Gradient Biotech

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

DataRequiredPurpose
Episodes with outcomes (event_observed, time_to_event_days)YesThe compared endpoint
Interventions (product, start/stop time)For treatmentLandmark treatment groups
Vaccination status (participant/episode attribute)For VEVaccinated vs unvaccinated groups
Linked resistance resultsOptionalTreatment × 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

Related guides