Clinical Guidelines to LLM Implementation Framework

ASCVD Risk Assessment and Management

🔍 Project Overview

This framework converts clinical guidelines into structured prompts and Multiple Choice Prompts (MCPs) for Large Language Models (LLMs) to follow in clinical decision support. This proof-of-concept implementation focuses on Atherosclerotic Cardiovascular Disease (ASCVD) risk assessment and management guidelines.

🎯 Objectives

• Convert clinical guidelines into step-by-step prompts for LLMs
• Classify steps requiring external tools versus reasoning
• Create reusable MCPs for standardized decision making
• Demonstrate effectiveness through a sample case walkthrough

📊 ASCVD Guideline Breakdown

Workflow Classification

Step	Process	Type
1. Patient Data Collection	Gather demographic and clinical variables	External Tool
2. ASCVD Risk Calculation	Calculate 10-year risk percentage	External Tool
3. Risk Stratification	Categorize risk level	Reasoning
4. Assess Risk Enhancers	Evaluate additional risk factors	Reasoning
5. Treatment Recommendation	Determine appropriate intervention	Reasoning
6. Follow-up Planning	Establish monitoring schedule	Reasoning

💻 LLM Implementation Strategy

Step 1: Patient Data Collection

# Patient Data Collection Prompt

You are assisting with ASCVD risk assessment. Please systematically collect:

1. **Age** (in years)
2. **Sex** (Male/Female)
3. **Race** (White, African American, Other)
4. **Total Cholesterol** (mg/dL)
5. **HDL Cholesterol** (mg/dL)
6. **Systolic Blood Pressure** (mmHg)
7. **BP-lowering medication status** (Yes/No)
8. **Diabetes status** (Yes/No)
9. **Smoking status** (Yes/No)

For each parameter, provide a brief explanation of its relevance and validate responses within appropriate clinical ranges.

Step 2: Risk Calculation

# ASCVD Risk Calculation Prompt

Based on the collected patient data, calculate the 10-year ASCVD risk:

**[EXTERNAL TOOL INTEGRATION]**
- Input all collected parameters into validated ASCVD Risk Calculator
- Return precise 10-year risk percentage

Parameters:
- Age: {age}
- Sex: {sex}
- Race: {race}
- Total Cholesterol: {total_cholesterol} mg/dL
- HDL Cholesterol: {hdl} mg/dL
- Systolic BP: {systolic_bp} mmHg
- BP medication: {bp_med}
- Diabetes: {diabetes}
- Smoker: {smoker}

Step 3: Risk Stratification MCP

# Risk Stratification MCP

With the calculated 10-year ASCVD risk of {risk_percentage}%, identify the appropriate risk category:

A) Low risk (<5%)
B) Borderline risk (5-7.4%)
C) Intermediate risk (7.5-19.9%)
D) High risk (≥20%)

Select the single best option according to established guidelines.

Step 4: Risk Enhancers Assessment

# Risk Enhancers Assessment Prompt

For this {risk_category} risk patient, evaluate each risk enhancer as "Present," "Absent," or "Unknown":

1. **Family history** of premature ASCVD (males <55 years, females <65 years)
2. **Persistently elevated LDL-C** ≥160 mg/dL
3. **Chronic kidney disease** (eGFR 15-59 mL/min/1.73 m²)
4. **Metabolic syndrome**
5. **Inflammation** (elevated hsCRP ≥2.0 mg/L)
6. **Persistently elevated triglycerides** (≥175 mg/dL)
7. **Premature menopause** (if applicable)
8. **Ethnicity-specific risk factors**

Summarize with total count: {number_present} present, {number_absent} absent, {number_unknown} unknown

Step 5: Treatment Recommendation MCP

# Treatment Recommendation MCP

**Patient Risk Profile:**
- 10-year ASCVD Risk: {risk_percentage}% ({risk_category})
- Risk Enhancers: {number_present} present, {number_absent} absent, {number_unknown} unknown

Based on this comprehensive assessment, select the most appropriate initial management:

A) **Lifestyle modifications only** (healthy diet, regular exercise, weight management, smoking cessation if applicable)
B) **Lifestyle modifications + consider moderate-intensity statin** after shared decision making
C) **Lifestyle modifications + initiate moderate-intensity statin therapy**
D) **Lifestyle modifications + initiate high-intensity statin therapy**

Select one option and provide a brief evidence-based rationale.

Decision Logic Map for Treatment

graph TD
    A[Calculate ASCVD Risk] --> B{Risk Category}
    B -->|<5%| C[Low Risk]
    B -->|5-7.4%| D[Borderline Risk]
    B -->|7.5-19.9%| E[Intermediate Risk]
    B -->|≥20%| F[High Risk]
    
    C --> G[Lifestyle Modifications Only]
    
    D --> H{Risk Enhancers?}
    H -->|≥2 Present| I[Consider Moderate Statin]
    H -->|<2 Present| G
    
    E --> J[Moderate-Intensity Statin]
    F --> K[High-Intensity Statin]

Step 6: Follow-up Recommendations MCP

# Follow-up Recommendations MCP

For this {risk_category} patient on {recommended_therapy}, select the appropriate follow-up plan:

A) Reassess lipid panel and risk factors in **3-6 months**
B) Reassess lipid panel and risk factors in **6-12 months**
C) No specific follow-up needed, reassess at next **routine visit**
D) Referral to **cardiology** for advanced risk assessment

Select one option according to current clinical practice guidelines.

📋 Implementation Workflow

1. Data Collection → Use Step 1 prompt to gather all necessary patient information
2. Risk Calculation → Apply external calculator (Step 2) to determine 10-year risk
3. Risk Categorization → Apply Step 3 MCP to stratify risk level
4. Risk Enhancement → Use Step 4 prompt to identify additional risk factors
5. Treatment Planning → Apply Step 5 MCP with decision logic to generate recommendations
6. Follow-up Planning → Apply Step 6 MCP to establish monitoring schedule

🧪 Case Study Example

Patient Profile:

• 55-year-old White male
• Total Cholesterol: 220 mg/dL
• HDL: 42 mg/dL
• Systolic BP: 138 mmHg
• On BP medication: Yes
• Diabetes: No
• Smoker: Yes

Risk Assessment:

• ASCVD Risk Calculation: 13.6% → Intermediate risk
• Risk Enhancers:
  • Family history: Present
  • Elevated LDL: Present (calculated LDL ~156 mg/dL)
  • Other factors: Unknown

Clinical Decision:

• Treatment Recommendation: Option C (Lifestyle modifications + moderate-intensity statin)
• Follow-up Plan: Option A (Reassess in 3-6 months)

🔄 Future Expansion

1. Additional Guidelines - Implement framework for other clinical algorithms
2. Enhanced Integration - Connect with EHR systems for automated data retrieval
3. Validation Studies - Compare LLM recommendations with clinician decisions
4. User Interface - Develop clinician-friendly front-end for real-time use
5. Patient Communication - Create complementary prompts for explaining results to patients

📈 Expected Benefits

• Standardized application of clinical guidelines
• Reduced clinician cognitive load
• Enhanced documentation of clinical reasoning
• Support for shared decision making
• Potential for integration with existing clinical workflows