Wedge Product: Inpatient Status Assessment

Decision support for the inpatient vs. observation determination

---

The Problem

When a patient is admitted to the hospital, someone has to decide: inpatient or observation status?

This decision happens at T+4-8 hours, when the physician places the admission order. It has major consequences:

• For the hospital: $2,000-$5,000+ revenue difference per case
• For the patient: Higher cost-sharing under observation, loss of SNF coverage
• For the system: Denials, appeals, rework, adversarial payer-provider dynamics

The gap: No one applies criteria at the decision moment. Physicians decide based on clinical intuition — they often don't understand Two-Midnight Rule or its implications. Case managers apply criteria 14-24 hours later, after the decision is locked. By then, it's retrospective review and damage control.

---

The Solution

Inpatient Status Assessment: AI-powered decision support that brings criteria-based reasoning to the decision point.

How it works:

1. Triggers at T+4-8 hours (when admission order is placed)
2. Pulls available clinical data (vitals, labs, diagnosis, patient factors)
3. Applies Two-Midnight reasoning: "Given this presentation, will this patient need hospital care spanning two midnights?"
4. Produces structured assessment showing:
   • Expected length of stay
   • Factors supporting inpatient vs. observation
   • Documentation suggestions
   • Confidence level based on data completeness

What it's NOT:

• Not a decision-maker — advisory only
• Not a billing tool — clinical reasoning, not coding
• Not InterQual/MCG — transparent reasoning from first principles, not proprietary black box

---

Example Output

Patient: 72-year-old female, pneumonia (J18.9), SpO2 89% on room air, WBC 14.2, fever 38.8°C

STATUS ASSESSMENT
=================
Current Status: Inpatient
Expected LOS: ≥2 midnights likely
Status Alignment: Appropriate
Confidence: High

Supporting Factors:
- Hypoxia requiring supplemental oxygen (SpO2 89% on room air)
- Elevated WBC suggesting active infection (14.2 K/uL)
- Fever indicating systemic infection (38.8°C)
- Age >70 with pneumonia — typically extended stay
- Pneumonia typically requires 3-5 day hospital stay

Documentation Suggestions:
- Document expected LOS (recommend stating "3-4 days anticipated")
- Document oxygen requirement and room air SpO2
- Reference elevated WBC and fever in assessment

Summary: Inpatient status appears appropriate. Clinical picture
strongly supports need for hospital care spanning two or more
midnights. Recommend documenting expected LOS and oxygen
requirement explicitly.

---

Why This Works

Technical Unlock

Two-Midnight Rule provides a tractable question: "Will this span two midnights?" AI can now reason about this using available clinical data. This makes tractable something that previously required expensive clinical expertise in real-time.

Removes Cognitive Burden

Physicians focus on clinical care. The system handles billing logic. No more coaching frontline staff on Two-Midnight implications — the tool encodes it.

Confidence to Classify Appropriately

Hospitals often default to observation defensively — to avoid retrospective audit denials. This gives confidence to classify as inpatient when criteria are met. Serves both hospital revenue and patient interests.

Transparent Reasoning

Unlike InterQual/MCG (proprietary, binary, black-box), this shows why it concluded what it did. What data was considered, how it maps to criteria, where it's clear vs. borderline.

---

Differentiation

---

The Bigger Vision

Phase 1: Provider-Side Decision Support (This Product)

AI-powered criteria assessment at the decision point, for the provider side. Helps physicians and case managers make better status determinations with better documentation.

Phase 2: Documentation Enhancement

Beyond surfacing the assessment, actively help improve documentation:

• Suggest specific language for physician notes
• Auto-generate medical necessity statements
• Pre-populate case management documentation

Phase 3: Shared Payer Visibility

The more radical possibility: what if the payer could see the same assessment?

If both provider and payer are looking at the same AI-generated, criteria-based reasoning at the time of admission:

• Cases that clearly meet criteria → auto-aligned, no dispute
• Cases that are borderline → both parties see it, can discuss
• Cases that don't meet criteria → provider knows upfront, can adjust

This shifts from adversarial retrospective review to shared prospective visibility. The "shared reasoning surface" — moving the conversation from "did this meet criteria?" (after the fact) to "do we agree on the criteria?" (more productive).

---

Why This Is a Strong Wedge

1. Concrete pain — Status misalignment is quantifiable: $2K-$5K per case, appeal costs, patient harm

2. Clear value — Bring criteria expertise to the decision point, not 24 hours later

3. Tractable — Two-Midnight provides a specific, answerable question (vs. fuzzy "optimize UM")

4. Differentiable — Not another InterQual/MCG. Automated, prospective, transparent

5. Expands naturally — Provider decision support → documentation → shared payer visibility

6. Proves the thesis — If we can do this for inpatient/observation, we can do it for other UM workflows

---

Deep Research

Full context scaffold with detailed sections on healthcare economics, utilization management, Two-Midnight Rule, current workflows, systems/data, and skill specification:

/Users/michaelyuan/Code3/research/inpatient-obs/

• context-scaffold.md — Overview of all sections
• sections/01-healthcare-economics.md — Payer-provider dynamics
• sections/02-utilization-management.md — UM as a function
• sections/03-inpatient-observation.md — The status distinction
• sections/04-current-workflow.md — How it works today
• sections/05-systems-and-data.md — Technical integration
• sections/06-intervention-point.md — Where AI fits
• sections/07-the-skill.md — Full skill specification
• design-insights.md — Critical learnings from domain expertise