Evidence fusion in pharmacogenomics (PGx) is the process of merging guidelines, drug labels, curated databases, and literature into one graded, reviewable recommendation for each gene–drug pair in a report. Instead of an analyst opening CPIC, DPWG, PharmGKB, and FDA pages one at a time, the reporting system matches the patient's genotype and medication list to every applicable source, tags each statement by origin, reconciles conflicts, and presents a single result that your medical director can approve. This article explains how that works, where it fails when done by hand, and what a defensible implementation looks like in a US clinical laboratory.
TL;DR
- Evidence fusion combines guidelines (CPIC, DPWG), FDA labeling, and curated knowledgebases such as PharmGKB into one source-tagged, graded recommendation per gene–drug pair.
- Each source grades evidence differently, so a fused report must keep the original grade and source visible rather than flattening everything into one score.
- Genotype alone is not the answer: phenotype, concomitant medications, and patient factors change what the recommendation means.
- Evidence changes constantly, so fusion without versioning and reanalysis becomes outdated; a medical director's sign-out stays the final safeguard.
- SignalPGx builds this workflow into white-label reporting infrastructure for laboratories.
Table of Contents
- What evidence fusion means in a PGx report
- Which sources a PGx interpretation should combine
- How evidence strength is assessed
- From genotype to recommendation: where the reconciliation happens
- What manual evidence management costs a laboratory
- What a defensible fusion workflow looks like
- Our perspective: fusion is a clinical control, not a convenience feature
- Putting fusion into production with SignalPGx
- FAQ
- Sources
What evidence fusion means in a PGx report
A PGx report answers a narrow question: given this genotype and this medication list, what should the prescriber consider? The evidence behind that answer is scattered. Clinical guidelines translate genotype to phenotype and phenotype to a therapeutic recommendation. Drug labels carry regulatory language. Knowledgebases score the underlying studies. Each has its own format, update cycle, and vocabulary.
Fusion is the engineering and clinical discipline of bringing these into one structure. Open-source tooling shows the basic mechanics well: PharmCAT matches genotyping results to applicable CPIC and DPWG recommendations, tags them by source, groups them by drug, and converts annotations into a common format. That is the minimum: normalize, attribute, and group. A production laboratory report goes further by adding evidence grading, medication context, and reviewer approval.
Two features separate fusion from simple aggregation:
- Attribution. Every statement in the final report can be traced to the guideline, label, or database record it came from, including the version.
- Reconciliation. When sources disagree, or when one is silent, the system applies a documented rule and surfaces the disagreement to a human reviewer instead of hiding it.
Which sources a PGx interpretation should combine
No single source covers every gene–drug pair a laboratory reports. A practical fusion layer draws on four families.
| Source family | What it contributes | Typical limitation |
|---|---|---|
| Clinical guidelines (CPIC, DPWG) | Genotype-to-phenotype tables and therapeutic recommendations with a stated strength | Only covers drugs and genes with a published guideline; updates arrive on each group's own schedule |
| FDA labeling and the FDA pharmacogenetic associations table | Regulatory language on affected subgroups and gene–drug interactions | Wording varies from actionable dosing to informational mention |
| Curated knowledgebases (PharmGKB/ClinPGx) | Scored clinical annotations that aggregate variant-level studies | Scores describe the strength of an association, not a prescribing instruction |
| Primary literature and variant resources (for example ClinVar) | Newer findings and variant-level context | Needs expert appraisal before it influences a recommendation |
The FDA table is worth singling out because it is easy to overlook. It is a published list of genetic variants that can affect drug concentrations, therapeutic response, or adverse events, with examples such as HLA-B*57:01 and abacavir, where the label advises against use in positive patients. A laboratory that reports only guideline content may miss label-level information that a prescriber or auditor expects to see.
The Association for Molecular Pathology position statement states that laboratories should ensure health-related PGx claims have clinical validity supported by robust evidence from peer-reviewed literature, FDA-approved drug labels, or established clinical practice guidelines. That sentence is effectively the scope definition for any fusion engine: if a claim cannot be traced to one of those three categories, it does not belong in the report.
Internationally, the picture is less settled. Where no national PGx regulation exists, institutions must compare available global resources to decide which guidance fits their local context. For US laboratories that usually means CPIC and FDA labeling as the anchor, with DPWG as a second reference where recommendations differ.
How evidence strength is assessed
Fusing sources is only useful if the strength of each is preserved. Three grading systems matter most, and they measure different things.
PharmGKB levels of evidence. PharmGKB's quantitative scoring uses published variant annotations plus qualifying guideline and drug-label annotations, and weighs study attributes such as phenotype category, P value, cohort size, effect size, and study type. Levels run from 1A to 4, and Level 4 means the evidence does not support an association. This tells you how well established the association is.
CPIC recommendation strength. CPIC grades therapeutic recommendations as strong, moderate, or optional, drawing on preclinical functional and clinical data, existing consensus guidelines, and other relevant evidence. This tells you how firmly a group of experts recommends acting on the association. The same authors note that randomized trials comparing genotype-guided dosing with conventional dosing are rare for most drug–gene pairs, which is why CPIC weighs a mix of evidence types.
FDA labeling language. The label states what the regulator has approved, which may be stronger, weaker, or simply different in scope from a guideline.
A strong association (PharmGKB 1A) does not automatically produce a strong recommendation, and a strong recommendation does not mean the label says the same thing. If a fusion engine collapses these into one number, a reviewer loses exactly the information needed to judge a conflict. We recommend displaying the grade from each source side by side and reserving any composite label for internal triage. For a deeper walk-through of the factors involved, see our post on PGx evidence grading, and for mapping CPIC, PharmGKB, and FDA content to a defensible report, see evidence-graded PGx.
From genotype to recommendation: where the reconciliation happens
Fusion starts after the genotype is called, but it does not end with a lookup. The path looks like this:
- Call and normalize the genotype. Variants are extracted from the VCF or assay output and assigned to haplotypes and diplotypes. Tools such as PharmCAT extract variants, infer haplotypes and diplotypes, and map them to curated CPIC recommendations. Genes with structural complexity, such as CYP2D6, need extra care; our CYP2D6 copy number guide covers that problem for US labs.
- Translate to phenotype. CPIC publishes tables translating genotype to phenotype and phenotype to therapeutic recommendation, which gives you a consistent starting vocabulary.
- Apply medication context. The same review explains that results are interpreted alongside age, renal and liver function, and concomitant medications. A strong CYP2D6 inhibitor such as fluoxetine can make the genotype result irrelevant because the interaction may leave limited to no CYP2D6 activity. This is the step where medication history is not optional.
- Match every applicable source. Guidelines, labels, and annotations are matched to each drug on the patient's list, tagged by origin, and grouped.
- Flag conflicts and gaps. Where CPIC and DPWG diverge, or where a label says something a guideline does not, the system flags the pair for review.
- Clinical review and sign-out. A qualified reviewer approves, edits, or rejects the output.
The last step is not decoration. The same CPIC-based guidance reminds clinicians to use judgment and to consider whether a stable, effective medication should be changed at all. A fused report should support that judgment, not replace it. Examples such as codeine and CYP2D6, CYP3A5 and tacrolimus, and clopidogrel in CYP2C19 poor metabolizers show how each gene–drug pair carries its own nuances that a reviewer must be able to see.
What manual evidence management costs a laboratory
Most laboratories begin with a spreadsheet or a set of internal documents: one row per gene–drug pair, with text copied from a guideline and a note of when someone last checked it. This works at a handful of genes and fails as menus grow, for predictable reasons.
- Version drift. A guideline is updated, a label changes, or a knowledgebase re-scores an annotation, and nobody notices for months. Reports issued in that window cite superseded content.
- Inconsistent wording. Two analysts summarize the same CPIC recommendation differently, and the lab's own reports disagree.
- Slow sign-out. Reviewers spend their time re-verifying sources instead of exercising clinical judgment.
- Weak audit trail. When a prescriber or an inspector asks why a report said what it said on a given date, the answer lives in somebody's memory.
The literature is blunt about the maintenance burden. Maintenance of PGx knowledgebases and clinical decision support tools is essential as the literature evolves, and it requires specialized PGx expertise to prevent alert fatigue and preserve clinical effectiveness. That same review recommends evaluating PGx alerts regularly. If you maintain a checklist for guideline changes today, our CPIC implementation checklist is a practical starting point.
There is also a skills gap on the receiving end. A UK professional body's guidance observes that interpreting pharmacogenomic information remains a specialized skill beyond the training of most health professionals, which makes clear, lab-to-clinic communication critical. The same document notes that PGx information can reach clinicians as a formal report or as rules-based decision support that alerts at the point of care. A fused, source-tagged evidence base can feed both.
What a defensible fusion workflow looks like
If you are evaluating whether to build or buy, these are the properties worth testing against any approach.
Source transparency
Every recommendation in the report should show which sources contributed, at which version, and with which grade. A reader should never have to trust an unlabeled summary.
Documented conflict rules
The rule for choosing between CPIC and DPWG, or between a guideline and a label, should be written down, approved by your medical director, and applied consistently. Disagreements should reach a reviewer rather than being resolved silently.
Validation framework
The CDC's ACCE model (analytical validity, clinical validity, clinical utility, and ethical, regulatory, and social implications) is a framework for evaluating and integrating genomic tests into practice. The same logic applies to the software layer that interprets results; for a broader look at how clinical utility is weighed for a single genetic test, see this discussion of MTHFR testing and clinical utility. Under CLIA, laboratories must also document analytical validity, maintain a quality management system, and have tests interpreted by board-certified molecular laboratory professionals. Labs pursuing accreditation can pair this with our ISO 15189 guide for genomics labs.
Living reanalysis
A report is a snapshot. A fusion system should record the evidence version behind each report and be able to tell you which previously issued reports are affected when a source changes. Then your team decides, with clinical input, what to do about them.
Human sign-out and audit trail
Automation should prepare the evidence and your medical director should approve it. Professional guidance describes the pharmacist's judgment as the final safety and quality-control step before PGx information changes a medication regimen; the laboratory's own sign-out plays the same role upstream. Each approval, edit, and rejection should be logged.
Report content and delivery
At minimum the AMP statement says reports should include metabolizer status, affected drugs, and a generalized statement alerting providers to consider dosage adjustments or treatment alternatives. Beyond that, delivery into the clinician's own system matters; our guide to pharmacogenomics in EHRs recommends mapping one gene–drug pair in FHIR first before scaling.
Our perspective: fusion is a clinical control, not a convenience feature
We see laboratories treat evidence fusion as a time-saving feature. We think that undersells it. Done well, it is the control that makes a PGx report defensible: the reason your reports say the same thing on Tuesday as they did on Monday, and the reason you can explain any difference.
That is how we built SignalPGx. Our platform is designed to bring genotype data, medication histories, and clinical evidence from more than 15 authoritative sources into one physician-reviewed report, with each statement traceable and each report auditable. Living reanalysis is designed to alert your team as guidelines evolve, so version drift becomes a managed workflow rather than a surprise. Medication intelligence simulation lets your reviewers explore how a medication list changes the picture before sign-out. And the final approval always stays with your medical director, because we believe automation earns trust by keeping clinicians in control, not by sidelining them.
White-labeling is more than a logo in the corner. The wording, the evidence tags, the review workflow, and the delivery path should all reflect your laboratory's standards, and your lab's name is on the result. We built the infrastructure so you can deploy under your own brand, typically within 5 to 7 days, and integrate with EHRs through HL7/FHIR, while keeping HIPAA and GDPR compliance in view.
— Tarek
Putting fusion into production with SignalPGx
If your team is maintaining guideline summaries by hand, or planning to launch a PGx service and wants the evidence layer done right from the start, we can scope it with you. Our White-Label PGx Reporting service is offered in Pilot, Standard, Scale, and Enterprise plans, and our project services cover discovery, design, and deployment: Discovery & Strategy from $7,000, System Design from $9,000, and Deployment & Integration from $15,000, each one-off per project. Plan pricing is not published, so request details on the White-Label PGx Reporting page.
Billing is part of the business case too; our pharmacogenomics reimbursement guide outlines what lab billing teams should plan for.
FAQ
Is evidence fusion the same as a PGx database?
No. A database stores records; fusion is the process of matching a specific patient's genotype and medications to the relevant records from several sources, reconciling them, and producing a graded, source-tagged recommendation. The database is one input, and the reviewed report is the output.
What happens when CPIC and DPWG disagree?
The system should flag the pair rather than choose silently. Your laboratory then applies a documented rule, approved by your medical director, and the report states which source it follows. Where no national regulation settles it, institutions are expected to compare global resources and decide what suits their context.
Does automation remove the need for a medical director's review?
No. Automation handles retrieval, matching, and formatting, which are the repetitive parts. Clinical judgment, such as whether a stable, effective therapy should change, stays with qualified reviewers, and a CLIA laboratory still needs board-certified professionals interpreting results.
How often should fused evidence be refreshed?
Continuously in practice, because guidelines, labels, and knowledgebase scores change on their own schedules. The practical approach is to version every evidence record, alert reviewers when a source changes, and identify which issued reports it touches.
Sources
- Matching Recommendations, PharmCAT
- Pharmacogenetic Gene-Drug Associations: FDA Perspective on What Physicians Need to Know, AAFP
- Association for Molecular Pathology Position Statement: Best Practices for PGx (2019)
- Pharmacogenomics in practice: a review and implementation guide
- An Evidence-Based Framework for Evaluating Pharmacogenomics Knowledge
- Evidence and resources to implement Pharmacogenetic knowledge
- Pharmacogenomics Clinical Annotation Tool (PharmCAT)
- Decoding Pharmacogenomic Test Interpretation
- Pharmacogenomic Clinical Decision Support: A Review, How-to Guide
- Guidance for using pharmacogenomic information in clinical practice, PRSB
- 3.5 Laboratory Report Interpretation, Council on Pharmacy Standards