To identify and document clinical tasks and decision-points from referral to discharge, we mapped operating procedures within three PGMP genomic medicine practices and discussed procedures and workflows with medical geneticists, genetic counselors, program managers, and clerical staff (Figure 2 and Supplementary material section 1). In the context of prior thought, (7–11) these clinicians synthesized the biological principles that govern their decision-making in daily practice (Figure 1). The resulting workflow was used to design, prototype, and implement an electronic data management system as a quality improvement initiative within the PMGP of British Columbia.

Having mapped the processes of genomic medicine practice and defined the logic of the biological and utilitarian practice decisions, we built the genomic medicine practice process into an instance of the research electronic data capture (REDCap) system and implemented this infrastructure as a quality improvement initiative within the PMGP. Previously identified data inefficiencies and vulnerabilities were addressed within the REDCap platform by building instruments to facilitate referral management, triage, patient assessment and phenotyping, testing coordination, and results tracking (Supplementary material section 2). During the first year of use, the implemented platform was iteratively optimized with input from over 50 clinicians and staff and was used to manage 10,000 patient records.

The process of genomic medicine practice engages five key players: the patient, the principal care provider (PCP), the clinical genomic specialist (CGS), the laboratory genomic specialist (LGS), and the payer. The roles and responsibilities of each entity are likely assumed by a team (Table 1). The CGS delivering genome-informed care might be a team of a specialist physician, genetic counselor, and support staff, and, in the context of this work, the LCS predominantly represents those engaged in the interpretation and reporting of genomic variation while recognizing that such is built on the analytical validity of the data generated by the laboratory.

Within the process map, the decisions and tasks for referral, triage, assessment, testing, and management are grounded in a biological framework that assesses the likelihood of a unifying genetic cause or a contributing genetic factor. The biological principles and the corresponding clinical attributes of the patient are depicted in Figure 1. The clinical attributes of the patient (i.e., the phenotype), which are ascertained through the process of medical and family history taking, examination, and investigation, are evidence in a Bayesian likelihood framework (Figure 3A and Supplementary material section 4A). Currently, because of the lack of population and aggregate data, this framework is a semi-quantitative approach providing logical guidance, rather than a fully quantitative approach.

In the Bayesian approach, the clinical attributes provide a likelihood of genetic disease, and the specificity of the features suggest the likelihood of a particular disease. Together these represent the prior probability. Clinically, these likelihoods reflect the physician’s acumen in differentiating individuals more likely to benefit from genomic testing and the physician’s suspicion for a specific genetic diagnosis prior to genetic testing.

The prior informs the probability of a variant’s pathogenicity and thus the yield of genetic testing in the patient (Figure 3A; 12, 13). The biological framework governing genomic medicine is thus intimately linked with Bayesian logic and relies on the CGS to apply the biological framework with Bayesian logic when assessing the likelihood of genetic disease, selecting testing, determining diagnosis, interpreting results, and managing patients.

Given that resources in every society are limited, a utilitarian framework allocates resources such as personnel, cognitive investment, and testing for genomic medicine practice (Figure 3B). Utilitarian principles assess the expected benefits to the patient, family members, payer, and society in order to maximize service and minimize opportunity costs (14). Applied concurrently with biologic principles, this approach optimally prioritizes those most likely to benefit from genetic testing, i.e., achieve a diagnosis or precise therapy (Figures 1, 3B; 15, 16). Genomic medicine is complimentary to societal utilitarianism in that rapid and precise diagnoses lead to more informed care and better utilization of resources.

This stage pertains to steps 1–14 of Figure 2. A phenotype-first approach, which identifies those most likely to have genetic disease and benefit from testing, remains common in genomic medicine practice. Evidence gathered through collaboration of the patient, PCP, and CGS informs the likelihood of a genetic disease and identifies potential diagnoses prior to genomic testing.

As for many clinical processes, care begins either when a patient or a family member of an affected individual communicates a concern to the PCP or the PCP notes a concern on a routine health check. Although dependent on the medical care organization within a society, in many societies the PCP determines whether the clinical presentation meets a threshold of perceived probability and importance that warrants a genomic assessment. Whether patient or PCP initiated, referrals to a CGS are typically framed as requests for further expertise or for a service to address a clinical question. In the latter scenario, the PCP refers the patient to a CGS when (i) the patient’s clinical presentation warrants evaluation that is outside of the PCP’s knowledge of genomics, (ii) genetic investigation is warranted and test selection and/or result interpretation is outside of PCP expertise, or (iii) genetic counseling is warranted and is outside of PCP expertise. CGS referrals requesting diagnostic services must contain (i) sufficient documentation of the clinical attributes suspicious for genetic disease as per biologic principles and (ii) a referral question that is potentially resolvable by CGS assessment and/or by genetic testing.

Electronic collection of additional information, consent forms, pre-appointment intake questionnaire, and possibly photographs from patients prior to the clinical appointment empowers informed, efficient, and effective patient-provider interactions and increases patient self-advocacy and satisfaction (17, 18). Prior collection of medical, developmental, and family histories and the review of systems increase the efficiency of CGS clinic visits and might suggest a phenotypic pattern indicative of specific diagnoses that warrants a targeted evaluation.

This stage pertains to steps 15–22 of Figure 2. Following the assessment and an evaluation of the likelihood of a genetic disease, the CGS considers the potential clinical utility of genomic testing. For transparent and objective determination of utility, we generated and piloted a tool for patients with intellectual disability (Supplementary material section 3). This tool demonstrates the thought process to assess whether testing is indicated for a patient (step 15 in Figure 2) and how missing clinical history or patient features skew perceptions of the potential utility of testing. Assessment by the CGS often occurs at a single time or over a short interval. Awareness that perceived utility of testing fluctuates with phenotypic evolution is, therefore, critical to understanding negative test results for those whose symptoms resolve and for those who need re-referral when symptoms persist or progress (Supplementary material sections 4A, B).

Test selection is informed by disorders that the CGS considers as possible causes for the clinical presentation, by the molecular mechanisms by which these disorders arise, and by the laboratory test characteristics. A non-specific phenotype has a broad differential and often warrants genomic profiling (WES or WGS). For example, appropriate testing for a patient with infantile hypotonia includes WES and assessment for SMN1 deletion, expansion of the CTG trinucleotide repeat in the non-coding region of DMPK, and SNRPN methylation; this highlights the need to understand the molecular mechanisms of disease and the technological limitations of each test. As a result of its comprehensiveness, exome (and genome) sequencing is a cost-effective first tier test for many indications (15, 16, 19), although an exome-first approach does conflict with some jurisdiction-imposed tiered genetic testing strategies (e.g., requiring a chromosomal microarray prior to funding whole exome sequencing).

A patient’s decision to pursue testing is informed by perception of the test’s utility (Figure 3B). Patients understand and assess utility as presented by the CGS via the process of informed consent. This includes an appraisal of the possible benefits, risks, limitations, results, and implications of the genetic findings to the patient and family members (5, 20). Additionally, disclosure of results is guided by understanding a patient’s values regarding the utility of testing.

The decision whether to fund a proposed test is made by the payer, i.e., the patient or an insurer. The payer assesses precedent and considers a utilitarian framework to decide whether to allocate limited resources. Providing the evidence of medical necessity is the responsibility of the CGS. Pre-determined genomic testing eligibility criteria or case-by-case utility assessments guide payer decisions (21). Decisional transparency by payers is essential to ensure equity and advocacy, which is needed when testing is supported by the biologic principles but the payer declines funding (14, 22).

This stage pertains to steps 23–24 of Figure 2. The next stage of genomic medicine practice encompasses genomic analysis and interpretation of the variation to evaluate potential molecular diagnoses. The LGS ensures the quality of sequencing, alignment, genotyping, annotation, variant filtering, classification, and reporting. Filtering and prioritization of variants for interpretation are guided by the patient phenotype and by the variant attributes (Figure 4 and Supplementary material section 5). Prioritizing variants based on the concordance of the patient phenotype with that of the disease(s) arising from variation at a locus optimizes the sensitivity and specificity of genetic testing (23, 24). Phenotype is often provided as a CGS-curated list of pertinent positive and negative signs, traits, and symptoms using the vocabulary of Human Phenotype Ontology (HPO). As ontological entities with information content and relationship, HPO terms provide a standardized language for communication, enable software-prioritization (e.g., Exomiser and LIRICAL) of genomic variants and facilitate standardized electronic exchange of phenotype as phenopackets (25, 26). In lieu of selecting HPO terms, natural language processing is increasingly used for comprehensive phenotype recognition from medical notes to ensure a high recovery of clinical keywords for downstream genetic analysis (e.g., Virtual Geneticist). These terms are commonly supplemented with prose describing the natural history and family history.

Variant attributes considered include frequency in affected and unaffected populations, potential deleteriousness of the variant, congruence with the established disease mechanism, conservation of the reference nucleotide and amino acid across species, and segregation of the variant with the disease in question (17). Variants are classified by the LGS according to the criteria of professional bodies such as the American College of Medical Genetics (ACMG) (17). The consistent application of these criteria should result in a variant receiving the same laboratory classification regardless of the LGS completing the analysis or the clinical features of the tested individual.

Regarding which variants to report, the LGS is guided by expert opinion and professional consensus, i.e., published guidelines (22). For example, the LGS generally does not report a single pathogenic allele in a gene that causes recessive disease, unless there is sufficient evidence to suggest concordance between the patient’s phenotype and the disease in question (Figure 4 and Supplementary material section 5). Similarly, the LGS generally limits reporting of variants of uncertain significance to those with the potential to cause diseases concordant with the patient phenotype. In both situations, the extent of phenotypic concordance needed for reporting of a variant is left to the judgment of the LGS. Within the ideal state of collaboration, the CGS provides a sufficiently detailed phenotype to allow optimal variant prioritization, and the LGS, reciprocally, provides evidential transparency for reported variation.

Given the nuances of genomic testing and variant reporting, the CGS contextualizes the laboratory-reported variants to establish a diagnosis. To formulate a clinical conclusion, variant interpretation considers factors inherent to a variant in parallel with the existing knowledge about the locus (gene-disease causality, disease-mechanism concordance) and the patient’s clinical presentation (phenotype-disease concordance, segregation) (Figure 5 and Supplementary material section 6 for case examples). The evidence from this logical framework modifies the prior of the Bayesian equation (Figure 3A) such that the posterior probability informs the likelihood that a variant, Variant A in Locus A, contributes to the patient’s presentation.

Completion of the logical framework often requires further evaluation or testing such as segregation, reverse phenotyping, or functional testing. For instance, variant segregation in the family determines if the identified variants (i) for a recessive disorder are in cis or trans, (ii) arose de novo, or (iii) segregate with the phenotype in the family. Segregation data requires an understanding of the penetrance and expressivity of a condition; for example, when the locus in question has previously shown incomplete penetrance for a disease, segregation to unaffected individuals in a small pedigree is insufficient to prove that a variant is benign. The likelihood that the variant in question contributes to a particular disease is modified by assessment of the patient for phenotypic concordance with the disease in question and by functional assays (Figure 5 and Supplementary material section 6). Of note, even if these studies do not lead to variant reclassification within the ACMG guidelines, the data might be sufficient for the CGS to decide whether a variant contributes to the patient’s presentation and if contributing, whether it contributes a full or partial explanation.

Overall, approximately 25–30% of exome sequencing results require further testing by a variety of methods to support the diagnostic process (18, 27). Of variants with ambiguous pathogenicity after these additional evaluations, most will not require further patient intervention until their significance is clarified; current understanding predicts that over time most of these variants will be re-classified as benign (20). Rarely, the CGS might judge ambiguous variants actionable and make management recommendations. Appropriate patient counseling and collegial consensus in these rare scenarios is paramount.

A challenge for the diagnosis of rare disorders is determining whether features previously unobserved with a disease represent an expansion of the disease phenotype or a second disease. Multiple molecular diagnoses, which are predicted to occur in 14–26% of probands (21, 28), are under-reported due to parsimony-based analytical pipelines and cognitive bias (27). Cataloging of the phenotype associated with each variant by submission to repositories like ClinVar and by publication of case and cohort reports facilitates the phenotypic expansion of each genetic disease. The documentation of the clinical variant interpretation within databases like ClinVar complements the ACMG classifications and builds distributed cognition.

This stage pertains to steps 25–31 of Figure 2. Regardless of the outcome of genomic testing, collaboration between the referring provider, the CGS, and the patient ensures the best care plan and follow-through on management recommendations. Patients and families diagnosed by genetic testing commonly report uncertainty about the meaning of a genetic diagnosis (29) despite being informed by the CGS of the diagnosis and its molecular etiology, clinical features, and implications (e.g., recurrence risk). To assist the patient’s adjustment to the diagnosis, the CGS often coordinates the implementation of medical care by other care providers, facilitates genetic counseling, and potentially pursues cascade testing in family members. When specialty clinics are unavailable, the PCP often coordinates the long-term support and management (30–35).

When interpreting negative results, the CGS considers the patient’s prior probability of genetic disease (stage 1) and the context of testing (including the perceived utility by the patient, stage 2) to determine whether the result represents a true or a false negative (Supplementary material section 4B). For instance, a negative genetic test result might increase the likelihood of an alternate non-genetic diagnosis and is, therefore, informative for diagnosis, management, and recurrence risk. Given that the current diagnostic yield of WES is only 35–55% among patients with traits suggestive of a genetic etiology (36–38), the CGS might remain suspicious of an underlying genetic etiology and suspect a false negative result. False negatives can occur due to limitations in knowledge or in technology.

For individuals suspected to have a genetic disorder but who do not have a molecular diagnosis, genetic counseling assists the patient/family’s adjustment to the absence of an objective explanation (6). Additionally, their awareness that the additive and iterative capture of phenotype evolution (including resolved features) promotes constructive engagement in the diagnostic odyssey. Expansion of medical knowledge also promotes diagnostic resolution such that after 1 year, new genomic locus-disease or phenotype-disease associations increase exome diagnostic yield for Mendelian disorders by approximately 10% (39–42). Concurrently or alternatively, the patient characteristics might be sufficiently compelling that the CGS offers additional clinical testing or research-based investigations. Examples of such clinical testing include whole genome sequencing or testing of another tissue to identify a mosaic variant. Examples of research-based investigations include long-read sequencing, de novo genome assembly, methylome analysis or transcriptome analysis (18). Also, exchange of suspicious variants (PhenomeCentral¹, GeneMatcher², or MatchMaker exchange³) facilitates novel disease gene discovery through generation of cohorts of patients with similar phenotypes and variants.

Mapping the genomic medicine process across the PMGP of British Columbia identified processes that limited genomics-informed healthcare in each of the four stages of genomic care (Table 2). These limitations to genomic medicine practice are present globally to varying degrees and with cultural nuances. An example of a culturally nuanced limitation is the differences among payers (43).

Because genomic medicine relies on an evolving collaboration of patients, PCPs, CGSs, LGSs, and payers, its practice necessitates distributed cognition (i.e., knowledge embedded in objects, individuals, and tools) and integrative technology that enables facile data capture, sharing, management, and analysis. As a quality improvement initiative, the PMGP of British Columbia prototyped an electronic platform for genomic medicine to identify key functions facilitating genomic medicine practice. The customized REDCap electronic data management platform facilitated CGS-patient communication, provided secure data (phenotype, genotype, family history) storage, enabled triage through conditional logic, generated standardized reports, improved data management and querying, and automated data transfer. Replacing multiple ad hoc methods of data collection (i.e., paper and Excel spreadsheets), the electronic platform introduced workflow efficiencies. Furthermore, a patient-facing module removed barriers and allowed secure sharing of medical information, photographs, and consent forms. Using customized, automated emails, the platform reduced clerical tasks and achieved a completion rate of 89% for the pre-appointment questionnaire and of 77% for the release-of-information consent form. To standardize funding applications for genomic testing, eligibility and phenotyping instruments captured and communicated pertinent patient details to the payer. This expedited funding approval and access to testing by several weeks. In addition, the capture of information (phenotype, testing eligibility, genomic sequencing results, and outcomes) across a patient’s journey identifies patient cohorts with unique care needs, promotes sharing of cognitive resources in the interpretation of variants, refines testing criteria to maximize utility, and facilitates assessment of quality improvement. The platform is a prototype that will require further development and adaptation to meet the needs of each health system.