Cancer Bayesian Causal Network — Pearl SCM Modeler

Baseline 5-yr Mortality

—

no intervention

Naive Combined HR

—

Π HRᵢ (independent)

Pearl-Adjusted HR

—

do(X) · backdoor · ρ̄

Adjusted Mortality

—

5-yr probability

Absolute Risk Reduction

—

ARR · NNT = —

E-Value

—

unmeas. confound. bound

PN — Necessity

—

Tian–Pearl bound

PS — Sufficiency

—

Tian–Pearl bound

PNS — Both

—

monotonicity assumed

PAF — Pop. Attrib.

—

ensemble-wide

Causal DAG — Pearl SCM

do(X) · backdoor-adjusted

Intervention Forest

HR · 95% CI (selected only)

Pareto Frontier · Minimum Effective Set

log-HR reduction vs interventions added

Sensitivity Tornado

leave-one-out Δlog(HR)

Dose–Response Saturation

Emax model · selected

Cross-Correlation ρ — Class Matrix

eigenvalue-corrected

Selected Intervention Ledger

individual contributions to log-HR reduction

Intervention	Category	HR	95% CI	log(HR)	Dose	Adj log(HR)	E-value	Grade	Reference

§ Antithesis & Critical Counter-Arguments

Methods (abbreviated — full appendix in companion PDF)

Framework. Hazard ratios (HR) sourced from SEER 18 (2014–2020), guideline-endorsed pivotal RCTs (ASCO/ESMO/NCCN reference list), and Cochrane/EBCTCG meta-analyses. Combined effect computed under Pearl's Structural Causal Model: do(X) = intervention application, using cited, confounder-adjusted effect sizes (backdoor adjustment is the source study’s) for measured confounders (age, sex, stage, biomarker status) and de-bundling via cross-correlation deflation. Naive combination assumes independence (log-HRs additive). Pearl-adjusted combination applies deflation factor f(N, ρ̄) = 1 − ρ̄·(N−1)/N to the additive log-HR sum, then re-exponentiates. Mean ρ̄ ≈ 0.30 (user-default) reflects moderate within-category bundling (surgery+adjuvant ≈ 0.55; lifestyle cluster ≈ 0.45; targeted+immunotherapy ≈ 0.20).

Counterfactual quantities. ARR = P(death|baseline) − P(death|do(X)). NNT = 1/ARR. Probability of Necessity (PN), Sufficiency (PS), and joint (PNS) computed under monotonicity from Tian–Pearl bounds. Population Attributable Fraction (PAF) from per-intervention risk shares. E-values per VanderWeele & Ding (Ann Intern Med 2017) for unmeasured-confounding robustness.

Variant interaction. Molecular-variant modifiers multiplicatively scale stage-baseline mortality. Variant-restricted interventions (e.g., olaparib for BRCA, osimertinib for EGFR) enter the bundle only if the matching variant is selected; otherwise the slot is null. Multiple variants compose multiplicatively (capped at biological plausibility limits).

Pearl J. Causality: Models, Reasoning, and Inference, 2nd ed., Cambridge UP, 2009.
VanderWeele TJ, Ding P. E-value methodology. Ann Intern Med 2017;167:268–274.
Tian J, Pearl J. Probabilities of causation: bounds and identification. Ann Math AI 2000;28:287–313.
SEER*Explorer, NCI 2024. Stage-specific 5-yr cause-specific mortality, 2014–2020 diagnoses.
NCCN Clinical Practice Guidelines in Oncology (v.2024–2025 series); ESMO Living Guidelines.
Chinn S. SMD–HR conversion. Stat Med 2000;19:3127–31.
EBCTCG, IDEA, ATLAS, FLAURA, KEYNOTE-024/177/189, CheckMate-067/214/649, PRODIGE-24, POLO, EV-302, INDIGO, ADAURA, VIALE-A, PERSEUS, CARTITUDE-4, NIAGARA, MIRASOL — full bibliography in PDF appendix.