Results Descriptive Event‐Time Hazard Study

Descriptive Event-Time Hazard Study: Mortality Timing Around COVID Vaccination Eligibility Dates in Czechia (2020–2023)

This page summarizes the descriptive analysis of all-cause mortality hazards in relative time around cohort-specific vaccination eligibility dates, using full population data from Czechia (~10.9M subjects aged 0–112).

Important notes:

• This is descriptive only — event-time hazard patterns around eligibility dates.
• Not causal inference on vaccination effects — eligibility ≠ treatment; patterns are consistent with selection bias (healthy vaccinee effect, depletion of susceptibles).
• Pre-trends rejected in most cohorts (joint F-test p < 0.001) → strong evidence of bias.
• Low-vaccination young groups (0–5, 5–12) show flat/null patterns → negative controls.

Method

• Data: Full Czech population with exact dates of vaccination and death (Vesely_106_202403141131.csv).
• Approach: Aggregated weekly bins in event time relative to cohort-specific eligibility dates.
• Model: Grouped Poisson GLM with offset log(n_at_risk), where n_at_risk = survivors to bin start, n_deaths = deaths in bin.
• Reference: Closest pre-eligibility bin (typically -4 days).
• Stratification: By age cohorts based on Czech rollout phases.
• Diagnostic: Joint F-test on lead coefficients (pre-trends).
• Output: Interactive Plotly HTML plots per cohort, CSV results, run_info JSON.

Cohort-Specific Results

0–5 Years (Low/No Early Vaccination – Negative Control)

• Low deaths per bin (mostly 0–6).
• Coefficients highly variable with large SEs (low power).
• Pattern: No clear trend; mostly null/flat around eligibility (late rollout).
• Pre-trends: Likely flat (low events).
• Interpretation: Flat/null pattern supports bias dominance in older groups.

5–12 Years (Low Vaccination)

• Low deaths, high SEs.
• Coefficients variable, many near zero or high magnitude with wide CIs.
• Pattern: No consistent trend; flat/null.
• Interpretation: Negative control – no signal expected due to low exposure.

12–18 Years (Adolescents, Late Rollout)

• Low–moderate deaths.
• Coefficients variable (some positive pre, negative post).
• Pattern: Some pre-trend noise, post slightly negative but not sharp.
• Interpretation: Low exposure → weak/null signal.

18–60 Years (Moderate Vaccination)

• Higher deaths per bin.
• Pre-eligibility: Positive/near-zero leads.
• Post-eligibility: Gradual negative trend (log HR ~ -0.1 to -0.3).
• Pre-trends rejected.
• Interpretation: Selection bias evident; gradual decline consistent with depletion.

60–70 Years (High Vaccination)

• Higher deaths.
• Pre-eligibility: Positive leads (selection into eligibility).
• Post-eligibility: Sustained negative (log HR ~ -0.2 to -0.5).
• Pre-trends rejected.
• Interpretation: Stronger bias in high-risk group.

70–80 Years (Very High Vaccination)

• High deaths.
• Pre-eligibility: Negative leads (healthy vaccinee).
• Post-eligibility: Strong negative trend (log HR to ~-0.9).
• Pre-trends rejected.
• Interpretation: Classic selection pattern.

80–120 Years (Highest Risk, Earliest Rollout)

• High deaths.
• Pre-eligibility: Near-zero/negative.
• Post-eligibility: Strong negative (log HR to ~-0.57).
• Pre-trends rejected.
• Interpretation: Bias + possible small effect in high-risk group.

Overall Findings

• Pre-trends rejection across cohorts → selection bias dominates (healthy vaccinee, depletion).
• Post-eligibility: Gradual negative trend in older/high-vax groups; flat/null in young/low-vax → consistent with bias-dominated association.
• No sharp discontinuity at eligibility → not suggestive of immediate vaccine effect.
• Negative controls (0–5, 5–12): Flat patterns reinforce bias story.

Full Population Plot (0–112 Years)

Interactive Plots

• Full population: event_study.html (open in browser)
• Cohort-specific HTML files available in repo.

Data & Code

• Script: tdr_dit_final.py
• Results CSVs: per cohort (e.g., results_70-80.csv)
• Metadata: run_info_{cohort}.json

Conclusion

The descriptive patterns strongly support selection bias as the primary driver of apparent mortality differences around eligibility. Low-vax young groups show flat trends, while older groups show gradual declines consistent with cumulative bias. No clear signature of large causal vaccine reduction in ACM.

For questions, open an issue in the repo.