clinical-research

Overview of clinical-research skill

What clinical-research is for

clinical-research is a study-design support skill for teams shaping a prospective clinical study before protocol submission. It helps structure endpoint selection, two-arm sample size and power estimates, and feasibility phase-gate decisions. The core value is not “writing a protocol”; it is turning an early study idea into a defensible design synopsis with assumptions, risks, and human owners made explicit.

Best-fit users and decisions

The clinical-research skill fits clinical operations, R&D, medical monitor, biostatistics, and product teams working on drug, device, biologic, diagnostic, or digital-therapeutic studies. It is most useful when you need to compare candidate endpoints, pressure-test an assumed effect size, estimate enrollment burden, or decide whether a plan is ready for GO, GO-WITH-CONDITIONS, REDESIGN, or NO-GO.

What makes it different from a generic prompt

This skill includes deterministic Python tools, not just instructions. endpoint_selector.py scores endpoint candidates, sample_size_estimator.py estimates sample size for means, proportions, and survival designs, and phase_gate_scorer.py evaluates feasibility. The reference files also enforce important clinical-research discipline: estimand-first thinking, surrogate endpoint caution, owner sign-off, and enrollment realism.

Important boundaries before adoption

Outputs are estimates and decision support, not medical advice, clinical fact, statistical sign-off, or a submission-ready protocol. A clinician, biostatistician, and regulatory owner must review final decisions. If your actual need is regulatory quality management, ISO 13485, EU MDR, FDA 510(k), PMA, or QSR submission planning, this skill is adjacent but not the right primary workflow.

How to Use clinical-research skill

clinical-research install and first files to read

Install in a compatible Claude/Codex-style skills environment with:

npx skills add alirezarezvani/claude-skills --skill clinical-research

Then inspect the skill directory at research-ops/skills/clinical-research. Read SKILL.md first for workflow scope, then assets/protocol_synopsis_template.md to understand the required inputs. For decision logic, read references/endpoint_and_power.md, references/study_design_canon.md, and references/trial_operations.md. Review scripts last so you know which outputs are generated by rules rather than language-model judgment.

Inputs the skill needs for useful output

Prepare a short protocol synopsis before asking for analysis. Strong inputs include: study phase, product profile, indication, population, arms, allocation ratio, candidate endpoints, whether any endpoint is surrogate or PRO, validation evidence, assumed effect size with citation, alpha, target power, dropout, enrollment months, site count, eligible population, visits, invasive procedures, and budget. Missing effect-size justification and unrealistic recruitment assumptions are the two inputs most likely to weaken the result.

Turning a rough goal into a complete prompt

A weak prompt is: “Help design my trial.” A stronger clinical-research usage prompt is:

“Use the clinical-research skill to evaluate this phase 2 parallel-group RCT synopsis for a drug in [indication]. Candidate endpoints are [list], including [surrogate/PRO status and validation]. Assumed treatment difference is [value] based on [citation/source]. Alpha is 0.05, power 0.80, dropout 15%. We plan [sites] sites, [months] enrollment, eligible pool [number], budget [amount]. Classify endpoints, estimate sample size for [means/proportions/survival], score feasibility, and list assumptions requiring clinician, biostatistician, or regulatory owner review.”

This gives the skill enough context to invoke endpoint scoring, sample-size estimation, and phase-gate reasoning coherently.

Practical workflow with scripts

Start with the protocol synopsis template, then run or ask the agent to use the relevant scripts. Use python3 endpoint_selector.py --sample to inspect endpoint input shape, then provide your own JSON with --input. Use sample_size_estimator.py when the endpoint and effect metric are clear. Use phase_gate_scorer.py after enrollment, site, burden, and budget assumptions are available. If your team uses saved defaults, config_loader.py supports project config at .research-ops/clinical-research.json and global config at ~/.config/research-ops/clinical-research.json.

clinical-research skill FAQ

Is clinical-research for Clinical Research beginners?

Yes, if you treat it as a guided checklist and not as an authority. The references explain endpoint hierarchy, estimands, trial designs, feasibility, and operational burden in practical language. Beginners should still involve a biostatistician early, especially when choosing effect size, multiplicity strategy, adaptive design assumptions, or survival-analysis parameters.

When should I not use this skill?

Do not use it to diagnose patients, choose treatment, claim regulatory acceptability, finalize a protocol, or replace statistical analysis planning. It is also not ideal for retrospective database studies, real-world evidence studies, qualitative research, or complex adaptive simulations unless you extend the workflow. The built-in estimators are strongest for structured prospective two-arm design questions.

How is this different from asking an LLM directly?

A direct prompt may produce a plausible narrative but often hides assumptions. The clinical-research skill forces explicit endpoint classification, surrogate flagging, sample-size assumptions, feasibility scoring, and owner routing. That structure makes review easier and reduces the risk of accepting a polished but underpowered or operationally impossible plan.

Does it fit regulated clinical development?

It fits the pre-submission design discussion, not the final regulated deliverable. The references align with concepts such as ICH E9(R1), GCP, endpoint validation, risk-based monitoring, and feasibility gates, but your organization’s SOPs, statistical standards, medical governance, and regulatory strategy remain controlling.

How to Improve clinical-research skill

Improve clinical-research results with better assumptions

The most important upgrade is better evidence for effect size. Replace “expected medium effect” with a cited minimal clinically important difference, prior trial result, validated biomarker relationship, or natural-history benchmark. For proportions, provide baseline event rate and target absolute difference. For survival, provide hazard ratio, event assumptions, and follow-up expectations. Better assumptions improve both sample-size estimates and phase-gate credibility.

Common failure modes to check

Watch for endpoints that are measurable but not clinically meaningful, unvalidated surrogates proposed as primary endpoints, optimistic enrollment rates, too many burdensome visits, under-specified estimands, and sample sizes reverse-engineered from budget. The skill can flag these issues, but it cannot know whether your citation quality, site network, or competitive trial landscape is truly adequate unless you provide that context.

Iterate after the first output

Do not stop at the first verdict. Ask for a sensitivity pass: “Re-run the design with 10%, 15%, and 25% dropout,” or “Compare feasibility if enrollment takes 9 vs 15 months.” Ask the agent to separate design risks into clinical, statistical, regulatory, and operational owners. This turns a single clinical-research guide output into a decision memo that leadership can act on.

Customize the workflow for your organization

Use the config mechanism for default alpha, power, dropout, product profile, and named owners. Add organization-specific thresholds for feasibility, required endpoint evidence, monitoring complexity, or budget gates. If your program repeatedly uses the same indication or design family, add examples and validated assumptions to local references so future clinical-research outputs are less generic and easier to audit.