Evidence
How the engine behaves on a real cohort.
We ran the engine against the IWPC warfarin cohort of 5,700 patients. Predicted risk tiers track the doses patients were actually prescribed — a retrospective consistency check, not a clinical validation.
The IWPC dose gradient
Predicted risk tiers line up with actual prescribed weekly dose in a clean, monotonic gradient — roughly 45.8 mg/week at the low tier, 33.63 at standard, down to 21.51 mg/week at the high tier — across 827, 1,267 and 1,904 patients respectively.
A dose gradient is not a clinical outcome. It shows that the engine’s tiers track real prescribing — not that any patient was dosed wrongly, and not that the engine would have improved their care. We make no outcome claim from this cohort. Run with anukriti-pgx-core 0.11.0 — the engine production serves — re-run on 16 November 2026 against IWPC (Klein et al., NEJM 2009; CPIC 2017). The three figures are identical to the decimal on engines 0.7.1, 0.9.0 and 0.11.0, so the gradient reflects the CPIC rules rather than any one engine release.
What’s behind the numbers
- IWPC warfarin cohort — 5,700 patients.
- 2,504 real 1000 Genomes phase-3 samples across five superpopulations.
- 4 live workflows, all CPIC level A.
Open findings
A companion CPIC-table audit ships with open findings disclosed, including a CYP2C9 table mismatch our own audit found. We publish what doesn’t work yet, not only what does — that transparency is part of the platform’s credibility.
Why population-aware PGx matters
Published evidence shows genotype-frequency differences materially change drug safety and efficacy — clopidogrel (CYP2C19), warfarin (CYP2C9 + VKORC1), simvastatin (SLCO1B1), and carbamazepine (HLA-B*15:02) are documented examples.