Nonrandomized intervention data and analysis

Specify populations estimation and sensitivity

Select baseline confounders using clinical and temporal knowledge. Assess overlap and plausible unmeasured confounding. Avoid automatic adjustment for post-treatment variables, and account for the assignment level in uncertainty estimates.

Flow diagram

Steps

Finalize intervention and outcome data

Resolve queries and freeze prespecified definitions.

Define analysis populations

Link inclusion, adherence and missing outcomes to estimands.

Apply planned adjustment

Use compatible outcome, weighting or matching models with diagnostics for baseline adjustment. Assess overlap and state consistency, exchangeability and positivity assumptions for a causal interpretation.

Handle missingness and uncertainty

Respect clustering and data loss assumptions.

Assess residual bias and report

Include assignment limitations, contamination and prespecified sensitivity.