Genomic Medicine for Undiagnosed Diseases

Summary

One of the primary goals of genomic medicine, utilizing genomics in clinical care, is to improve diagnosis through identification of genomic conditions to improve clinical management, prevent complications, and promote health. In this paper we explore how genomic medicine is being used to obtain molecular diagnoses for patients with previously undiagnosed diseases in the prenatal, pediatric, and adult clinical settings. We focus on the role of clinical genomic sequencing (exome and genome) in aiding patients with undiagnosed conditions despite extensive clinical evaluation and prior testing. In particular, we explore the impact of combining genomic and phenotypic data and integrating across multiple data types to improve diagnoses for patients with undiagnosed diseases, along with how these genomic sequencing diagnoses change clinical management.

Introduction

Genomic medicine is defined by the National Human Genome Research Institute (NHGRI) as “an emerging medical discipline that involves using genomic information about an individual as part of their clinical care (e.g., for diagnostic or therapeutic decision-making) and the health outcomes and policy implications of that clinical use.¹” An introduction to genomic medicine can be found in “Opportunities, Resources, and Techniques for Implementing Genomics in Clinical Care” in this series and key definitions in Panel 1. With the continuing decreases in the cost of DNA sequencing,² clinical exome and genome sequencing are being used across diverse clinical settings with the goal of increasing diagnostic rates and improving clinical management. In exome sequencing, the protein-coding regions (or exons) of the genome are sequenced, while genome sequencing includes both protein-coding and non-protein-coding regions of the genome. For more information on the clinical utility of exome and genome sequencing for genomic medicine see “Building Evidence and Measuring Clinical Outcomes for Genomic Medicine” in this series. In this paper, we use clinical genomic sequencing to refer to the clinical use of exome or genome DNA sequencing and diagnosis to refer to an etiological molecular diagnosis as a step beyond a descriptive diagnostic name for a condition with unknown cause.

Panel 1: Key genomic medicine definitions.

Exome sequencing – seqeuncing the protein-coding regions (or exons) of the genome

Genome sequencing – seqeuncing both protein-coding and non-protein-coding regions of the genome

Clinical genomic sequencing – clinical use of exome or genome DNA sequencing

Monogenic inheritance – a single gene is causative of the disease

Oligogenic inheritance – variants in more than one gene influence disease

Polygenic risk score – calculates the cumulative risk of many genetic variants that all have a small effect on disease risk by using a weighted sum

Allelic heterogeneity – different variants in the same gene leading to variable phenotypes and a spectrum of disease severity

Clinical utility – results can be used to inform clinical decisions and management

Personal utility – results have benefits to an individual or family beyond clinical care

One rapidly emerging area of genomic medicine involves establishing a diagnosis for patients with complex phenotypes (or combinations of phenotypes) that have defied conventional medical evaluation. Initial successes were reported from the NIH Undiagnosed Diseases Program³ and more recently from the Undiagnosed Diseases Network;⁴ in turn, this has led to the global Undiagnosed Diseases Network International (UDNI) effort including programs in 16 countries.⁵ A common element in all the global undiagnosed patient programs that make up the UDNI is the utilization of genomics as an important component of the diagnostic process. The International Rare Diseases Research Consortium (IRDiRC) has also recognized the importance of diagnosis in their global rare disease goals for 2017–2027 with goal 1 seeking to have “all patients coming to medical attention with a suspected rare disease…diagnosed within one year if their disorder is known in the medical literature; [and] all currently undiagnosable individuals…enter[ing] a globally coordinated diagnostic and research pipeline.”⁶ In this paper, we consider a patient to have an undiagnosed disease if the individual has received an appropriate, extensive clinical evaluation based on their presenting symptoms and signs yet remain without an etiologic diagnosis. Such individuals may also have received targeted genetic testing or low-resolution chromosomal copy number analyses (e.g., chromosomal microarray) based on their clinical presentation and/or may have a suspected diagnosis, but no genomic-based diagnosis of disease has been made.

Many patients with undiagnosed diseases are eventually found to have rare diseases. In the United States, the Orphan Drug Act of 1983 and Rare Disease Act of 2002 define rare diseases as conditions that affect fewer than 200,000 people in the United States.^7,8 However, while rare diseases are individually rare, there are so many of them (estimated at ~7,000) that, altogether, they affect 25–30 million people in the United States, amounting to nearly one in ten Americans.⁹ Based on patient surveys,¹⁰ rare disease patients in the United States spend an average of 7.6 years on their diagnostic odyssey to reach a diagnosis, and 5.6 years in the United Kingdom. Reflecting several years spent without a diagnosis, the NIH Undiagnosed Diseases Program reported that pediatric applications peaked in children ages 4–6 with congenital onset of disease and teenagers age 16–18 with onset of symptoms at early school age¹¹. Living with an undiagnosed disease also presents a significant burden on patients and their families. Patients visit an average of four primary care physicians and four specialists before reaching their diagnosis, repeat testing with an average of two to three misdiagnoses, have difficulty locating specialists, receive conflicting treatment guidance, and have difficulty coordinating care amongst clinicians.¹⁰ Beyond the lengthy diagnostic odyssey, adult patients with rare diseases report lower health-related quality of life compared with the population at large and patients with common chronic diseases.¹² While some parents of undiagnosed children have been shown to be tolerant of uncertainty (remaining actively engaged in health care, and having confidence in performing coping behaviors when faced with life challenges), 35–40% also experience anxiety and depression.¹³

With a desire to bring an end to the diagnostic odyssey as soon as possible, genomic sequencing has been investigated as a potential key diagnostic modality for patients with undiagnosed diseases. Given that at most 46% of patients presenting to medical genetics specialists and suspected of having a genetic disorder are currently diagnosed using traditional genetic diagnostic evaluations, comprehensive clinical evaluations, targeted genetic testing, and chromosomal copy number analyses,¹⁴ approaches for improving diagnostic rates are still needed. In this paper, we will explore the impact of combining genomic and phenotypic data and integrating across multiple data types to improve diagnoses in undiagnosed patients, along with how these genomic sequencing diagnoses change clinical management. Figure 1 illustrates a vision for implementing genomic medicine for patients with undiagnosed diseases integrating these points starting from the undiagnosed patient in box #1.

Figure 1.

A vision for implementing genomic medicine for patients with undiagnosed diseases

Combining Genomic and Phenotypic Data to Improve Diagnoses

One significant barrier to reaching a diagnosis in undiagnosed patients is the variable quality and quantity of phenotypic data available to the clinical sequencing laboratory searching for a causative genomic variant. Laboratories performing clinical genomic sequencing use such phenotype data while interpreting the sequence variants they encounter to determine pathogenicity and prioritization of variants in their clinical reports. However, many laboratories report receiving only limited and highly variable phenotypic information from the referring provider (Table 1). In addition, the benefits of data sharing, discussed further below, cannot be realized unless phenotypic data are collected in a structured and readily shareable fashion. One clinical diagnostic laboratory reported receiving variable phenotype information ranging from International Classification of Diseases–version 9 (ICD-9) codes, to completion of a phenotype checklist on the genomic sequencing submission form, to submission of a clinical summary, to multiple clinical notes and laboratory test results.¹⁵ Another laboratory reported receiving case summaries of two to five pages in free text.¹⁶ While many groups generated standardized Human Phenotype Ontology (HPO)¹⁷ terms to capture the primary clinical indication and sometimes other symptoms when comparing diagnostic rates (Table 1), only a few utilized application solutions to produce HPO terms rather than relying on manual curation.^18,19 Phenotype application solutions can aid in the collection of HPO terms and currently include Phenomizer²⁰to rank diseases using signs and symptoms producing a phenotype driven differential diagnosis (http://compbio.charite.de/phenomizer/) and BioLark²¹ a concept recognition tool used to produce HPO terms from clinical notes. Another tool, Phenolyzer²² uses phenotypes combined with prior biological knowledge to implicate genes and disease-associated genomic variants. In a family with multiple conditions, including Prader–Willi Syndrome, Hereditary Hemochromatosis, dysautonomia-like symptoms, Tourette Syndrome, and other conditions, 21 HPO candidate terms were identified.²³ Increasing the number of HPO terms analyzed with Phenolyzer²² in combinations of 1 to 6 of the 21 candidate HPO terms increased the chance of a known variant being prioritized as “high confidence”.²³ This increase in confidence highlights the need for more complete and systematic collection of structured phenotype information, such as HPO terms, along with improved tools for collecting phenotype data rapidly and in an automated fashion, to facilitate interpretation of genomic sequencing for undiagnosed diseases. In Figure 1 this is shown with the collection of structured phenotype data in box #2 before the interpretation of genomic sequencing results in box #3.

Table 1:

Genomic sequencing and phenotyping methods from studies of undiagnosed patients unselected for a specific phenotype published in 2014–2018

Study PMID	First author	Year	# Patients	Age group	Type of sequencing	Individuals studied	Phenotyping method
30304647	Splinter K	2018	382	pediatric and adult	exome and genome	families when available	cross-disciplinary team selects HPO terms
25326635	Yang Y	2014	2000	prenatal to adult	exome	proband with parent Sanger confirmation when available	clinical data provided by referring physician
25326637	Lee H	2014	814	pediatric and adult	exome	proband, trio, and other family members	referring physician reported primary clinical indication and differential diagnosis
28496993	Bick D	2017	22	pediatric	genome	proband with parent Sanger confirmation when available	referring physician case summary
26633542	Retterer K	2016	3040	pediatric and adult	exome	proband and up to 4 family members when available	referring physician provided primary clinical diagnosis and ICD-9 code, used to select HPO terms
26938784	Stark Z	2016	80	infants	exome	proband	diagnostic investigations from referring clinicians and medical records, HPO terms collected at enrollment
28567303	Stavropoulos D	2016	100	pediatric	genome	proband	PhenoTips collection after clinical geneticist exam using HPO terms
26633545	Posey J	2016	486	adults	exome	proband with parent Sanger confirmation when available	available clinical information used to generate HPO terms
25356970	Farwell K	2015	500	prenatal to adult	exome	trios when available	clinical and test history from referring provider summarized by molecular geneticist or genetic counselor
30266093	Normand E	2018	146	prenatal	exome	proband or trio	fetal phenotype converted to HPO categories using Phenomizer
27848944	Trujillano D	2017	1000	prenatal to adult	exome	trios when available	referring provider clinical data used to generate HPO terms
28973083	Meng L	2017	278	infants	exome	proband or trio	BioLark and manual review of clinical notes used to generate HPO terms

One setting in which combining genomic approaches with structured phenotypic information has been adopted to accelerate the diagnostic process is in neonatal medicine for critically ill infants. There is a particularly acute need for accurate phenotyping to support rapid diagnoses in the neonatal intensive care unit (NICU) setting to reduce infant morbidity and mortality.²⁴ The critical impact of early diagnosis and effective treatment on neonatal development has helped drive the adoption of natural-language-processing and machine-learning algorithms to extract structured phenotypes from electronic medical records.²⁴ Combining detailed genomic sequencing and structured phenotypic data can help clinicians direct their diagnostic searches towards suspected genes and diseases (many of which can be pre-defined based on past experience), allowing more rapid diagnoses that can change clinical management. By extracting phenotypes directly from the medical record of infants in the NICU, Farnaes et al. demonstrated derivation of rich phenotypic data to enhance the interpretation of genome sequences leading to provisional diagnoses in as little as 26–48 hours from the time of blood sample receipt.²⁴ Such computational approaches can provide more complete phenotypes as structured data that, when coupled with differential diagnosis gene lists tied to commonly encountered patterns of phenotypes, allow for the creation of computer simulated gene panels, thereby speeding up genomic sequence interpretation in the clinical setting. While such techniques may speed up diagnosis through the use of automation, it should be noted that some novel findings may be missed by such an approach that may be picked up using more traditional clinical genomic sequencing approaches. Implemented as a sequential process, automation can enhance the rapid diagnosis of sick neonates who have a disorder that is familiar to neonatologists, while moving those neonates for whom a diagnosis is not apparent into the more time intensive clinical discovery interpretation pipeline. Sharing data utilizing structured phenotypes also allows for knowledge generation about disease mechanisms through research studies combining genomic data and standardized phenotype terms. Research studies have used multiple HPO terms to describe complex clinical disease phenotypes in a structured fashion allowing for the identification of novel genomic variants associated with specific phenotypes.²⁵ HPO terms can also be used to align human clinical phenotypes to model organism phenotypes to aid understanding of disease mechanisms.²⁵

However, when analyses are limited by observed phenotypes, determining a priori what phenotypes have a genetic etiology in an undiagnosed patient may be difficult. A patient’s clinical presentation may reflect multiple diagnoses (2 or more), yielding a blended phenotype that does not fit exactly with a single condition.^4,26–29 Multiple diagnoses were encountered in 3–7% of diagnosed cases in 5 studies of patients unselected for phenotype.^15,26–29 While the genes implicated in these diagnoses may be related to distinct phenotypes seen in the patient that combine to account for some or all of a patient’s clinical findings, related or overlapping phenotypes may also be caused by genes interacting in the same pathway. Analyses also often assume a monogenic inheritance model in which a single gene is causative of the disease; however, undiagnosed disease may be caused by oligogenic inheritance in which variants in more than one gene influence disease. Many common variants can also be associated with rare diseases, as has been demonstrated for rare severe neurodevelopmental disorders using polygenic risk scores.³⁰ A polygenic risk score calculates the cumulative risk of many genetic variants that all have a small effect on disease risk by using a weighted sum. Additionally, novel diagnoses may come from newly described findings or phenotypic expansions of known conditions^31,32 that may not fit with phenotypic expectations. Therefore, deep phenotyping, that systematically catalogues signs and symptoms of disease rather than focusing on a single primary diagnosis, may assist with disentangling the genetic contributions of undiagnosed diseases. In Figure 1, deep phenotyping is also included in box #2 with an emphasis on the importance of data sharing with box #4 to interpret clinical genomic sequencing in box #3.

Phenotyping can also be a challenge due to the timing of clinical genomic sequencing, as is the case in the prenatal setting. Prenatal imaging, including fetal computed tomography scanning, echocardiography, magnetic resonance imaging, and ultrasonography, can be utilized to detect prenatal phenotypes and help confirm suspected diagnoses. Current practice guidelines recommend chromosomal microarray analysis and karyotyping for fetal anomalies detected by fetal imaging, identifying aneuploidy, chromosomal rearrangements, and copy number variants (deletions and duplications) responsible for these detected anomalies in 30–40% of pregnancies.^18,33,34 However, such testing leaves approximately 60% of fetuses with detected anomalies undiagnosed.¹⁸ Such undiagnosed cases may benefit from rapid prenatal clinical genomic sequencing. Normand et al. demonstrated a 35% diagnosis rate (22/62 fetuses with at least one structural anomaly detected by fetal imaging) using prenatal exome sequencing of trios for ongoing pregnancies often after negative karyotype and microarray.¹⁸ Further studies to enhance our knowledge of prenatal phenotypes and structured phenotype sharing will continue to improve our ability to link genomic variation to these early onset conditions.

Phenotypes can also change over time and have variable intensities in their presentation. For example, some conditions exhibit allelic heterogeneity, i.e., different variants in the same gene leading to variable phenotypes and a spectrum of disease severity. Lysosomal storage disorders (e.g., Fabry, Gaucher, and Pompe disease) exhibit allelic heterogeneity leading to variable age at onset from the newborn period through childhood and into adulthood, highlighting that conditions may not always be easily classified into newborn, childhood, or adult onset. This points to the critical importance of linking longitudinal phenotypic and genotypic data. In addition to collecting up-front phenotypic data, further phenotyping based on the results of genomic sequence data can improve our understanding of how conditions progress and develop from the prenatal to pediatric and adult time periods, and the associated changes in phenotypes. Such data will also allow for determination if a variant explains all of a phenotype improving our biological understanding to help achieve diagnoses and connecting these diagnoses to therapeutic strategies across the lifespan. Up to date information will therefore be critical, requiring that databases of phenotypic and genetic variant information be updated regularly and shared broadly. Figure 1 demonstrates this link between box #2 structured phenotype data and box #3 clinical genomic sequencing interpretation both feeding into box #4 data sharing to generate diagnoses box #6 across the lifetime.

Improving Diagnoses by Integrating Multiple Data Types

Genomic sequencing data can also be integrated with additional data types (such as model organism, metabolome, and transcriptome data) to improve diagnoses as shown in Figure 1 box #5. In the first 20 months of the Undiagnosed Diseases Network, combining clinical exome and genome sequencing data with functional information from studying drosophila and zebrafish animal models led to diagnoses in eight patients, while metabolomics data contributed to diagnoses in three others out of 132 diagnoses.⁴ Clinical assessment by a medical geneticist to obtain additional targeted phenotypic and molecular data, based on information derived from genomic sequence data, has also been shown to lead to more accurate diagnoses and a net increase in diagnoses from 36% to 43% (16 diagnoses promoted to definitive and 5 demoted from definitive to possible or unlikely yielding a net 67 diagnoses out of 155 cases).³⁵ The inclusion of transcriptome (RNA sequencing) data when combined with genomic sequencing information has also improved diagnostic yield, including in cases where the disease specific tissue may be inaccessible resulting in the use of more accessible cell sources such as fibroblasts or peripheral blood mononuclear cells.^36–38 The sequencing of RNA in muscle biopsies from 50 undiagnosed patients with muscle disease yielded 17 new diagnoses in families where prior DNA sequencing had yielded no genomic diagnosis.³⁶ Additionally, sequencing of RNA from peripheral blood mononuclear cells supported the first long-read genomic sequencing diagnosis of a novel pathogenic deletion in PRKAR1A in a patient with Carney complex.³⁸ In Carney complex variation in PRKAR1A leads to degradation of a subunit of protein kinase A, thereby triggering increased activation leading to uncontrolled cell growth, and an increased risk of benign tumors.³⁸ The sequencing of fibroblast RNA additionally provided support for the pathogenicity of a novel, de novo heterozygous IGF2 splice site variant in an Australian Aboriginal, leading to a diagnosis of Silver-Russell syndrome (a syndrome characterized by poor growth before and after birth).³⁷ This finding represents the first Australian Aboriginal family known to be diagnosed with IGF2-related Silver-Russell syndrome and highlights the importance of genomic testing in diverse populations to clarify phenotypic features within and across populations with different ethnic backgrounds. Utilizing additional data (such as those from animal models, metabolomics studies, clinical re-assessment, and RNA sequencing) can thus add to the clinical genomic sequencing results to improve diagnoses.

For rare disease diagnoses, data sharing (Figure 1, box #4) is also critical for identifying additional patients with the same or similar phenotypes, strengthening evidence that variant in a gene is associated, especially when a gene has not previously been associated with any human disease. Matchmaker Exchange (https://www.matchmakerexchange.org/) is an excellent example of a federated data sharing model with multiple connected databases, or nodes, that is designed to allow patients, clinicians, and researchers to share information.³⁹ With 7 databases currently sharing data, Matchmaker Exchange matches patients based on genes and phenotypes. Patients can access Matchmaker Exchange through GeneMatcher (https://genematcher.org/) and MyGene2 (https://www.mygene2.org/), while all nodes support data entry and matching by clinicians and researchers. Sharing phenotypic and genomic data has also been critical to the success of multiple global initiatives to diagnose previously undiagnosed patients in the Undiagnosed Disease Network International (http://www.udninternational.org/).^4,5,40–43 Gainotti et al. highlighted the importance of patient and family participation in undiagnosed research, recommending that patients’ active participation should be maximized. This allows for patients to describe their own phenotype, choose how much information they share in matchmaking databases, and express preferences for what genomic results they find relevant to return to them.⁴³ Such active participation in data sharing drives patient matching and clinical diagnosis and furthers discovery related to the etiology of these previously undiagnosed conditions. Databases, such as the Genome Aggregation Database (gnomAD, http://gnomad.broadinstitute.org/) that share standardized information about allele frequencies seen in the general population and ClinVar (https://www.ncbi.nlm.nih.gov/clinvar/) that share relationships between variants and phenotypes, are also important resources for determining if a suspected genomic variant is seen in the general population and for reporting new variant-phenotype assertions.

Changing Clinical Management Based on Genomic Sequencing Diagnoses

Utilizing collected phenotypic information, clinical genomic sequencing (exome or genome) aims to diagnose patients and change clinical management (Figure 1, box #7). Diagnostic rates for clinical exome and genome sequencing vary dramatically depending upon the patient population (Table 2); however, multiple studies have shown overall diagnostic rates of 25–35% for pediatric and adult undiagnosed diseases.^{4,15,16,18,19,26,27,44–46} In general, diagnostic rates tend to be higher in children and lower in adults. Using proband-only exome sequencing, Stark et al. reported one of the highest diagnostic rates in infants suspected of monogenic disorders at 58% (46/80).⁴⁷ On the other end of the age spectrum, Posey et al. saw one of the lowest diagnostic rates in adults at 18% (85/486).²⁸ The diagnostic rate in adults over age 30 (10%, 24/231) was lower than in adults age 18–30 (24%, 61/285), potentially due in part to the lower availability of parental samples for analysis in older adults. This lack of parental samples limits the ability to detect de novo variants (i.e. variants not inherited from either parent) as part of the clinical report (box #3).²⁸ Environmental effects may also have a greater impact on adult conditions leading to increased non-genetic variation contributing to the phenotype, making it more difficult to identify causative genetic variants. Fewer recurrent molecular diagnoses of different variants in the same gene were seen in adult (11%, 9/85) versus pediatric patients (57%, 266/463) p-value < 0.0001 suggesting that greater genomic diversity may underlie adult disorders.^26,28 Additionally, phenotypes seen in infants may be more severe than those seen in adults due to survivability.

Table 2:

Diagnostic rates across different ages from studies of undiagnosed patients unselected for a specific phenotype published in 2014–2018

Study PMID	First author	Year	Age group	Diagnostic rate
30304647	Splinter K	2018	pediatric and adult	132/382 (35%)
25326635	Yang Y	2014	fetus	6/11 (55%)
			<5 years	247/900 (27%)
			5–18 years	210/845 (25%)
			>18 years	41/244 (17%)
			overall	504/2000 (25%)
25326637	Lee H	2014	pediatric and adult	213/814 (26%)
28496993	Bick D	2017	pediatric	8/22 (36%)
26633542	Retterer K	2016	pediatric and adult	876/3040 (29%)
26938784	Stark Z	2016	0–2 years	46/80 (58%)
28567303	Stavropoulos D	2016	<1 month – 18 years	34/100 (34%)
26633545	Posey J	2016	adults	85/486 (18%)
25356970	Farwell K	2015	prenatal	2/2 (100%)
			0–3 months	6/12 (50%)
			<1 year	7/36 (19%)
			1–5 years	67/194 (35%)
			5–12 years	30/117 (26%)
			12–18 years	19/58 (33%)
			18–40 years	14/45 (31%)
			>40 years	5/36 (14%)
			overall	152/500 (30%)
30266093	Normand E	2018	prenatal	46/146 (32%)
27848944	Trujillano D	2017	prenatal	4/23 (17%)
			<1 year	42/141 (30%)
			1–5 years	128/394 (32%)
			5–15 years	73/285 (26%)
			15–30 years	23/81 (28%)
			>30 years	10/38 (26%)
			unknown age	27/38 (71%)
			overall	307/1000 (31%)
28973083	Meng L	2017	infants (<100 days)	102/278 (37%)

A single phenotype category, as may be defined by HPO terms, can also exhibit variability in diagnostic rate.^{15,18,19,26,28,44–47} Neurological phenotypes are one of the most frequent primary indications for referral, with diagnostic rates from 27% to 42% in six studies.^{15,18,19,26,45,46} Additionally, there are differences in diagnostic rates by phenotype. ^{15,18,19,26,28,44–47} Variability in the phenotype yielding the highest diagnosis rate can also be seen across studies, with 8 studies reporting 8 different phenotypes: hearing (55%)¹⁵, craniofacial (46%)¹⁸, abnormalities of blood (65%)¹⁹, retinal disorders (48%)⁴⁴, obstetric (43%)⁴⁵, connective tissue (44%)⁴⁶, neurometabolic disorder (74%)⁴⁷, and neurodevelopmental (28%)²⁸. It is also important to consider the denominator used when comparing diagnosis statistics between different phenotypes and studies, as illustrated by Trujillano et al., whose study of 1000 families included 771 (77%) with neurological abnormalities, of which 229 received a diagnosis.⁴⁶ Thus, while only 30% (229/771) of the families with neurological abnormalities received a diagnosis, 75% (229/307) of the 307 families diagnosed had neurological abnormalities.⁴⁶

Moreover, multiple studies reported changes in management for 33–94% of patients diagnosed including changes in therapy from receiving a diagnosis (Table 3).^{4,15,16,18,19,27,47} In the prenatal setting, having a diagnosis before birth provides the opportunity for direct intervention, including surgical procedures before or immediately after birth. For example, Deprest et al. noted the opportunity for rapid clinical genomic sequencing to provide diagnostic information about whether congenital diaphragmatic hernia is isolated or associated with other fetal abnormalities, helping to decide if fetal endoluminal tracheal occlusion would be beneficial as a prenatal intervention.⁴⁸ Currently, abnormal genomic findings are used as exclusion criteria for fetal endoluminal tracheal occlusion trials due to a worse prognosis. Diagnostic information can also be used by families to prepare for when and where delivery will occur and to give insights about condition specific care challenges. In Normand et al., prenatal detection of a pathogenic COL1A1 variant allowed the parents time to learn about osteogenesis imperfecta, a condition characterized by bones that break easily, and to connect with other families about strategies to prevent such breaks.¹⁸ Delivery strategies to minimize trauma can also be employed. Diagnostic information may additionally be useful for family planning purposes and have a positive psychosocial impact for parents.¹⁸ In the unfortunate situation where an inevitably fatal condition is identified, care planning includes avoidance of unnecessary and futile intensive care, with development of plans for palliative comfort care that can reduce suffering and financial stress. Rapid genomic sequencing of critically ill infants in the NICU has been shown by multiple groups to change clinical management, ranging from 33–72% in three studies.^19,24,47 Of note, some of the management changes are to palliative care (19/53 or 36% of management changes in Meng et al. and 1/18 or 6% in Farnaes et al.); however, even this change can be of personal utility to the family, as it brings an end to the diagnostic odyssey and clinical utility as it allows for a peaceful withdrawal of invasive interventions that would prove ineffective.^19,24 A test has clinical utility if the results can be used to inform clinical decisions and management, while a test with personal utility has benefits to an individual or family beyond clinical care. Management changes, such as changes in medications and surgical procedures, also led to avoided morbidity in 26% (11/42) of infants in Farnaes et al. including seizure control with a change in medication in an infant with Early Infantile Epileptic Encephalopathy type seven and avoidance of a surgical Kasai procedure in an infant with Alagille syndrome.²⁴ In the Undiagnosed Diseases Network, diagnosis of pediatric and adult patients led to a change in therapy in 28 patients (21% of the 132 individuals diagnosed), a change in care other than therapy (such as changes in diagnostic strategy) in 49 patients (37%), and variant-specific genetic counseling in 48 patient (36%).⁴ Having a diagnosis can affect care from the prenatal to the adult setting with changes in medical management (including starting or stopping therapies or other interventions based on the diagnostic results) and modified genetic counseling based on recurrence risk, which may influence reproductive planning.^{4,15,16,18,19,27,47}

Table 3:

Examples of management changes after diagnosis for studies of undiagnosed patients unselected for a specific phenotype published in 2014–2018

Study PMID	First author	Year	Age group	Management changes
30304647	Splinter K	2018	pediatric and adult	21% (28/132) change in therapy, 37% (49/132) change in care other than therapy, 36% (48/132) variant-specific genetic counseling
28496993	Bick D	2017	pediatric	75% (6/8) impact on medical management or surveillance, 4 changes in medication, 6 medical surveillance
26633542	Retterer K	2016	pediatric and adult	5 reported with suggested intervention or treatment
26938784	Stark Z	2016	0–2 years	33% (15/46) clinical management changed (3 additional treatment started, 5 treatments stopped or modified)
28567303	Stavropoulos D	2016	<1 month – 18 years	94% (32/34) change in clinical management
30266093	Normand E	2018	prenatal	4 medical management changes, 15 reproductive planning, 10 recurrence risk of 19 cases with information
28973083	Meng L	2017	infants (<100 days)	52 infants affected medical management

There are also ethical considerations (Figure 1, box #4) regarding the use of clinical genomic sequencing for diagnosis of previously undiagnosed conditions. The American College of Genetics and Genomics (ACMG) recommends a minimal gene list for analysis as secondary findings when clinical genomic sequencing is conducted across the age spectrum.⁴⁹ Genes on this list as well as additional medically actionable findings including incidental, secondary, and carrier-status findings as determined by the clinical genomics laboratories, are provided on clinical reports.^{15,16,18,19,26–28,44} In the pediatric and prenatal setting, genomic sequencing results returned to parents may have implications for their own health or family planning decisions. Which incidental or secondary results should be returned in the clinical context is also a matter of debate, with professionals having different opinions about what results should be returned in the pediatric setting.^49–55 For adults, carrier results may have direct relevance for reproductive planning, and incidental or secondary findings may have broader implications for other family members. Furthermore, even if a condition is not currently medically actionable, having a diagnosis may still have clinical and personal utility for the family, including an end to the economic and psychological cost of the diagnostic odyssey, the possibility of family member testing, information for reproductive decision-making, and social support related to the diagnosis.⁵⁶ Data protection (Figure 1, box #4) is also critical to ensure that trust in data sharing is maintained while not preventing access to information that may help lead to diagnoses and changes in management.

Conclusion

Genomic medicine can help undiagnosed patients and their families end their diagnostic odysseys. While adult patients may currently exhibit lower diagnostic rates, structured longitudinal phenotyping (Figure 1, box #2) may improve our understanding of the connection between genomic variants and phenotypes across the lifespan. Challenges also remain for improving structured phenotype collection methods, getting laboratories access to the latest data on genes and phenotypes globally, improving methods to integrate additional datatypes with clinical genomic sequencing, and making information on diagnoses sharable in a way that protects patients. Figure 1 also illustrates the central role that data sharing plays in genomic medicine to link phenotypes with clinical genomic sequencing report, additional data types, and diagnoses, leading to changes in clinical management.

Such diagnoses matter, not just for the potential treatments they may lead to, but also for the closure and the peace of mind provided by finally receiving a diagnosis. Diagnoses derived from genomic sequencing findings provide a variety of benefits to patients and their families, including new opportunities for therapeutic interventions and condition-specific management that can lead to improved outcomes and quality of life. The diagnostic results can be used for family planning, cascade testing of other family members, justifying social and educational services, and connecting to other families and condition-specific support groups. Data sharing and data standards are critical to the future success of genomic medicine, since clinical data can inform research discoveries that, in turn, can return new knowledge to inform clinical care. This virtuous cycle is a core component of genomic medicine and will only be enhanced by the new technologies and innovations being brought to bear for the population of patients affected with rare and undiagnosed diseases.