Case study · Pillar 2 · Companion

Pembrolizumab + Lenvatinib — Efficacy Framework.

The efficacy companion to the safety flagship. Max-of-arms × indication-specific synergy coefficient. The framework's hardest test was the LEAP-002 negative trial — v1.0 predicted positive (s = 0.30 transferred from RCC); the trial was negative; v2.0 calibrated s = 0.10 for HCC and now reproduces the negative outcome.

Open draft report → Read the methodology
Why a separate document

Safety and efficacy use different arithmetic.

Safety AEs are independent events from two mechanisms (irAE from pembro, VEGFR from lenva). Additive arithmetic. Efficacy responses overlap on the same responder population. Max-of-arms × synergy arithmetic. Mixing the two breaks both predictions. The efficacy report is a companion to the safety report, not a section of it.

The three efficacy insights

Different from the safety insights.

1 · Max-of-arms × synergy beats additive for ORR.

IO and TKI partly target the same responder population — additive arithmetic over-predicts. The synergy coefficient (s = 0.10–0.55 across indications) captures the marginal contribution above the better monotherapy.

2 · IO+TKI amplifies depth and durability of response.

CLEAR shows 16% CR (3× pembro mono, 8× lenva mono) and 26.7 mo median DoR. Combination converts shallow PRs into CRs more often than either single agent.

3 · PFS gain decouples from OS gain.

CLEAR mPFS gained 14.7 mo (HR 0.39) but mOS gained only −0.6 mo (HR 0.79 NS) because the control arm allowed crossover to active IO. Predicting OS requires modeling subsequent therapy.

The hardest test

LEAP-002 — when synergy lifted from RCC was wrong.

Metricv1.0 pred (s = 0.30)v2.0 pred (s = 0.10)Observed
ORR32%27%26.1%
mPFS10.5 mo8.4 mo8.2 mo
mOS23.5 mo21.0 mo21.2 mo
OS HR vs lenva mono0.74 (positive)0.85 (NS)0.84 (NS)
Trial verdictpredicted positive ✗predicted negative ✓negative
The lesson: synergy coefficients reflect tissue-specific immune-vascular biology, not portable drug-class properties. Cirrhotic immune dysfunction collapses IO contribution in HCC. The post-trial v2.0 calibration to s = 0.10 reproduces the negative outcome.
Validation summary

Four trials, ±5% on ORR, ±2 mo on mPFS.

TrialORR ΔmPFS ΔmOS ΔSynergy coef
CLEAR (RCC, fit)+2%+1.4 mo+3.7 mo0.45 (calibration)
KN-775 (endometrial)−1.7%+0.1 mo+1.3 mo0.55 (highest)
LEAP-002 (HCC) — negative−0.9% (v2.0)−0.2 mo (v2.0)+0.2 mo (v2.0)0.10 (post-hoc)
LEAP-012 (HCC + TACE)+3% (v2.0)+1.1 mo (v2.0)awaiting OS0.30 (TACE rescue)

Open the draft report.

Companion 2,500-line analysis with calculator, response grid, comparator, attribution trees, and Supabase comments. Mirror structure to the safety flagship.

Open draft report →