clinical-trial-design
Designing clinical trials — phases, randomization, endpoints, sample size, and CONSORT-compliant reporting.
- License 许可
- MIT
- 最近更新
- 2026-11-07
- Stars
- ★ 0
- Forks
- 1
01 前言 Front Matter
- name
- clinical-trial-design
- category
- scientific
- description
- Designing clinical trials — phases, randomization, endpoints, sample size, and CONSORT-compliant reporting.
03 SKILL.md 正文
Overview
clinical-trial-design covers the architecture of interventional clinical studies: choosing the phase and design (parallel, crossover, factorial, adaptive), defining endpoints, randomizing and blinding, sizing the sample, and planning analyses that regulators and journals accept. A trial's credibility is fixed at the design stage — no analysis rescues a trial that measured the wrong thing in the wrong population.
When to use
- Choosing trial phase and design for a new intervention.
- Defining primary/secondary endpoints and estimands.
- Randomization schemes: simple, blocked, stratified, minimization.
- Blinding: who, how, and when it's impossible (surgical/behavioral trials).
- Sample-size calculation and interim analyses with alpha spending.
- Adaptive designs: group sequential, sample-size re-estimation, platform trials.
- Writing protocols and CONSORT-compliant reports.
Core concepts
- Phases. Phase I: safety/tolerability (often healthy volunteers, dose escalation). Phase II: preliminary efficacy + dose finding. Phase III: confirmatory efficacy vs standard of care (large, randomized, the basis for approval). Phase IV: post-marketing surveillance. Skipping phases is occasionally justified, never casual.
- Estimands (ICH E9 R1). Define precisely what treatment effect you estimate: population, treatment conditions, endpoint, handling of intercurrent events (discontinuation, rescue medication), summary measure. "Intention-to-treat" is an estimand choice, not a default virtue — state it.
- Endpoints. Primary: one (or a co-primary pair), clinically meaningful, measured reliably. Surrogate endpoints (biomarkers) are acceptable only when validated — lowering a lab value is not the same as helping patients. Patient-reported outcomes belong in most modern trials.
- Randomization. Blocked randomization keeps arms balanced over time; stratification balances key prognostic factors (site, disease severity); minimization for many strata. Use a central, concealed system — envelope randomization is a known failure point.
- Blinding. Double-blind where possible; when impossible (surgery, psychotherapy), blind the outcome assessors and statisticians at minimum, and use objective endpoints. Assess blinding success — failed blinding with subjective endpoints is a major bias source.
- Sample size. Powered on the primary endpoint's minimum clinically important difference, with realistic variance and dropout inflation (10-20%). Underpowered trials are unethical — they expose patients to risk for uninterpretable results.
- Interim analyses. Group-sequential designs with alpha-spending functions (O'Brien-Fleming, Pocock) allow early stopping for efficacy or futility. Unplanned peeking at the data inflates Type I error — all interim looks must be prespecified with a data monitoring committee.
- Design variants. Crossover (each patient gets both treatments — efficient, but only for stable chronic conditions without carryover); factorial (two interventions at once, tests interaction); non-inferiority (margin must be clinically justified, not statistically convenient); platform/adaptive trials (shared infrastructure, dropping arms — efficient but operationally demanding).
Practical workflow
- PICO. Population, Intervention, Comparator, Outcome — one sentence that the whole trial hangs on. If the comparator isn't the real clinical alternative, the trial answers the wrong question.
- Estimand and endpoints. Primary endpoint, intercurrent-event strategy, analysis population. Register on ClinicalTrials.gov before enrollment.
- Design choice. Parallel-group default; crossover/factorial/adaptive only with justification.
- Size it. Power 80-90% on the MCID; inflate for dropout; document assumptions.
- Randomize and blind. Central system, stratified blocks; blinding plan with fallback for unblindable interventions.
- Monitor. Independent DMC, prespecified interim analyses, safety stopping rules.
- Analyze as planned. ITT primary; per-protocol as sensitivity; handle missing data with prespecified methods (not last-observation-carried-forward).
- Report CONSORT. Flow diagram, baseline table, primary result with effect size + CI, harms by arm. Publish regardless of outcome — publication bias kills evidence synthesis.
Common pitfalls
- Surrogate primary endpoints without validation (the drug works on the biomarker, not the patient).
- Underpowered trials ("exploratory" as an excuse for n=30 confirmatory claims).
- Unplanned interim looks inflating false positives.
- Per-protocol presented as primary (breaks randomization).
- Non-inferiority margins chosen to guarantee success.
- Comparator that's a straw man (placebo where an active standard exists).
- Selective outcome reporting — publishing a different primary endpoint than registered.
04 同仓库更多 Skills
web-scrape
Turn web pages into LLM-ready markdown via the Firecrawl API (renders JS, anti-bot handling). Use when browser_open/fetch fails on JS-heavy pages, or when you need clean markdown/structured extraction from a site. Requires FIRECRAWL_API_KEY.
design-taste-frontend
tasteskill v2, the anti-slop frontend skill. Art direction for websites, distilled from a study of 247 curated gallery sites (One Page Love, Awwwards Sites of the Day). Deals researched directions (idea, composition, type, palette, photographs, section blueprints) from a library, enforces a strict no-slop house style with a real-browser inspector, and runs a look-and-fix loop on screenshots. Use this whenever you build, design, redesign, restyle, polish or review a landing page, marketing site, portfolio, product page, one-pager, hero or any public-facing web page, in any stack (HTML/CSS, React, Next.js, Vue, Svelte, Astro, Tailwind) — also when the user says a page looks generic, templated, "AI-made", sloppy, or asks to make it look designed, premium, award-level or less like every other site. Not for dashboards or app screens.
test-driven-development
Use when implementing any feature or bugfix, before writing implementation code
systematic-debugging
Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes
requesting-code-review
Use when completing tasks, implementing major features, or before merging to verify work meets requirements
brainstorming
You MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation).