PGx evidence grading is the set of curated frameworks and numeric scoring systems, PharmGKB/ClinPGx, CPIC, DPWG, FDA labels, and payer policies, used to judge whether a gene-drug association should change prescribing. In practice, we treat CPIC Level A/B recommendations and PharmGKB/ClinPGx Level 1A/1B annotations as the actionable tier. Everything scored lower belongs in the research column until stronger data arrives.
TL;DR:
- PharmGKB/ClinPGx 1A and 1B annotations support action flags; 2A and lower belong in monitoring or research, with 2A tied to Tier 1 genes.
- CPIC A or B supports a prescribing action, but recommendation strength is separate; DPWG’s potentially beneficial tier may justify preemptive testing without changing a prescription.
- A 2015 review found clinical validity for 46% of 166 FDA drug biomarker pairs, clinical utility for 29%, and evidence for both in only 27%.
- CMS MolDX coverage requires demonstrated analytic validity, clinical validity, and clinical utility; combinatorial panels face added scrutiny when each gene drug link lacks independent validation.
Table of Contents
- How the major PGx evidence frameworks divide the work
- PharmGKB / ClinPGx scoring: how annotation scores map to LOE
- CPIC vs DPWG: turning evidence into a prescribing action
- Why evidence grading sits downstream of assay validation
- Turning graded evidence into clinician-facing report fields
- Getting evidence grading governance right
- SignalPGx: lab-focused PGx reporting and evidence fusion
- FAQ
- Sources
How the major PGx evidence frameworks divide the work
Every framework answers a different question, and confusing them is where most implementation delays start. PharmGKB/ClinPGx curates and scores the underlying association evidence into a level of evidence (LOE). CPIC takes that evidence and turns it into a specific prescribing recommendation, graded separately by strength. DPWG runs a parallel clinical-implication process, often weighted toward preemptive genotyping programs common across Europe. FDA labels flag a biomarker or pharmacogenomic relationship without always specifying a dosing action. Payer policies, including CMS MolDX, sit downstream of all of them and decide what gets reimbursed.
- PharmGKB/ClinPGx answers: how strong is the evidence behind this specific gene-drug association?
- CPIC answers: given sufficient evidence, what should a clinician actually do with the genotype?
- DPWG answers: what is the clinical consequence, and does it justify preemptive testing?
- FDA labels answer: does regulatory language exist, and is it descriptive or prescriptive?
- Payer policies answer: will this test and its downstream action be covered?
Knowing which question you're asking determines which source you open first, and labs that blend these outputs without labeling the source create reports that read as more definitive than the underlying evidence supports.
PharmGKB / ClinPGx scoring: how annotation scores map to LOE
PharmGKB/ClinPGx assigns a numeric score to each variant annotation using six weighted attributes: phenotype category, p-value, cohort size, effect size, study type, and association significance. The summed score determines the level of evidence, and the scoring method documented in a 2021 evaluation framework lays out the exact bands curators apply.
| LOE | Score range | What it requires |
|---|---|---|
| 1A | high | Guideline or FDA label with variant-specific prescribing guidance, plus at least one supporting publication |
| 1B | moderate | High evidence strength but no formal prescribing guidance |
| 2A | low | Moderate evidence in a Tier 1 Very Important Pharmacogene (VIP) |
| 2B | low | Moderate evidence outside a Tier 1 VIP gene |
| 3 | very low | Low evidence, typically a single modest study |
| 4 | none | Association not supported by available data |
Rare variants are scored against a separate, lower threshold because small cohort sizes mechanically limit statistical power, and PharmGKB's clinical annotation update confirms that curators can manually override a computed score when clinical judgment or newer literature warrants it, as long as the override carries written justification. For lab reporting purposes, this matters: a 2B association in a non-VIP gene and a 2A association in a VIP gene can carry the same point total but different reporting weight.
Pro Tip: Use the 1A/1B boundary as your primary filter for which associations earn a "recommended action" flag in a report; treat 2A and below as "monitor" entries pending further evidence.
CPIC vs DPWG: turning evidence into a prescribing action
CPIC and DPWG both convert graded evidence into clinical guidance, but they use different scales and different endpoints. A comparison of the two systems draws a clear line between evidence strength and recommendation strength, a distinction that is easy to lose in a busy lab report.
- CPIC Level A or B means sufficient evidence exists for at least one prescribing action; Level C or D generally does not support a specific action.
- CPIC recommendation strength (Strong, Moderate, Optional) is graded independently of the evidence level and reflects how confidently clinicians should act on it.
- DPWG uses a clinical-implication framework built around severity of effect and a number-needed-to-genotype calculation, which leans toward justifying preemptive panel testing rather than reactive single-gene action.
- Practical mapping: a CPIC A/Strong pairing should drive an immediate dosing or drug-selection decision, while a DPWG "potentially beneficial" tier may justify adding a gene to a preemptive panel without changing today's prescription.
Neither system should be read as interchangeable with PharmGKB's LOE. A gene-drug pair can hold a 1A annotation while CPIC has not yet issued formal guidance simply because CPIC hasn't reviewed that pair.
Why evidence grading sits downstream of assay validation
Grading evidence means nothing if the underlying genotype call is wrong, which is why analytic validity, clinical validity, and clinical utility function as three separate gates. Analytic validity asks whether the assay accurately and reproducibly detects the genotype. Clinical validity asks whether that genotype reliably predicts the phenotype or drug response. Clinical utility asks whether acting on that result improves a patient outcome. FDA guidance on genotyping systems is explicit that evidence interpretation cannot compensate for a poorly validated assay, which is why grading only ever applies after analytic performance is established.
- Confirm the assay's analytic performance across runs, sites, and operators.
- Establish clinical validity, meaning the genotype-phenotype link is reproducible in the relevant population.
- Only then layer on clinical utility evidence to justify a prescribing action.
A 2015 review cited in PharmGKB's clinical annotation update found convincing or adequate clinical validity evidence in 46% of the 166 FDA label drug-biomarker pairs reviewed, but clinical utility in only 29%, and both validity and utility together in just 27%. The CMS MolDX local coverage determination reflects this gap directly, requiring demonstrated analytic validity, clinical validity, and clinical utility before covering a PGx test, and it applies particular scrutiny to combinatorial algorithms that lack independent validation of each gene-drug component.
Turning graded evidence into clinician-facing report fields
Translating a numeric evidence grade into a usable clinical report is where most of the operational work happens, and it benefits from a repeatable sequence rather than ad hoc judgment calls.
- Confirm the genotype call meets your assay's validated analytic performance.
- Map the variant to its current PharmGKB/ClinPGx LOE.
- Check CPIC, DPWG, and FDA label status for an existing prescribing recommendation.
- Annotate the recommended action alongside its evidence provenance, never as an unsourced instruction.
- Document your reanalysis policy and route the report through medical director sign-off before release.
Each report should carry, at minimum: genotype, inferred phenotype, evidence level, recommended action, links to the source guideline, the date of grading, and a contact point for clinician questions. Our guide to converting genotype calls into phenotype fields walks through how to standardize that mapping across a panel, and a CPIC implementation checklist is useful for labs building out the action-annotation step specifically.
Pro Tip: Keep the EHR-facing field set minimal: phenotype, an actionability flag, and a direct link to the guideline source are usually enough to let clinical decision support trigger without forcing anyone to parse narrative text.
Getting evidence grading governance right
The biggest mistake we see in PGx program design is treating every gene-drug association as equally reportable. Prioritize CPIC A/B actions and PharmGKB/ClinPGx 1A/1B annotations first; everything else is a research note until the evidence base catches up, and saying so plainly protects clinicians from over-acting on thin data.
The harder, less visible work is governance: living reanalysis that catches guideline updates, HL7/FHIR integration that gets the update into the chart instead of a PDF nobody rereads, and a documented audit trail with medical director sign-off on every evidence-based change. Grading frameworks move. Reports that don't move with them become liabilities quietly.
— Tarek
SignalPGx: lab-focused PGx reporting and evidence fusion
We built our system to automate the evidence fusion work described above across multiple authoritative sources, paired with living reanalysis that flags guideline changes and HL7/FHIR integration that pushes updates directly into the chart. 
Our white-label PGx reporting plans let labs deploy branded, evidence-graded reports without building this infrastructure from scratch, and our deployment and integration services are available for teams that want help connecting it to existing lab or EHR workflows. If you're weighing drug-gene interaction coverage as part of this build, our clinical reference guide on drug-gene interactions is a useful companion, and teams managing broader medication risk reviews alongside PGx findings may also find structured interaction-risk checks relevant to that workflow.
This article is general information, not a substitute for advice from a qualified doctor. Consult a qualified healthcare professional about your own circumstances before acting on anything here.
FAQ
How accurate is PGx testing?
Accuracy depends on the assay's analytic validity, which must be established before any evidence grading applies, as FDA genotyping guidance makes clear. A validated assay paired with a CPIC Level A or PharmGKB Level 1A association gives the most reliable basis for a prescribing decision.
Does insurance pay for PGx testing?
Coverage varies by payer and test, but CMS MolDX policy generally requires demonstrated analytic validity, clinical validity, and clinical utility, with closer scrutiny for combinatorial panels lacking independent evidence for each component. Our reimbursement guide for lab billing covers how to document that case for payers.
What does PGx stand for?
PGx is shorthand for pharmacogenomics, the study of how an individual's genetic makeup affects their response to medications. It's used broadly across clinical and laboratory literature as an abbreviation rather than a distinct term.
What does PGx testing show?
PGx testing identifies genetic variants that influence how a patient metabolizes or responds to specific medications, which labs then translate into a phenotype and, where evidence supports it, a prescribing recommendation. Not every variant detected carries an actionable recommendation, which is why evidence grading determines what belongs in the clinical section of a report versus a research note.
Sources
- An Evidence‐Based Framework for Evaluating Pharmacogenomics Knowledge for Personalized Medicine - PMC
- FDA guidance on drug-metabolizing enzyme genotyping systems
- CMS MolDX LCD on Pharmacogenomic Testing (LCD 38394)