Drug-specific signals in oncology: an internal check (May 2026)
An internal check of two drug-specific signals on 85 oncology programs with known outcomes. Biomarker quality lined up with outcomes; Phase 1 response rate added little. It is a dated record, not a validation of the model or peer-reviewed research. The model uses the published biomarker values, and every value can be changed.
What we checked
In May 2026 we looked at two drug-specific signals on a set of 85 oncology programs whose outcomes are known: biomarker quality (whether a program selects patients by a genomic alteration) and the objective response rate reported in Phase 1.
What we saw
Biomarker quality lined up with outcomes in this set. The model does not use a value fitted to these programs: it uses the published values from Schwaederle et al. 2016 (1.35× genomic, 0.85× protein-only), and only after a program’s classification is reviewed.
Phase 1 response rate added little once biomarker quality was known, and the set was too small to tell either way. It is shown as a flag beside the valuation and does not change PoS.
What this is not
The set was small, the analysis was ours, and it was not peer reviewed. It is not a validation of the model and says nothing about any individual asset. The full report is kept unchanged as a dated record; some of its wording predates Methodology M5.
01Results recorded
What we saw, and what the model does.
Biomarker quality
As recorded. +5.2pp held-out AUC over the structural baseline, stable across cohort sizes. The genomic_validated cohort odds ratio was 5.59 vs the Schwaederle (2016) literature anchor of 1.35 — we ship the lower anchor with the overshoot disclosed.
What the model does. Uses the published Schwaederle 2016 values (1.35× genomic, 0.85× protein-only) for oncology solid tumor at Phase II/III, after review.
Phase 1 objective response rate (ORR magnitude)
As recorded. A joint biomarker x ORR-bucket model beat the biomarker-only baseline by only +0.5pp held-out AUC (paired DeLong p=0.48), and the comparison is statistically unpowerable — detecting a +3pp gain at this baseline needs ~830 drugs; the cohort is 85. This followed the Phase 1 finding that the two signals combined fell below baseline (-0.3pp at 43-drug coverage) — the double-counting that motivated the joint-table test.
What the model does. Shown as a flag beside the valuation. Does not change PoS.
Limitations
Small set (85 programs; the biomarker result is from NSCLC), analysed by PhaseFolio and not peer reviewed. Failed programs report early-phase response rates less often. Not a validation of the model.