Intervention effectiveness
Compare clinical outcomes across treatment and vaccination groups with bias-aware, descriptive methods — a landmark design for treatment and observational VE for vaccination. These are association-only research estimates, never treatment effectiveness for prescribing or vaccine efficacy for policy.
Research question
Do outcomes differ by treatment or by vaccination status, once immortal-time bias is addressed and confounding is made visible?
Who this is for
- Anti-infective development teams evaluating treatment response and resistance context
- Vaccine researchers defining breakthrough cohorts and effectiveness estimates
- Researchers who need confounding and time-bias caveats surfaced, not hidden
Data requirements
| Data | Required | Purpose |
|---|---|---|
Episodes with outcomes (event_observed, time_to_event_days) | Yes | The compared endpoint |
| Interventions (product, start/stop time) | For treatment | Landmark treatment groups |
| Vaccination status (participant/episode attribute) | For VE | Vaccinated vs unvaccinated groups |
| Linked resistance results | Optional | Treatment × resistance stratification |
Workflow
Model episodes, outcomes, interventions, vaccination
→ Treatment outcome (landmark) and/or Vaccine effectiveness
→ Review estimates + caveats → AI interpretation
Step 1 — Treatment outcome
The Treatment outcome analysis groups episodes by recorded interventions using a landmark (treatment status fixed as of index_date + landmark_days, events before the landmark excluded) to defuse immortal-time bias. It reports Kaplan-Meier time-to-event, log-rank, risk ratio/odds ratio, and a treatment × resistance cross-tab from AMR linked from Microbiology.
Step 2 — Vaccine effectiveness
The Vaccine effectiveness analysis compares vaccinated vs unvaccinated episodes on a severe outcome, reporting VE = 1 − risk/odds ratio with a confidence interval, a breakthrough-infection summary, an optional test-negative design, and subgroup views.
Step 3 — Read the caveats
Both surface explicit caveats — confounding by indication, immortal-time mitigation, healthy-vaccinee bias, unmodeled waning, and small groups — and never claim effectiveness, efficacy, or a prescribing/policy recommendation.
Expected outputs
- Landmark treatment groups with KM/log-rank, RR/OR, and resistance cross-tab
- VE with confidence interval, breakthrough summary, and optional test-negative/subgroup views
- Bias and confounding caveats attached to every estimate