Results and Result‐Files AG70

This page provides the raw output files for each methodology.

Every method is tested against the real dataset, a stress-test reclassification, and a null-hypothesis simulation to ensure scientific rigour.

---

AC) Hernán Style Pooled Logistics (RMST)

Scientific gold standard (except that I did not use covariates by design)

What it asks: "Based on a statistical model, what would the difference in survival time have been if everyone in the population was vaccinated versus if no one was vaccinated?" How it works: It uses person-day regression (Logistic Model) to estimate the counterfactual. It smooths the raw data to find the underlying treatment effect while adjusting for time-varying factors.

• Real Data (Czech-FOI): Standard AG70 Analysis (.txt)
  • ΔRMST: 17.32 days | 95% CI: [15.13, 19.55] | VE: 62.9% | Survival: 97.4% vs 93.0%
• Stress Test (5% UVX->VX): Reclassified Dataset Analysis (.txt)
  • ΔRMST: 6.27 days | 95% CI: [-3.94, 10.74] | VE: 40.0% | Survival: 95.8% vs 93.1%
• Null Hypothesis (HR=1): Simulation Dataset Analysis (.txt)
  • ΔRMST: -0.48 days | 95% CI: [-3.37, 1.33] | VE: -1.6% | Survival: 96.2% vs 96.3%

---

AE) Empirical Dynamic CC RMST (Selection Bias Diagnostic)

What it asks: "Based on raw history, what is the survival gap if we compare every person to an identical 'digital twin' starting at Time 0?" How it works: A Clone-Censor (CC) design. It compares the Historical ΔRMST (raw record) against the Clone–Censor ΔRMST (the "Historical Perfect Experiment").

• Real Data (Czech-FOI): Standard AG70 Log (.log)
  • Historical ΔRMST: +15.04 days
  • Clone–Censor ΔRMST: +14.80 days | 95% CI: [+12.63, +16.68] | VE: +57.5%
  • ΔΔRMST (Bias/Selection Diagnostic): -0.24 days
• Stress Test (5% UVX->VX): Reclassified Dataset Log (.log)
  • Historical ΔRMST: -17.47 days
  • Clone–Censor ΔRMST: -20.10 days | 95% CI: [-44.29, -5.01] | VE: -14.6% | Survival: 92.2% vs 93.0%
  • ΔΔRMST (Bias/Selection Diagnostic): -2.63 days
• Null Hypothesis (HR=1): Simulation Dataset Log (.log)
  • Historical ΔRMST: -1.23 days
  • Clone–Censor ΔRMST: -1.80 days | 95% CI: [-4.69, +0.74] | VE: -6.7% | Survival: 93.3% vs 93.3%
  • ΔΔRMST (Bias/Selection Diagnostic): -0.56 days

---

AE) C.S. Pierce Evidence Weighted RMST (experimental)

What it asks: "How much 'weight' does each death carry, and what is the robust signal when we adjust for the density of evidence?" How it works: An information-theory approach. It filters survival signals through a "surprisal" pipeline (Abductive Logic) to distinguish between systematic effects and noise.

• Real Data (Czech-FOI): Download Log
  • Final Empirical ΔRMST: 12.63 days | VE: 46.1% | Survival: 96.8% vs 94.1%
  • Final Weighted ΔRMST: 2.49 days | VE: 9.1% | Survival: 95.5% vs 95.1%
• Stress Test (5% UVX->VX): Download Log
  • Final Empirical ΔRMST: -0.688 days | VE: -3.5% | Survival: 95.2% vs 95.8%
  • Final Weighted ΔRMST: -0.747 days | Interpreted Signal: "Evidence favoring Unvax"
• Null Hypothesis (HR=1): Download Log
  • Final Empirical ΔRMST: 1.843 days | VE: 4.8% | Survival: 96.2% vs 96.3%
  • Final Weighted ΔRMST: 0.324 days | Interpreted Signal: "Noise / Low Evidence"

---

AE) Landmark RMST

What it asks: Starting from the first vaccination (the landmark), is there a difference in survival between vaccinated people and those not yet vaccinated?
How it works: Everyone alive at the landmark is included.

• Primary analysis: keeps all participants even if they later change vaccination status, reflecting real-world conditions.
• Sensitivity analysis: removes people who switch groups, which can overstate the benefit because the remaining group may be healthier.

• Real Data (Czech-FOI): Standard AG70 Log (.log)
  • Primary ITT-like
    ΔRMST (difference in restricted mean survival time): +1.31 days | 95% CI: [0.94, 1.97] | VE approx: 0.4% | Survival at 365 days: 96.3% vs 95.1% (vx vs uvx)
  • Sensitivity censoring crossover
    ΔRMST: +4.67 days | 95% CI: [3.56, 5.32] | VE approx: 1.3% | Survival at 365 days: 96.3% vs 95.0% (vx vs uvx)
• Stress Test (5% UVX->VX): Reclassified Dataset Log (.log)
  • Primary ITT-like
    ΔRMST: +1.15 days | 95% CI: [0.85, 1.93] | VE approx: 0.3% | Survival at 365 days: 94.4% vs 95.1% (vx vs uvx)
  • Sensitivity censoring crossover
    ΔRMST: +4.13 days | 95% CI: [3.30, 5.21] | VE approx: 1.2% | Survival at 365 days: 94.4% vs 95.0% (vx vs uvx)
• Null Hypothesis (HR=1): Simulation Dataset Log (.log)
  • Primary ITT-like
    ΔRMST: +0.01 days | 95% CI: [-0.31, 0.40] | VE approx: 0.0% | Survival at 365 days: 96.0% vs 96.0% (vx vs uvx)
  • Sensitivity censoring crossover
    ΔRMST: +0.25 days | 95% CI: [-0.01, 0.57] | VE approx: 0.1% | Survival at 365 days: 96.0% vs 96.0% (vx vs uvx)

Bias Note:

• Primary analysis is least biased: all participants stay in the comparison, though later changes in vaccination can slightly reduce the observed effect.
• Sensitivity analysis may overstate benefit: removing switchers leaves a healthier subset, even though the groups remain “pure.”

---

🏠 Back to Home