Pharmacogenomics: Clinical Implementation
A practical guide to implementing pharmacogenomics in health systems — pre-emptive panel testing, CPIC guideline integration into EHR clinical decision support, phenotype translation, the highest-priority gene-drug pairs, implementation challenges, and the evidence base for health system PGx programmes.
Pre-emptive vs reactive PGx testing
Pharmacogenomic testing strategies fall into two models with distinct workflows, economics, and clinical impact:
Reactive (on-demand) testing
Genotyping is ordered when a clinician prescribes a drug with a known gene-drug interaction. Most common current practice. Limitations: test results take 1–14 days (lab turnaround), causing prescribing delays or empirical prescribing while awaiting results; genotype data generated is not available for future drug decisions; creates workflow friction. Reactive testing is appropriate for high-cost one-time tests (e.g. HLA-B*57:01 before abacavir in HIV — test is ordered only once).
Pre-emptive panel testing
Genotyping a patient for 10–20+ clinically actionable gene-drug pairs in a single panel test, before any specific drug is prescribed — typically at health system enrolment, primary care, or perioperative assessment. Results stored in the EHR as permanent, searchable discrete data. Cost per actionable result is lower than reactive testing when panel cost is distributed across all future drug prescribing events. The CPIC ↗ and PharmGKB ↗ Level 1A/1B gene-drug pairs are the pre-emptive panel targets.
Major pre-emptive PGx programmes demonstrating feasibility and clinical utility: RIGHT Protocol (Mayo Clinic), PREDICT (Vanderbilt), 1200 Patients Project (University of Florida / IGNITE Network), Ubiquitous Pharmacogenomics (U-PGx) PREPARE trial (multicenter EU RCT — demonstrated 30% reduction in clinically relevant adverse drug reactions in patients receiving PGx-guided prescribing vs standard care, Lancet 2023).
CPIC guidelines: translating genotype to prescribing action
CPIC guidelines differ from standard drug prescribing guidelines in one critical way: they assume the clinician already has the genotype result and provide drug-specific recommendations based on that result. They do not address whether to test — only what to do when a result is known.
| Gene | Phenotype | Affected drugs | CPIC recommendation |
|---|---|---|---|
| CYP2D6 | Poor metaboliser (PM) | Codeine, tramadol | Avoid — no analgesia (prodrug not activated); risk of accumulation if URM |
| CYP2D6 | Poor metaboliser (PM) | Tricyclic antidepressants (nortriptyline, amitriptyline) | Use 50% of normal starting dose; consider alternative |
| CYP2D6 | Ultrarapid metaboliser (URM) | Codeine | Avoid — excessive morphine conversion; deaths reported (especially children) |
| CYP2C19 | Poor metaboliser (PM) | Clopidogrel | Use alternative antiplatelet (prasugrel, ticagrelor) — reduced active metabolite, higher MACE risk |
| CYP2C19 | Rapid/Ultrarapid metaboliser | SSRIs (citalopram, escitalopram, sertraline) | May need higher doses; consider pharmacodynamic response monitoring |
| DPYD | Poor/Intermediate metaboliser | 5-FU, capecitabine, tegafur | Reduce dose by 25–50% (IM) or avoid (PM); mandatory pre-treatment testing in EU |
| TPMT/NUDT15 | Poor metaboliser | Azathioprine, mercaptopurine, thioguanine | Reduce dose by 50–90% or use alternative — severe myelosuppression risk |
| HLA-B*57:01 | Allele present | Abacavir | Contraindicated — hypersensitivity reaction (now standard pre-treatment screening) |
| HLA-B*15:02 | Allele present | Carbamazepine | Avoid — SJS/TEN risk (Southeast Asian patients; FDA and EMA label requirement) |
| SLCO1B1 (*5 variant) | Decreased function | Simvastatin (high dose) | Use ≤40 mg/day simvastatin or switch to rosuvastatin/pravastatin (lower SLCO1B1 dependence) |
Full, searchable CPIC guidelines with prescribing recommendations, supplementary data, and CDS resources are freely available at cpicpgx.org ↗. See our pharmacogenomics basics guide for metaboliser phenotype definitions and star allele nomenclature.
EHR integration: from genotype to CDS alert
Implementing PGx CDS in an EHR requires four technical layers:
- Discrete data storage: Genotype results must be stored as discrete coded data elements — not as PDF lab reports or free text. The HL7 FHIR Genomics specification ↗ and LOINC codes for genetic variants provide the standardised representation. Star alleles (e.g. CYP2D6 *1, *4, *5) must be mapped to CPIC diplotype tables.
- Phenotype translation: Diplotype (pair of haplotypes) → phenotype (Poor/Intermediate/Normal/Rapid/Ultrarapid Metaboliser) translation must be standardised. CPIC provides translation tables for each gene. EHR systems or middleware (Genomic Prescribing System, G2MC) perform this translation and store the phenotype as a separate discrete field.
- CDS rule authoring: Phenotype-triggered alerts are authored using the EHR's clinical rule engine (Epic's BPA framework, Cerner CareAware CDS, or HL7 CDS Hooks for vendor-agnostic deployment). Rules fire when a pharmacist or prescriber orders a drug with a CPIC Level A recommendation for the patient's stored phenotype.
- Alert design: Effective PGx alerts include: phenotype displayed (e.g. "CYP2D6 Poor Metaboliser"), specific recommendation (alternative drug or dose), evidence grade (CPIC Level A), link to full CPIC guideline, and an easy acknowledgement pathway for clinicians who proceed with a justified rationale. Alert fatigue from non-actionable or low-evidence alerts is the primary barrier to PGx CDS adoption — see our ASHP Informatics Summit 2027 for CDS governance approaches.
The PharmGKB clinical annotations ↗ complement CPIC by providing Level 1A (highest evidence), 1B, 2A, 2B evidence ratings for gene-drug associations beyond those with full CPIC guidelines.
Priority gene-drug pairs for implementation
Not all CPIC guidelines are equal in clinical urgency. Priority for health system implementation is based on: clinical consequence severity, prescribing frequency, test turnaround relevance, and label requirements. The highest-priority pairs are:
DPYD / Fluoropyrimidines
EMA mandates DPYD testing before 5-FU and capecitabine. DPYD poor metabolisers have life-threatening early-onset toxicity (mucositis, myelosuppression, neurotoxicity) at standard doses. 4 DPYD variants routinely tested (c.1905+1G>A, c.1679T>G, c.2846A>T, c.1236G>A/HapB3). See PGx basics.
HLA-B*57:01 / Abacavir
Hypersensitivity reaction (fever, rash, GI symptoms, respiratory) occurring in ~5–8% of HLA-B*57:01 carriers. Potentially fatal on rechallenge. Now standard pre-treatment screening in all HIV treatment guidelines worldwide.
HLA-B*15:02 / Carbamazepine
Stevens-Johnson Syndrome/TEN in Southeast Asian patients. Risk 10× higher in carriers vs non-carriers. FDA and EMA label requirement for testing in at-risk ethnic backgrounds. See our drug hypersensitivity guide.
CYP2C19 / Clopidogrel
CYP2C19 poor metabolisers have reduced active metabolite formation, higher platelet reactivity, and significantly higher MACE (major adverse cardiovascular events) post-PCI compared to normal metabolisers. CPIC: use prasugrel or ticagrelor as alternatives in PM patients. Point-of-care CYP2C19 genotyping available.
TPMT / NUDT15 / Thiopurines
Poor metabolisers at high risk of severe myelosuppression from azathioprine and mercaptopurine. TPMT phenotyping (enzyme activity) or genotyping both acceptable. NUDT15*3 variant particularly prevalent in East Asian patients. Dose reduction of 50–90% required. See PGx guide.
CYP2D6 / Codeine, Tramadol
CYP2D6 poor metabolisers: no analgesic effect from codeine (prodrug). URM: excess morphine → respiratory depression and death (FDA Black Box Warning in children). Tramadol: same pathway. See opioid safety guide.
DPYD spotlight: mandatory pre-treatment testing for fluoropyrimidines
DPYD (dihydropyrimidine dehydrogenase) is the rate-limiting enzyme in the catabolism of fluoropyrimidine chemotherapy agents — 5-fluorouracil (5-FU), capecitabine (prodrug of 5-FU), and tegafur. DPYD activity accounts for ~80–85% of 5-FU elimination. DPYD poor metabolisers accumulate toxic 5-FU concentrations, resulting in early-onset, severe, potentially fatal toxicity:
- Grade 3–4 mucositis (oral and GI)
- Severe myelosuppression (neutropenia, thrombocytopenia)
- Hand-foot syndrome
- Neurotoxicity (cerebellar — rare)
- Cardiotoxicity (coronary vasospasm — 5-FU specific)
The EMA mandated pre-treatment DPYD genotyping for 5-FU and capecitabine across the EU from 2020 (EMA label update). The four DPYD variants with the highest evidence (CPIC/DPWG Level A) are:
- c.1905+1G>A (IVS14+1G>A, DPYD*2A) — splice site variant; most common high-impact variant in European populations; associated with complete DPYD deficiency
- c.1679T>G (DPYD*13) — missense (I560S); severe activity reduction
- c.2846A>T (D949V) — missense; partial activity loss
- c.1236G>A/HapB3 — haplotype marker; intermediate activity loss
Dose adjustment: DPYD intermediate metabolisers (one functional allele) receive 25–50% dose reduction; poor metabolisers (no functional alleles) should use alternative chemotherapy regimens. DPYD activity score (0, 1, 1.5, 2) from CPIC guides dose selection. See the CPIC DPYD guideline ↗.
Implementation challenges
Despite strong clinical evidence, health system PGx implementation faces real barriers:
Result recontact
When a new CPIC guideline is published after a patient was already genotyped, health systems face the challenge of identifying and notifying affected patients. Most current systems do not have automated recontact workflows — this is an active area of informatics development.
Equity and access
Pre-emptive PGx panels are more cost-effective in patients prescribed many medications over time — often those with chronic disease. However, access to genotyping is unequal across socioeconomic groups and healthcare settings, and most PGx research cohorts are predominantly European-ancestry.
Clinician education
Surveys consistently show that prescribers have limited knowledge of PGx — they may not interpret a CDS alert correctly or understand the phenotype terminology. Point-of-care resources, consult pharmacist support, and actionable alert messaging are key to bridging the knowledge gap.
EHR interoperability
Genotype results generated in one health system are often unavailable when a patient transfers to a different institution. FHIR Genomics and the Genome Medical Alliance's data portability initiatives address this, but interoperable PGx data exchange is not yet universal.
Insurance coverage (US)
Coverage for pre-emptive PGx panel testing is inconsistent across US payers. Medicare covers a multi-gene panel (MolDX LCD) only for specific indications. Some Medicaid programs and commercial insurers cover DPYD and HLA-B testing mandatorily. Pre-emptive panel coverage remains limited outside of health system-funded research programmes.
Variants of uncertain significance
Most clinical PGx panels focus on well-characterised variants. As sequencing expands (WGS-based PGx), more variants of uncertain significance (VUS) will be identified. Communicating VUS appropriately without causing unnecessary prescribing restriction or patient anxiety is a major challenge for clinical informatics.
FAQ
- What is pre-emptive pharmacogenomic testing?
- Panel genotyping of 10–20+ clinically actionable gene-drug pairs before any specific drug is prescribed, so results are available in the EHR for future prescribing decisions. More cost-effective per actionable result than reactive (on-demand) testing. Demonstrated in RIGHT Protocol (Mayo Clinic), PREDICT (Vanderbilt), and the U-PGx PREPARE RCT (30% reduction in adverse drug reactions, Lancet 2023).
- What are CPIC guidelines and how do they support PGx implementation?
- CPIC (Clinical Pharmacogenomics Implementation Consortium) guidelines provide genotype-specific prescribing recommendations for over 40 gene-drug pairs. They assume the genotype is already known and advise what to do with it — not whether to test. Freely available at cpicpgx.org. Evidence graded A (high) to D (low); recommendation type: strong, moderate, optional.
- How are pharmacogenomic results integrated into EHR clinical decision support?
- Four layers: (1) discrete genotype storage (FHIR Genomics + LOINC codes); (2) phenotype translation (diplotype → metaboliser phenotype via CPIC tables); (3) CDS rule authoring (phenotype-triggered alert fires on drug order); (4) alert design (clear recommendation, CPIC evidence grade, avoid alert fatigue). HL7 CDS Hooks enables EHR-vendor-agnostic deployment.
- Which gene-drug pairs have the highest clinical priority for implementation?
- Mandatory/urgent: DPYD/fluoropyrimidines (EMA mandated), HLA-B*57:01/abacavir (standard of care). High priority: HLA-B*15:02/carbamazepine, CYP2C19/clopidogrel (post-PCI), TPMT/NUDT15/thiopurines. Moderate: CYP2D6/codeine and tramadol. All prioritised by clinical consequence severity, prescribing frequency, and available alternatives.