Clinical Cohort Protocol Designer

Problem

Clinical cohort design often fails because the research question, source population, time-zero, baseline variables, and follow-up endpoints are mixed together. This skill targets retrospective or prospective clinical cohort protocols, not patient-specific medical advice, randomized trial protocols, case-control designs, or finished manuscripts.

How it works

It converts a clinical research question into a reviewable protocol framework:
- Decides whether a cohort design is appropriate and states the interpretation level: descriptive, associative, predictive, or quasi-causal.
- Chooses a retrospective or prospective cohort structure based on EHR, registry, claims, or prospective recruitment realities.
- Defines the source population, inclusion/exclusion criteria, index date, baseline window, and enrollment logic.
- Specifies follow-up start, observation horizon, censoring rules, competing events, and endpoint ascertainment.
- Separates baseline variables, exposures, predictors, effect modifiers, and post-baseline variables to avoid treating follow-up measurements as baseline confounders.
- Selects a main analysis line matched to the endpoint structure, such as time-to-event, binary, longitudinal, or competing-risk models.

Boundaries

The skill clarifies missing inputs instead of overbuilding a protocol. If disease, population, exposure/predictor, endpoint family, or follow-up horizon is unclear, it may ask 2–6 high-yield questions; if a one-shot output is requested, it labels assumption-dependent elements. It is not for individualized care, pure literature review, diagnostic accuracy studies, or mechanistic wet-lab plans.