Audience: utilization review nurses, case managers, and other clinical end users.

Introduction

What CareAssist Is

CareAssist is an AI-powered utilization management application built specifically for hospital UM teams. It reviews each admitted patient's clinical record, evaluates how sick they are and how intensively they're being treated, and recommends whether their stay should be classified as inpatient or observation/outpatient. Alongside that recommendation, it produces the clinical narrative and supporting evidence you need to defend the decision — to a Physician Advisor, to a payer on a peer-to-peer, or in an eventual appeal.

In practical terms, CareAssist is designed to do the first-pass chart review that a UR nurse or Physician Advisor would otherwise do manually — except it does it for every patient, on a continuous cadence, and produces a structured, defensible write-up you can read in minutes instead of reconstructing the picture from raw notes yourself.

CareAssist is not a replacement for clinical judgment. The classification it recommends is a starting point — a structured argument grounded in the chart — that you review, validate, and either accept, refine, or escalate.

What CareAssist Is For

CareAssist is built to help with three specific problems UM teams face every day:

1. Time pressure. A thorough manual review of a complex admission can take 30 minutes or more. With caseloads of 20–40 concurrent patients, there usually isn't time for that on every case. CareAssist compresses the first pass to minutes, so you can spend your time on the parts of the review that need a human — validating the AI's reasoning, gathering additional context, and engaging with the treating team or payer.

2. The middle ground. Clearly-inpatient stays (heart transplant, prolonged ICU care) and clearly-outpatient stays (short ED visits) rarely generate disputes. The disagreements with payers — and the revenue at stake — live in the middle: short stays, observation-eligible presentations, and admissions where severity and intensity are moderate. CareAssist is engineered to reason about that middle ground, and to surface the winnable cases that would otherwise get triaged superficially.

3. Defensible documentation. The quality of the documentation produced during concurrent review directly determines how defensible the case is if the payer pushes back. CareAssist generates a structured clinical rationale — with specific findings, references to the chart, and alignment to medical necessity criteria — that can be attached to a criteria review, used to prep for a peer-to-peer, or handed off to the appeals team.

What CareAssist Does Not Do

A few things worth being explicit about:

• CareAssist does not change clinical care. The status recommendation is financial and administrative. Care decisions remain with the treating team, made on the patient's needs alone. This is the same firewall between clinical care and financial classification that already exists in your UM program.

• CareAssist is not a denial-management or appeals tool. Once a classification is denied by a payer and you're moving into formal appeals, that work lives in a sibling product, AppealAssist. CareAssist's job ends at the concurrent UM decision and the documentation that supports it.

• CareAssist does not pull longitudinal multi-month history. Its evaluation is scoped to the current encounter — the admission you're reviewing right now. Historic data matters only when it informs the current clinical picture.

How CareAssist Reasons About a Patient

Understanding the model behind CareAssist's recommendation makes the application much easier to use — because every part of the output maps back to it. The reasoning has three layers.

The Encounter

CareAssist evaluates patients in the context of a single encounter — the time window during which a patient is admitted for one specific reason. An encounter starts at admission and ends at discharge. An inpatient stay that lasts a week is one encounter; an ED visit that turns into an admission is typically one encounter as well.

Everything CareAssist pulls — admission notes, progress notes, vitals, labs, imaging, medications ordered, procedures, consultations — is scoped to that encounter. The product is, by design, about the current situation: how sick is this patient right now, and how intensively are they being treated right now.

The Two Axes: Severity of Illness and Intensity of Services

Every CareAssist evaluation looks at a patient along the same two axes a Physician Advisor would:

• Severity of Illness (SI) — how medically unwell the patient is. Are vitals stable or dangerous? Are there active, dangerous conditions? Are labs abnormal in ways that matter?

• Intensity of Services (IS) — how resource-intensive the care being provided is. IV medications? Continuous monitoring? Procedures? Surgery?

The combination determines the appropriate level of care:

• High SI + High IS → clearly inpatient. Sick patient, intensive care.

• Low SI + Low IS → clearly outpatient. Stable patient, minimal intervention.

• High on one, low on the other → the middle ground. This is where the product earns its keep, and where most of the cases you'll actually use CareAssist to think through will live.

This two-axis framing is the same logic that underpins InterQual and MCG. The difference is how CareAssist applies it: not as a rule-based decision tree, but as the kind of clinical synthesis a Physician Advisor performs — reading the chart, weighing the findings, and reasoning toward a defensible classification.

Clinical Rubrics

The SI and IS axes are evaluated through a set of clinical rubrics — discrete clinical questions, each scoped to one dimension of the patient's condition or care. There are roughly nine to ten rubrics running on each case:

Severity-of-Illness dimensions:

• Respiratory criteria — breathing status, respiratory distress, oxygen requirements, ventilation

• Infection criteria — sepsis, bacteremia, fever, infection severity and treatment response

• Laboratory criteria — lab values indicating organ dysfunction, electrolyte derangements, markers of acute illness

• Diagnostic criteria — what diagnoses are in play; how severe and active they are

• Hemodynamic criteria — blood pressure, heart rate, perfusion, shock indicators

• Neurological criteria — mental status, neurological deficits, seizure activity, altered consciousness

• Risk of complications — factors that elevate the risk of clinical deterioration

Intensity-of-Services dimensions:

• Medications — IV medications, high-risk drugs, treatments that require inpatient administration. The MAR (Medication Administration Record) — the list of medications actually given to the patient in the hospital setting — is a key input here.

• Procedures — surgeries, interventions, or procedures requiring hospital resources

• Monitoring — need for continuous observation, telemetry, frequent reassessment

Each rubric is independent. Each one looks at the chart and produces a finding for that dimension, with supporting evidence pulled from the record. The rubrics are also where CareAssist's clinical expertise lives — they encode how a Physician Advisor would reason about that specific dimension, what findings matter, and what counts as "concerning" versus "stable."

Putting It Together

A CareAssist evaluation runs in roughly this sequence:

1. Encounter assembly — the clinical record for the current encounter is pulled together.

2. Rubric execution — each rubric runs against the record, producing a finding and supporting evidence for its dimension.

3. Scoring and synthesis — the rubric outputs are combined into an overall severity score and intensity score, and a recommended classification is derived.

4. Narrative generation — a day-by-day summary of the encounter (admission → day one → day two → …) is generated, framed in the context of the classification.

5. Report generation — two outputs are produced: an in-app view for quick review, and a fuller PDF report for use in defending the classification with payers.

Because the chart keeps evolving, evaluations don't happen once. CareAssist re-runs on a configurable cadence — typically hourly — so the classification recommendation stays current as new notes, labs, and orders land.

What You'll See — Outputs from a CareAssist Evaluation

Every evaluation produces:

• A classification recommendation — inpatient, outpatient, or observation

• A severity score and an intensity score — quantitative summaries across the rubrics

• A dimension-by-dimension breakdown — what each rubric found, with the specific evidence from the chart

• A day-by-day narrative — how the encounter has progressed

• A UM-facing PDF report — the complete write-up, designed to be handed off or attached as supporting documentation

The in-app view is your daily working surface. The PDF is what you'll attach, share with a Physician Advisor, or use to prep for a payer call.

Where CareAssist Is Heading

CareAssist today is shaped like a report generator: an analysis runs, a structured output is produced, you read it and decide what to do. That's where the product is now, and it's already useful in that form.

Where it's heading is toward real-time decision support — a product that lives alongside you during the moments where the value is actually realized: the ten minutes before you decide to escalate a case, the prep before a peer-to-peer, the live conversation with a payer medical director. The shift over the next year will be toward:

• Interactive querying of the chart and the analysis — asking specific questions in the moment, not just reading a precomputed report.

• Continuous re-evaluation as the chart evolves, so the recommendation always reflects the latest clinical picture.

• Payer-facing argument support — surfacing the strongest evidence on demand, anticipating likely counter-arguments, and helping you map your case to the criteria language payers use.

• MCG/InterQual awareness — pulling the incumbent tools' outputs in alongside CareAssist's analysis, so you can cite them when they agree and prepare for the objection when they don't.

• Deeper Epic integration — more of the workflow happening inside the EMR you already use, not in a separate window.

You don't need to do anything to get ready for those changes. They'll arrive as releases roll out. The clinical reasoning that drives the recommendation will stay the same; the surface you use to engage with it will get progressively richer.