Plan a causal cohort analysis
Start with a treatment decision and a population, then align eligibility, exposure assignment and follow-up. Record the adjustment rationale and support limitations before interpreting a fitted association as a causal effect.
Worked decision
Hypothetical decision: A cohort labels participants exposed if they start treatment within 90 days after diagnosis, but counts follow-up from diagnosis. A person who starts on day 60 must remain eligible until that day to enter the exposed group. The team pauses the baseline comparison, aligns eligibility, assignment and follow-up using a defensible design, and records how grace periods and time-varying treatment will be handled. Adding more confounders to the original model would not correct the time-zero error.
Record your decision
Can this cohort support the intended treatment comparison? Resolve time-zero misalignment, impossible exposure comparisons and unclear covariate roles first. A balanced measured sample does not establish that unmeasured confounding is absent. Use the cohort SAP to record the estimator, assumptions and sensitivity analyses; retain associational wording when causal identification is not defensible.
Work through the evidence
- Specify the population and treatment strategies (TXT)
- Align eligibility, assignment and follow-up (TXT)
- Build a defensible confounder adjustment rationale
- Record covariate roles and measurement timing (TXT)
- Inspect standardized differences in measured baseline covariates
- Document unsupported exposure comparisons (TXT)
- Complete the observational cohort analysis plan (DOCX)
- Read the authors’ free Causal Inference textbook