Bayesian Causal Network: All-Cause Lymphoma Mortality Reduction

Pearl SCM Framework — E-values · Twin Network · Balke-Pearl Bounds · Mediation · Simpson's Paradox

Factor Table
E-Value Sensitivity
Twin Network
Balke-Pearl Bounds
Mediation
Simpson's Paradox
Dementia Risk
Causal DAG
Tools Used

All 32 Lymphoma Mortality Risk & Protective Factors

FactorHR95% CIDirectionCategorySource

Hazard Ratios — Forest Plot

Factor Distribution by Category

E-Value Sensitivity Analysis (VanderWeele & Ding 2017)

E-value = minimum confounding strength to explain away the observed effect. E > 2.0 = robust.

E-Value Detail Table

FactorHRE-value (point)E-value (CI)Robust

Twin Network Counterfactual Analysis (Pearl 2000)

PN = Probability of Necessity; PS = Probability of Sufficiency. Baseline 10-yr mortality = 12%.

Counterfactual Detail Table

FactorActual 10yrCounterfact 10yrARDPNPSNNT

Balke-Pearl Sharp Bounds on ACE

Bounds calibrated via E-value confounding sensitivity. All 32 factors are sign-definite.

NDE/NIE Mediation Decomposition (Pearl 2001)

Simpson's Paradox Scenarios & do-Calculus Resolution

All-Cause Dementia Risk — Cross-Correlated vs Independent Model

20 modifiable factors with literature-informed correlation matrix. Multivariate probit framework.

Individual Factor PAFs

Causal Directed Acyclic Graph (DAG)

Tools & Methods Used