Clinical Decision Support

Clinical Decision Support

Safety Boundary

Do not present recommendations as current medical fact without checking authoritative, current sources. Use this skill for research and documentation support, not for direct diagnosis, prescribing, or emergency decisions. For individual care plans, use treatment-plans and keep clinician review explicit.

Workflow

1. Define document purpose: cohort analysis, evidence summary, treatment algorithm, biomarker strategy, or regulatory/research support.
2. Identify population, intervention/exposure, comparator, outcomes, time horizon, and required jurisdiction or guideline body.
3. Search or verify current authoritative sources before clinical claims: guidelines, regulator labels, trial publications, systematic reviews, and institutional policy.
4. Grade evidence and uncertainty. Separate observed data, model-derived estimates, expert consensus, and assumptions.
5. Build only the required document sections: executive summary, evidence table, statistical methods, recommendation rationale, limitations, and review/sign-off.
6. Validate privacy, de-identification, citation completeness, and whether all recommendations are traceable to dated sources.

References

• Read references/source-notes.md before citing guidelines, labels, or standards.
• Read references/evidence-checklist.md when grading evidence, writing recommendation strength, or auditing safety language.
• Read references/legacy-full-skill.md only when you need the older LaTeX layout patterns, figure suggestions, or report-structure examples.

Validation

• Every clinical recommendation has a source, date, population boundary, and uncertainty statement.
• Patient-level identifiers are absent unless the user explicitly provides de-identified material.
• Budget warnings about this domain are context warnings only; they do not block installing the complete medical domain.