Clinical Decision Support Studio

A scientific/medical skill for generating professional clinical decision support (CDS) documents. Clinical Decision Support Studio helps create evidence-based clinical guidelines, decision algorithms, drug interaction alerts, and diagnostic support tools used by healthcare IT systems and pharmaceutical companies.

When to Use This Skill

Choose Clinical Decision Support Studio when:

• Creating clinical decision algorithms and flowcharts
• Writing evidence-based treatment guidelines with recommendation levels
• Building drug interaction alert logic for clinical systems
• Developing diagnostic decision trees with sensitivity/specificity data

Consider alternatives when:

• You need clinical trial data (use ClinicalTrials.gov tools)
• You need patient-level medical records (use EHR-specific tools)
• You're doing basic medical literature search (use PubMed)
• You need regulatory submission documents (use regulatory-specific tools)

Quick Start

Copy
claude "Create a clinical decision support algorithm for Type 2 Diabetes management"

Copy
# CDS Algorithm: Type 2 Diabetes Initial Management

## Evidence Level: Grade A (Strong Recommendation)
**Source:** ADA Standards of Medical Care 2024, NICE NG28

## Decision Tree

### Step 1: Confirm Diagnosis
- HbA1c ≥ 6.5% on two separate occasions
- OR Fasting plasma glucose ≥ 126 mg/dL
- OR 2-hour plasma glucose ≥ 200 mg/dL (OGTT)

### Step 2: Initial Assessment
| Parameter | Target | Alert Threshold |
|-----------|--------|-----------------|
| HbA1c | < 7.0% | ≥ 9.0% (consider dual therapy) |
| Blood Pressure | < 130/80 mmHg | ≥ 140/90 mmHg |
| LDL Cholesterol | < 100 mg/dL | ≥ 190 mg/dL |
| eGFR | > 60 mL/min | < 30 mL/min (nephrology referral) |

### Step 3: First-Line Therapy
IF HbA1c < 9.0%:
  → Metformin 500mg BID, titrate to 1000mg BID
  → Lifestyle modifications

IF HbA1c ≥ 9.0%:
  → Metformin + second agent (SGLT2i if eGFR > 30, or GLP-1 RA)
  → Consider insulin if symptomatic hyperglycemia

Core Concepts

CDS Document Types

Type	Purpose	Example
Decision Algorithm	Step-by-step clinical pathway	Diabetes management ladder
Alert Rule	Automated warning logic	Drug interaction alert
Order Set	Pre-defined treatment orders	Sepsis bundle orders
Reminder	Preventive care prompt	Annual screening reminders
Infobutton	Context-sensitive reference	Drug dosing reference

Evidence Grading

Copy
## GRADE System for Recommendations

| Grade | Strength | Meaning |
|-------|----------|---------|
| 1A | Strong, high quality | Benefits clearly outweigh risks |
| 1B | Strong, moderate quality | Benefits likely outweigh risks |
| 1C | Strong, low quality | Benefits appear to outweigh risks |
| 2A | Weak, high quality | Balance of benefits and risks close |
| 2B | Weak, moderate quality | Balance uncertain |
| 2C | Weak, low quality | Very uncertain benefit-risk balance |

Alert Logic Structure

Copy
{
  "alert_id": "DI-001",
  "alert_type": "drug_interaction",
  "severity": "high",
  "trigger": {
    "condition": "AND",
    "rules": [
      {"medication_class": "SSRI", "status": "active"},
      {"medication_class": "MAOI", "status": "prescribing"}
    ]
  },
  "message": "CRITICAL: Serotonin syndrome risk. SSRIs and MAOIs are contraindicated. Washout period of 14 days required.",
  "override_reason_required": true,
  "evidence_level": "1A",
  "references": ["FDA Safety Alert 2023-XX"]
}

Configuration

Parameter	Description	Default
evidence_system	GRADE, Oxford, SORT	GRADE
output_format	Markdown, JSON, HL7 FHIR	markdown
guideline_source	ADA, NICE, WHO, ESC	Required
alert_severity_levels	High, moderate, low	[high, moderate, low]
include_references	Cite supporting evidence	true

Best Practices

1. Always cite evidence levels and sources. Every clinical recommendation must include its evidence grade and the guideline source. Ungraded recommendations risk being applied inappropriately and create liability concerns.

2. Design alerts with override workflows. Not every alert should be a hard stop. High-severity alerts (drug contraindications) should require documented override reasons, while low-severity alerts (dose adjustments) can be informational.

3. Include monitoring parameters. Decision support shouldn't end at treatment initiation. Include follow-up intervals, lab monitoring schedules, and escalation criteria so clinicians know when to reassess.

4. Use standard terminologies. Map medications to RxNorm, diagnoses to ICD-10, and labs to LOINC codes. Standard terminologies enable interoperability between clinical systems and consistent alert triggering.

5. Validate with clinical stakeholders. CDS content must be reviewed by practicing clinicians before deployment. Even evidence-based guidelines require local adaptation for formulary differences, patient populations, and available resources.

Common Issues

Alert fatigue from too many low-priority alerts. Reduce noise by tuning alert sensitivity — suppress alerts for known, documented combinations and reserve hard-stop alerts for truly dangerous situations. Track override rates: if > 90% of an alert type is overridden, it needs redesign.

Guidelines conflict between different medical societies. When ADA and NICE recommendations differ, document both perspectives and let the clinical team choose based on their practice context. Flag conflicts explicitly rather than silently choosing one guideline.

CDS logic doesn't account for special populations. Pediatric, geriatric, pregnant, and renally impaired patients require modified thresholds. Build population-specific branches into decision algorithms and clearly label which population each recommendation applies to.