Same regimen, 5× different fatal rate.
Lenvatinib 20 mg + pembrolizumab 200 mg Q3W. Same dose, same indication class. Fatal TRAE rate ranges from 1.1% in CLEAR (fit RCC) to 5.7% in KN-775 (fragile post-platinum endometrial) — a 5× range driven by population, not by regimen.
Conventional FDA labeling reports a single fatal rate based on the pivotal trial. Clinicians counseling a patient with a different population profile have no quantitative framework to adjust the label-based estimate. The framework fills that gap.
What the framework discovered.
1 · Additive mono model is the correct primary baseline.
~85% of AEs are predicted within 5% by independent-events arithmetic from monotherapy baselines. Failures: hypothyroid supra-additivity (dual-hit on thyroid by both arms), discontinuation attribution overhead.
2 · Lenva dose-titration explains sub-additive constitutional AEs.
By month 3, ~70% of patients are on 14 mg, not 20 mg. The dynamic dose envelope matters. Predicting from start dose alone over-estimates fatigue, diarrhea, weight loss.
3 · Population fragility > dose > drug interaction.
5× range in fatal TRAE (1.1% fit RCC to 5.7% fragile endometrial). Dose contributes ~1.5×, DDI is negligible. The patient mattered more than the regimen.
Four trials, ±5% on Gr3, ±1–2pp on fatal.
| Trial | Gr3 TRAE pred vs obs | Fatal TRAE pred vs obs | Disc pred vs obs |
|---|---|---|---|
| CLEAR (RCC fit, calibration) | 70.3% vs 71.6% | 2.5% vs 1.1%* | 26.5% vs 37.2% |
| KN-775 (endometrial fragile, v2.0) | 87% vs 88.9% (v2.0) | 5.85% vs 5.7% | 44.3% vs 33.5% |
| LEAP-002 (HCC CP-A, v2.0) | 56.4% vs 61.5% | 1.3% vs 1.0% (v2.0) | 24.9% vs 17.7% |
| LEAP-012 (HCC + TACE, v2.0) | 71.4% vs ~71% | 2.0% vs ~2.0% | ~20% vs ~14% |
* CLEAR fatal over-prediction (2.5% vs observed 1.1%) corrected in v2.0 with the "extra-fit downward coefficient" — RCC enrollment is fitter than the framework's default population.
Four documents shipped.
| Document | Audience | Format |
|---|---|---|
| Main safety analysis report | Clinical pharmacology advisors, drug-development teams | Interactive HTML · 210 KB · v2.3 calculator + comparator + attribution trees |
| Manuscript draft | Journal editors, peer reviewers | Peer-review-ready · target Clinical Cancer Research |
| KOL deck | Clinical advisory boards, KOL roundtables | 12-slide presentation · arrow-key navigation |
| Patient counseling one-pager | Shared decision-making visits, informed consent | Single page · plain-language risk framing · natural frequencies |
Open the live report.
Calculator, comparator, attribution trees, Supabase reviewer comments — all functional.
Open live report →