What is established

CYP2D6 and drug selection

The Clinical Pharmacogenetics Implementation Consortium’s 2021 guideline — still current — makes actionable recommendations for two drugs:

• Poor metabolizers (activity score 0): “current evidence supports the avoidance of codeine and tramadol and the use of an alternative analgesic due to the likelihood of suboptimal or lack of effect.” CPIC explicitly rejects dose escalation as a workaround — “There is insufficient evidence in the literature to recommend a higher dose of codeine or tramadol in poor metabolizers.”
• Ultrarapid metabolizers (activity score above 2.25): codeine or tramadol “should not be used, in order to avoid the risk of severe toxicity with label-recommended dosing.”

This is a safety-avoidance rule about which drug to choose. It is the strongest pharmacogenomic recommendation in this area and it is narrower than it is usually reported to be.

The clinical stakes are established. FDA acted in April 2017 to contraindicate codeine for pain and cough and tramadol for pain in children under 12, contraindicate tramadol under 18 after tonsillectomy or adenoidectomy, and warn against use in obese adolescents and those with obstructive sleep apnea or severe lung disease. The adverse event reports cited included 64 cases of serious breathing problems with codeine, 24 of them deaths, some after a single dose. The index case in the literature was a single case report of a two-year-old with obstructive sleep apnea who died after adenotonsillectomy, with an ultrarapid CYP2D6 genotype confirmed.

CYP3A matters as much, and gets ignored

A randomized, double-blind, placebo-controlled five-arm crossover study in 10 genotyped healthy volunteers given oral oxycodone with and without CYP2D6 blockade (quinidine) and CYP3A blockade (ketoconazole) found that CYP2D6 blockade reduced the subjective pain threshold by about 30%, and that CYP3A4 blockade had a major effect on every pharmacodynamic measure. It is a small experimental-pain study, and it is the cleanest demonstration that variability is not a CYP2D6 story alone. In practice this means co-medication review is at least as important as genotype.

And the outcome trials are negative

This is the part that a vendor selling pharmacogenomic testing will not lead with, so we will.

• A 2025 systematic review and meta-analysis of six randomized trials of pharmacogenomic-guided opioid therapy found reduced opioid consumption (SMD −0.38, 95% CI −0.67 to −0.08) but no difference in pain intensity (SMD −0.31, 95% CI −0.89 to 0.27, p = 0.30), with adverse-event data available from only one trial.
• A 2026 open-label randomized trial across eight US health systems enrolled 1,602 patients and analyzed 351 with an actionable phenotype. The intervention worked as an intervention — concordant prescribing rose from 27% to 64% (p < 0.001) — and outcomes did not differ (integrated analgesic score difference 2.8, 95% CI −18.3 to 23.8, p = 0.80). The authors concluded that the data “do not support a role for CYP2D6-guided opioid therapy in the contemporary postoperative setting of multimodal pain management.”
• A chronic-pain primary care randomized trial of 253 patients found no difference in three-month pain change (p = 0.74) among the 106 CYP2D6 intermediate and poor metabolizers.

One proposed explanation, from an analysis finding no association between CYP2D6 phenotype and opioid consumption or pain for hydrocodone, oxycodone or even the tramadol positive control: multimodal analgesia appears to attenuate pharmacogenetic effects. When a patient is receiving several analgesic mechanisms at once, the contribution of any single enzyme pathway to the final outcome is diluted.

How to hold both halves at once

The thirteen-fold dose variability described in step five is real, and genetic differences in hepatic metabolism account for three-fold or more of it. That is a fact about pharmacology and it is not in dispute.

It does not follow that a genotype predicts a dose. Explaining variability after the fact and predicting the right dose in advance are different problems, and the trials tested the second. This is why metabolic determination in this program is scoped to explaining an outlier and avoiding a known-dangerous drug choice — not to setting a dose.

A citation trap worth recording, because it inflates the apparent evidence base. The frequently cited COMT origin study of 207 cancer patients and a second paper often listed alongside it use the same 207-patient cohort. They are not independent replications.

Questions this raises

Should an agency fund routine pharmacogenomic testing for pain patients?

Not on the basis that it improves pain outcomes — the randomized evidence does not support that, including a 2026 eight-site trial that changed prescribing substantially and changed outcomes not at all. There is a narrower, well-supported case for CYP2D6 testing to avoid codeine and tramadol in poor and ultrarapid metabolizers. See what pharmacogenomics can and cannot do.

If genotype does not predict dose, why does dose vary so much?

Because dose requirement is driven by pain severity, drug interactions affecting CYP3A and CYP2D6, receptor-level differences, neural transmission and tolerance — a combination no single genotype captures. Explaining variability and predicting a dose are different problems. See step five.