Pharmacogenomics-based drug selection

A PGx recommendation is usable only when it cites two things: the genotype the lab actually reported, and the published drug-gene row that maps that genotype to an agent and dose. If either cite is missing, the recommendation stays empty. The workflow does not infer a metabolizer phenotype from race, does not fill a missing star allele from a family history, and does not write a labeled dose as if the gene were normal.

Quality is citation completeness. "Consider an alternative" with no diplotype and no table row is not a PGx recommendation. "CYP2C19 *2/*2 on the lab report matches the published clopidogrel poor-metabolizer row; avoid clopidogrel and use a listed alternative" is.

The prescriber still chooses the drug and the dose. Pharmacist work is to confirm both cites are present, refer to the same gene and drug, and remain true for the order under review. Review is not conversion of the recommendation into a signed order.

Load genotype as a structured lab result. You need the gene, the named alleles or diplotype as the lab reported them, the specimen or report date, and the reporting laboratory. A checkbox that says PGx was performed, a scanned PDF with no parsed alleles, or a consult sentence that says "genetics done" is not a genotype for this workflow. Empty stays empty.

Load the PGx table as the mapping your site is allowed to use: gene, drug or drug class, phenotype or activity score, and the recommended action (use as labeled, adjust dose, avoid, or switch to a listed alternative). That table may sit in a lab portal, a drug-knowledge base, or an order-entry alert. Tempus, Epic, FDB, and Oracle Health belong to the same class of systems that can carry genotype, drug-gene rows, or both. Use the feed your organization has validated. Do not rank those vendors and do not assume a feature you have not seen in your build.

Match on gene and drug first, then on the phenotype or diplotype the lab reported. Do not collapse "likely intermediate" into "intermediate" unless the published row you are citing uses that same label. If the table has no row for this pair, or the lab did not report this gene, stop. Leave agent and dose blank.

When both inputs match, write the recommendation as two citations plus an action: the lab diplotype (gene, alleles, report identifier), the table row (source and version your P&T committee adopted, gene, drug, phenotype, action), then the suggested agent and dose taken from that row. Do not add a third clinical impression that changes the action. If they do not match, write nothing in the recommendation field. The prescriber still chooses whether to follow it.

When the current order is for a drug with no gene row in your table, this workflow is silent on purpose. Use real-time drug interaction monitoring for kinetic conflicts that are not genotype-based. Use medication reconciliation agent so you interpret genotype against the home list, not only the new order.

A cardiology discharge includes clopidogrel 75 mg daily. The chart already holds a CYP2C19 lab report of *2/*2. The site-adopted PGx table includes a clopidogrel row for CYP2C19 poor metabolizer: avoid clopidogrel and use a listed alternative P2Y12 inhibitor at its labeled dose.

The recommendation can state: CYP2C19 *2/*2 (named lab report) matches the published clopidogrel/CYP2C19 poor-metabolizer row; recommended action is to avoid clopidogrel and select a listed alternative. The prescriber accepts, changes, or declines. You do not rewrite the order from the recommendation pane.

If the same discharge has no CYP2C19 result, you do not recommend 75 mg as a PGx dose, you do not recommend an alternative, and you do not write "assume normal metabolizer." The PGx fields stay empty. Clopidogrel may still be ordered on clinical grounds. That is a prescribing decision, not a filled-in genotype.

This example is only a pattern for the two cites that must appear together, and for the blank that must remain when one cite is missing. It is not a measured case and it does not imply a protocol, a catch rate, or a time saving.

Writing a dose when genotype is absent is a failure of this workflow. Putting the labeled dose into the PGx recommendation field "until results return" mixes a holding pattern with a gene-based choice. If the team needs a holding plan, document it in the progress note or in the ordinary dosing comment. Keep the PGx recommendation empty.

Inventing a star allele is the second failure. If the lab reported CYP2C19 *1/*17, you do not upgrade it to *2/*2 because poor metabolizer would change therapy. If the lab reported only "CYP2C19 variant detected" without named alleles, you do not pick *2/*2 to unlock the table row. If ancestry or a relative's result is in the chart, that is not this patient's diplotype.

Treating the recommendation as the order is the third. Dual cites do not sign the medication. Routing an unsigned recommendation into order-entry as a clopidogrel stop or an alternative start, without prescriber action, is an error path even when the cites are correct. The pharmacist can call, document the two cites, and hold a verify if local policy says so. The prescriber still chooses.

Do not use treatment response prediction as a stand-in for a missing genotype. A predicted response is a different input. It does not create a star allele and it does not satisfy the published PGx row.

PGx selection does not replace indication, renal dosing, interaction review, or duplicate-therapy checks. After a dual-cite recommendation appears, still confirm the alternative is on formulary, is not contraindicated, and does not collide with the rest of the list.

When the published row is a pointer to a guideline clause (dose reduction, alternative agent, or "no recommendation"), retrieve that clause instead of paraphrasing from memory. evidence-based guideline retrieval is the companion when the table row is not the full monograph.

If genotype was drawn for one indication and the new order is a different substrate of the same gene, re-match the table. A CYP2D6 result used for one antidepressant does not automatically authorize a dose for a new opioid without a published row for that pair.

The pharmacist's job at the point of prescribing is narrow: require both cites, refuse invented alleles, keep blanks blank, and leave the order to the prescriber.