Risk Stratification Indices

Risk Stratification Indices Reasoning

Overview

Guide selection and interpretation of clinical risk stratification indices (LACE, Charlson, Elixhauser) and SDOH factors (Z-codes, ADI) for population health management. Based on validated scoring systems from peer-reviewed literature.

Usage

• Activate when selecting a risk stratification index for a specific use case (readmission, mortality, utilization)
• Activate when interpreting LACE, Charlson, or Elixhauser scores
• Activate when incorporating SDOH factors (Z-codes, ADI) into population health stratification

Core Concepts

Response Format

Apply scoring logic internally. Present the recommended index with justification, scoring interpretation, and risk tier. Do not narrate the full scoring calculation unless asked.

1. Choosing a Risk Stratification Method

What is the use case?
├── 30-day readmission prediction
│   └── Use LACE index (designed for readmission)
├── Long-term mortality risk adjustment
│   └── Use Charlson Comorbidity Index
├── Hospital resource utilization / cost prediction
│   └── Use Elixhauser Comorbidity Index
├── Medicare risk adjustment (payment)
│   └── Use CMS-HCC (Hierarchical Condition Categories)
└── Population health stratification
    └── Combine clinical risk (Charlson/Elixhauser) + SDOH factors

2. LACE Index (Readmission Risk)

| Component | Scoring | Range | |-----------|---------|-------| | L — Length of stay | 1d=1, 2d=2, 3d=3, 4–6d=4, 7–13d=5, ≥14d=7 | 0–7 | | A — Acuity of admission | Emergent=3, Urgent=2, Elective=0 | 0–3 | | C — Comorbidity (Charlson) | 0=0, 1=1, 2=2, 3=3, ≥4=5 | 0–5 | | E — ED visits (prior 6 months) | 0=0, 1=1, 2=2, 3=3, ≥4=4 | 0–4 | | Total | Sum of L+A+C+E | 0–19 |

Risk tiers: Low (0–4), Moderate (5–9), High (10+)

3. Charlson Comorbidity Index

17 conditions with integer weights:

• Weight 1: MI, CHF, PVD, CVD, dementia, COPD, connective tissue disease, peptic ulcer, mild liver disease, uncomplicated diabetes
• Weight 2: complicated diabetes, hemiplegia, renal disease, non-metastatic cancer
• Weight 3: moderate/severe liver disease
• Weight 6: metastatic tumor, AIDS/HIV

Total score = sum of all applicable weights.

4. Elixhauser Comorbidity Index

| Attribute | Charlson | Elixhauser | |-----------|----------|------------| | Conditions | 17 | 31 | | Weighting | Fixed integer | Varies by model (van Walraven common) | | Scope | Mortality prediction | Mortality + resource use | | Mental health | Limited (dementia only) | Depression, psychoses, substance use | | Best for | Long-term mortality | Hospital utilization, readmission |

5. SDOH Risk Factors

ICD-10 Z-Codes for SDOH

Key ranges: Z55 (education/literacy), Z56 (employment), Z57 (occupational exposure), Z59 (housing/economic — Z59.0 homelessness, Z59.41 food insecurity), Z60 (social environment), Z62 (upbringing/abuse), Z63 (family circumstances), Z65 (psychosocial/legal).

Area Deprivation Index (ADI)

| ADI Percentile | Risk Level | Implication | |---------------|------------|-------------| | 1–25 | Low deprivation | Standard outreach sufficient | | 26–50 | Moderate | Enhanced reminder systems | | 51–75 | High | Care coordination, transportation assistance | | 76–100 | Very high | Intensive outreach, community health workers |

Incorporating SDOH into Stratification

Weight members higher when:

1. Any SDOH Z-code documented in claims
2. High ADI score (≥51st percentile)
3. Dual-eligible status (Medicare + Medicaid)
4. Language barrier (non-English preferred)
5. No PCP visit in 12 months (care disengagement)

Common Mistakes

• Wrong: Using Charlson for readmission prediction → Right: Use LACE (purpose-built for 30-day readmission)
• Wrong: Using Elixhauser for long-term mortality → Right: Use Charlson (validated for mortality prediction)
• Wrong: Ignoring SDOH factors in population stratification → Right: Combine clinical index + SDOH for actionable tiers
• Wrong: Applying same outreach to all risk tiers → Right: Tailor interventions by barrier type (transportation, literacy, engagement)
• Wrong: Using ADI alone without clinical risk → Right: ADI indicates access barriers; combine with clinical severity for full picture
• Wrong: Treating all Z-codes as equivalent risk signals → Right: Weight by relevance to the specific outcome (e.g., Z59.0 homelessness is higher barrier than Z56 employment)

When to Escalate

• When risk scores drive reimbursement or payment decisions (requires validated, audited implementation)
• When stratification results will determine resource allocation across populations
• When combining indices in novel ways not validated in literature

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: awslabs
• Source: awslabs/hcls-agent-skills
• License: MIT-0

Install and usage instructions live in the source repository linked above.