What pharmacogenomics can and cannot do

Pharmacogenomics in pain is sold on a strong version of a true premise, and the gap between the true premise and the sales claim is where public money goes to die.

The true premise

People metabolize opioids very differently. Enriched-enrollment randomized-withdrawal trial data suggests roughly 13-fold variability in the dose required, and genetic differences in hepatic metabolism can account for three-fold or more of that on their own. That is not controversial.

What is actually actionable

The Clinical Pharmacogenetics Implementation Consortium — the body whose job this is — makes recommendations for exactly two drugs, and they are about drug selection:

• CYP2D6 poor metabolizers: avoid codeine and tramadol, use an alternative. And explicitly do not escalate the dose instead — “there is insufficient evidence in the literature to recommend a higher dose.”
• CYP2D6 ultrarapid metabolizers: codeine and tramadol “should not be used,” to avoid severe toxicity at label doses.

The clinical stakes behind that are real. FDA contraindicated codeine and tramadol in children under 12 in 2017, citing 64 reports of serious breathing problems with codeine including 24 deaths, some after a single dose.

What is explicitly not actionable

CPIC, verbatim: “There are no therapeutic recommendations for dosing opioids based on either [OPRM1] or [COMT] genotype.” The OPRM1 variant’s effect on postoperative morphine requirement is “so modest as to not be clinically actionable” — a meta-analysis of 14 studies put it at Cohen’s d = 0.096. For COMT there is “no evidence” of association with adverse events and “mixed evidence” for analgesia.

And for the two opioids most relevant to chronic pain, oxycodone and methadone, CPIC gives no recommendation at all: “insufficient evidence and confidence.”

And the outcome trials are negative

• A 2025 meta-analysis of six randomized trials: reduced opioid consumption, no difference in pain intensity (p = 0.30).
• A 2026 randomized trial across eight US health systems: concordant prescribing rose from 27% to 64%, and outcomes did not differ (p = 0.80). The authors concluded the data “do not support a role for CYP2D6-guided opioid therapy” in contemporary multimodal postoperative care.
• A chronic-pain primary care trial of 253 patients: no difference in three-month pain change (p = 0.74).

One proposed reason: multimodal analgesia appears to attenuate pharmacogenetic effects. When several analgesic mechanisms are running at once, any single enzyme pathway contributes less to the final result.

How to hold both

Explaining variability after the fact and predicting a dose in advance are different problems. Genotype does the first usefully and the second not at all. So the defensible uses are narrow: avoid a known-dangerous drug choice, and document why an existing unusual dose is not unusual for this person. That is how metabolic determination is scoped, and we say so on the page rather than in a footnote.

Also worth remembering: a controlled crossover study found CYP3A blockade had a major effect on every pharmacodynamic measure for oxycodone. Co-medication review is at least as important as genotype, and it is free.

More: the full evidence page.