriers to genomic medicine adoption. Currently, adoption is limited by clinical inertia and inadequate strategies for accelerating clinical practice guideline adherence in instances where definitive diagnosis and treatment are available. For example, nearly half of FH patients do not receive the recommended treatment and fewer than 1 in 4 patients eligible for high-intensity statin treatment receive it or achieve treatment targets.⁶²

Across all study types, study investigators need to be alert to potential bias and limitations. Clinical outcomes may be markedly influenced by selection biases. Study cohorts that accumulate patients with specific phenotypes (e.g. an existing diagnosis of cancer or rare disease) will have an ascertainment bias that will not be correctable during analyses. Secondly, the age at enrollment of study participants may strongly impact outcome assessments; for example, an older study cohort may obscure increased mortality at younger ages due to survival bias. Thirdly, the need for observational, nonrandomized study designs to satisfy large recruitment requirements could increase the impact of confounding on the evidence base for the field. Finally, it may be difficult to determine that the outcome is attributable to the genomic result; for example, was a mammogram done in a woman after the return of a genomic result performed in response to the result, or as part of regularly scheduled preventive care? While some interventions can be confidently attributed to the return of the results based on timing and rarity of the test in routine care, (e.g. a serum ammonia level after return of a pathogenic variant in OTC), others such as the mammogram example warrant more discretion to avoid confounding. As the discovery of rare variants with predicted pathogenicity accelerates, the risk of false positive and false negative outcome associations increases.

The practice of genomic medicine is expected to expand from the identification and care of patients with single-gene Mendelian disorders to more common conditions with complex genetic associations. The development of polygenic risk scores to predict the onset of cardiovascular disease in adult patients is an early example.^63,64 As these disorders have multifaceted etiologies with established clinical risks, high dimensional genomic risks involving thousands or millions of variants, and potential epigenetic risks, outcome evaluations will need to compare clinical to clinical-genomic strategies at a scale that can differentiate the incremental benefit of adding genomic data to a standard clinical risk model.

Conclusion

Building evidence to use genomic sequencing to individualize preventive care strategies, improve early diagnosis of genomic syndromes, and to tailor therapeutic plans will require an extensive international effort to recruit and follow large, diverse study populations for clinical outcomes. Increased emphasis on implementation research will help achieve the scale and identify sustainable strategies for accelerating the adoption of guideline-recommended practices.

Glossary of Terms

Genome sequencing

sequencing of nearly all DNA

Exome sequencing

sequencing of protein-coding DNA

Germline

DNA that is present at birth

Genomic variant

an alteration in DNA, relative to the reference sequence

Minor allele frequency

rate of common (≥1%) or rare (<1%) variants within a population

Pathogenicity

a probabilistic assertion that a variant is causally related to a heritable disease

Penetrance

likelihood individual with variant will manifest the disease

Actionability¹²

the degree to which a variant should influence individual behavior or health care

Epigenetic effect

any process that alters gene activity without changing the DNA sequence

Secondary findings

identification of a variant that is associated with a condition other than the one for which testing was originally indicated

Pleiotropy

when a gene influences two or more phenotypic traits

Clinical utility¹³

How likely the test is to significantly improve patient outcomes