domains / research / clinical-decision-support-artifacts / clinical-decision-support

clinical-decision-support

Skill card

combined

The K-Dense-AI copy is a distinct, safety-first procedure for generating research-only CDS evaluation and governance artifacts using local Python scripts and strict data gates. The other three copies (Vibe-Skills, aipoch, OpenClaw) are near-identical reproductions of a different skill focused on generating publication-ready LaTeX/PDF documents for pharmaceutical research, including specific formatting and statistical visualization instructions. Because the K-Dense copy provides a unique workflow (script-based validation, data gating, artifact headers) that the others lack, and the others provide unique content (LaTeX/PDF generation, specific pharma use cases) that the K-Dense copy lacks, crowning one would hide a real alternative approach to the same named skill.

Repositories

Other copies

Combined skill

# Clinical Decision Support

## Safety Boundary
This skill produces research, evaluation, and documentation artifacts only. It must never be used to diagnose, treat, or make patient-specific clinical decisions. If a request affects care for a person, stop and route to a licensed healthcare professional.

## Workflow

### 1. Frame the Research Question
- Define the estimand or evaluation target.
- Distinguish descriptive, prognostic, predictive, and causal questions.
- Pre-specify outcomes, subgroups, and statistical methods.

### 2. Select Artifact Type
Choose between:
- **Research/Governance Artifacts**: Intended-use statements, evidence profiles, cohort table shells, survival analysis plans.
- **Pharmaceutical/Research Documents**: Patient cohort analyses, treatment recommendation reports, clinical guidelines.

### 3. Data Gate (For Research Artifacts)
Before generating any data-driven artifact:
1. Confirm input is synthetic or aggregate.
2. Reject patient rows, identifiers, or free text.
3. Keep source files local.
4. Record provenance and data cut date.

### 4. Generate Content
- **For Research Artifacts**: Use local Python scripts (e.g., `validate_cds_artifact.py`, `evidence_profile_check.py`) to validate and generate structured JSON artifacts. Ensure every artifact includes a header with intended use, limitations, and the statement: **Not for patient care or live clinical use.**
- **For Pharmaceutical Documents**: Generate publication-ready LaTeX/PDF files. Include:
  - Biomarker-stratified cohort analyses with statistical outcomes (HR, p-values, CIs).
  - GRADE evidence grading (1A-2C).
  - Treatment algorithm flowcharts (TikZ).
  - Regulatory compliance headers (HIPAA, ICH-GCP).
  - Professional formatting (compact margins, color-coded recommendations).

### 5. Human Review
- Require review by a methodologist/statistician for design.
- Require review by a domain expert for clinical context.
- Require review by a privacy officer for disclosure decisions.
- Outputs remain drafts until approved by qualified humans.