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

How-To

Quick-Start Workflow

The five-minute version of the daily workflow:

1. Open CareAssist and go to the patient census (the list of currently admitted patients).

2. Filter or sort to find the patients who need attention — typically the ones where CareAssist's recommended classification differs from the patient's current status, or where the status is uncertain.

3. Open a patient to see the evaluation: classification recommendation, severity and intensity scores, dimension-by-dimension findings, and the day-by-day narrative.

4. Review the evidence. For each finding that drives the recommendation, look at the supporting chart references and confirm they hold up.

5. Decide what to do — accept and document the recommendation, escalate to a Physician Advisor, request additional documentation from the treating team, or push back on the AI's reasoning if your read of the chart disagrees.

6. Action it — generate the PDF report, write the recommendation into the criteria review in Epic (when available for your site), or hand off to the appropriate person.

The rest of this section walks through each of these steps in detail.

1. Accessing CareAssist

CareAssist is a web-based application. Your administrator will provide the login URL and your credentials. Once signed in, you'll land on your facility's patient census view.

If your hospital has the Epic integration enabled, you may also be able to access CareAssist's status recommendation from inside Epic's criteria review workflow — see the section on Writing a Status Recommendation into Epic below.

Note: The exact login URL and SSO behavior vary by hospital deployment. Check with your internal CareAssist administrator or your IT support contact.

2. Navigating the Patient Census

The patient census is the central list of currently admitted patients at your facility. Each row represents one encounter and shows:

• Patient identifier

• Admission details (date/time of admission, admitting diagnosis or chief complaint, attending physician)

• Payer

• Current status (e.g., inpatient, observation)

• CareAssist's recommended classification

• A flag/indicator when the recommendation differs from the current status — these are your priority cases

How to work the list:

• Filter by payer, by current status, by ADT unit (campus or facility, useful when your health system spans multiple sites), or by mismatch state.

• Sort by severity score, intensity score, length of stay, or admission date to prioritize where to start.

• Open any patient row to drill into the full evaluation.

Note: The exact filters and columns available in the census view may vary based on your deployment configuration. If something described here isn't visible in your view, ask your administrator about display configuration.

Scoping the census to your unit

If your health system has multiple campuses, facilities, or units, CareAssist can scope the census view (and its automation rules) to the ADT units that belong to your team. This is controlled at the configuration level — work with your administrator to ensure your view shows the right population.

3. Reviewing a Patient

Open any patient from the census to see CareAssist's evaluation. The patient view shows you everything the model has produced for this encounter.

Reading the classification recommendation

At the top of the patient view, you'll see:

• Recommended classification — Inpatient, Outpatient, or Observation

• Severity of Illness score

• Intensity of Services score

These three together give you the headline read on the case. A high SI plus a high IS is a strong inpatient case. A low SI plus a low IS is a strong outpatient case. Anything in between — especially with disagreement between the axes — is a case worth reading carefully.

Reading the dimension-by-dimension breakdown

Below the headline, each rubric's finding is shown — Respiratory, Infection, Laboratory, Diagnostic, Hemodynamic, Neurological, Risk of Complications, Medications, Procedures, Monitoring. For each one, you'll see:

• The finding for that dimension (what the model concluded)

• The supporting evidence pulled from the chart (specific lab values, vital sign trends, notes, orders, MAR entries)

• The reasoning that connects the evidence to the finding

Your job here is validation. Read the evidence for each high-impact dimension. Confirm that what the model is citing is actually in the chart and that it's been interpreted reasonably. Most of the time it will hold up. When it doesn't, that's a signal — either a documentation gap (the chart doesn't support the finding the model made) or an actual disagreement that's worth raising.

Reading the day-by-day narrative

The day-by-day summary walks through the encounter chronologically — admission, day one, day two, and so on. It's framed in the context of the classification: how the clinical picture has progressed, what has changed, what is still in play. This is the section to read when you need to tell the story of the stay — to a Physician Advisor, in a peer-to-peer, or in a written appeal.

Reading the recommendation in context

Three checks to run before you act on the recommendation:

1. Does the evidence support the call? If the model says "inpatient" because of high severity in the respiratory dimension, can you point to the labs, vitals, and notes that justify it? If yes, the case is defensible. If no, work back to either better documentation or a different recommendation.

2. Are there findings the model missed? CareAssist works from the structured and free-text content in the chart. If a clinical reality exists outside the chart — a phone conversation with a consultant, a documented decision that hasn't been written up yet — the model can't see it. Add what's missing (see Adding Clinical Records Manually) and re-run if needed.

3. Does the day-by-day narrative match your read? If the trajectory the model describes feels off, dig in. The narrative is a useful sanity check on whether the model is reading the encounter the way you would.

4. Adding an Admission Manually

Most of the time, admissions flow into CareAssist automatically through the integration with your hospital's source systems. Sometimes — early in a deployment, when the integration hasn't picked up a case, or when you need to review a patient who isn't in the integrated feed — you'll need to add an admission manually.

To add an admission manually:

1. From the census view, choose the option to add an admission.

2. Enter the admission details: patient identifier, admission date/time, attending physician, admitting diagnosis or chief complaint, payer, ADT unit (if applicable).

3. Save the admission. It will appear in your census view.

Once the admission exists in CareAssist, you'll typically need to add clinical records so the evaluation has data to work from. See Adding Clinical Records Manually below.

Note: Manually-added admissions behave the same way as integrated ones for evaluation purposes — the rubrics run against whatever clinical data is present.

5. Adding Clinical Records Manually

Whether an admission came in via integration or you added it manually, you can supply additional clinical records by hand. This is useful when:

• The integration hasn't pulled in the latest notes, labs, or orders yet, and you need the evaluation to reflect them

• The encounter is in an unintegrated state and you need to provide the clinical content directly

• You want to add a record that wouldn't otherwise be in the feed (e.g., a printed consultation report)

To add clinical records to an admission:

1. Open the admission in the patient view.

2. Choose the option to add a clinical record (or equivalent label in your deployment).

3. Provide the record content — paste text, attach a file, or follow the upload flow your deployment supports.

4. Save. The record is now associated with the encounter and will be considered the next time the evaluation runs.

After adding records, you may want to trigger a re-evaluation so the recommendation reflects the new data.

6. Marking an Admission as Discharged

When a patient is discharged, the encounter closes. In most deployments, the discharge will flow in through the integration automatically. When it doesn't — or when you're working with a manually-added admission — you can mark the discharge by hand.

To mark an admission as discharged:

1. Open the admission in the patient view.

2. Choose the option to mark as discharged.

3. Confirm the discharge date.

Discharged admissions stop receiving new evaluations and drop out of the active census view. The encounter's history and the final evaluation remain available for reference.

7. Generating and Using the PDF Report

For every evaluation, CareAssist produces a complete PDF write-up designed for use outside the application — to attach to a criteria review, share with a Physician Advisor, hand off to denial management, or use as supporting documentation in a peer-to-peer.

To generate the PDF:

1. Open the patient view.

2. Select the Generate Report action (or equivalent).

3. The PDF is produced and made available to download or share.

What's in the PDF:

• Patient and encounter summary

• Classification recommendation

• Severity and intensity scores

• Each rubric's finding with supporting evidence

• The day-by-day narrative

• Citations to the chart

Note: Use the PDF as the canonical artifact when handing off the case. The in-app view is for your daily working surface; the PDF is the durable, shareable record of the evaluation.

8. Writing a Status Recommendation into Epic

If your hospital is on Epic and the integration is enabled, you can write CareAssist's status recommendation directly into the criteria review in Epic — the same object MCG and InterQual create when they post their criteria results.

This is the first step of CareAssist's Epic integration. Practically, it means the rest of your UM workflow inside Epic can see CareAssist's call alongside (or instead of) the incumbent criteria tool's call.

To write a status recommendation into Epic:

1. Open the patient view in CareAssist.

2. After reviewing and validating the recommendation, choose the option to send to Epic (or equivalent).

3. Confirm the action. CareAssist writes a criteria review to the encounter in Epic with its recommendation and supporting analysis.

Note: This is a one-direction write today — CareAssist posts to Epic; Epic doesn't post back into CareAssist. Deeper bidirectional integration is on the roadmap.

9. Working a Middle-Ground Case End-to-End

This walkthrough puts the pieces together for the kind of case that takes the most time and where CareAssist provides the most leverage.

Scenario. A patient was admitted yesterday with chest pain and a moderately elevated troponin. The attending placed them as inpatient. The payer has flagged the case as observation-eligible. The treating team is unsure. You have an hour before the payer wants a clinical update.

Step 1 — Open the case. Find the patient in the census; the mismatch flag is likely on. Open the evaluation.

Step 2 — Read the headline. Look at the recommended classification and the SI/IS scores. Suppose CareAssist recommends inpatient, with moderate severity (driven by the hemodynamic and laboratory dimensions) and moderate intensity (driven by medications and monitoring).

Step 3 — Validate the evidence. Scan the hemodynamic finding — what specifically is it citing? The troponin trajectory, any rhythm changes, any blood pressure concerns. Cross-check against the chart. Same for the laboratory finding (which labs, which thresholds), the medications finding (what's on the MAR, including any IV cardiac medications), and the monitoring finding (telemetry orders, frequency of vitals).

Step 4 — Read the narrative. Walk through the day-by-day summary. Does the trajectory support continued inpatient care? Are there discharge readiness signals?

Step 5 — Decide what to do.

• If the evidence holds up cleanly, you have a defensible inpatient case. Generate the PDF, write the recommendation into the criteria review in Epic, and use the narrative to prep for the payer call.

• If the evidence is thin in one area, identify the gap. Reach out to the treating team for documentation, or surface the case to a Physician Advisor for a clinical second opinion.

• If the evidence frankly doesn't support inpatient, you can use the analysis to coach the treating team on observation status — and produce a defensible observation case instead.

Step 6 — Document and act. Whatever you decide, the evaluation and its evidence are your audit trail. The PDF is the artifact you hand off.

10. Escalating to a Physician Advisor

CareAssist doesn't replace the Physician Advisor. It changes the shape of the escalation. Cases that previously went to a PA as raw chart reads now go to them pre-packaged — with the AI's classification, the supporting evidence, and the dimension-by-dimension reasoning all already structured.

When to escalate:

• The recommendation conflicts with the attending's documentation in a way you can't resolve with the chart

• The case sits squarely in the middle ground (one axis high, one low) and the financial stakes warrant clinical review

• The payer is likely to push back hard and you want clinical backup for the peer-to-peer

• The clinical picture is genuinely ambiguous and you want a physician's read before acting

How to escalate (workflow varies by site):

• Generate the PDF report

• Send or share it with your Physician Advisor along with a short note describing what you'd like them to weigh in on

• Use the evaluation as the basis for the conversation — not the whole case to re-litigate

The point of the PA review is to add clinical judgment on top of the structured analysis, not to redo the analysis from scratch.

11. When the Recommendation Disagrees with Your Read

CareAssist's recommendation is a starting point. It's not the final word. When your clinical read disagrees with the AI:

1. Identify which dimension is driving the disagreement. Look at the rubric findings. If you believe the patient is sicker than the model suggests, which dimension is underweighting the severity?

2. Look at the evidence. Is the model missing something — a note, a lab, a finding that isn't in the chart yet? If yes, see Adding Clinical Records Manually.

3. Check the reasoning. Is the model interpreting a finding in a way you'd interpret differently? Make that interpretation explicit when you document your call.

4. Document your override. Your call is the call, and your documentation should reflect both what the AI found and why your judgment differs. That's defensible documentation — exactly the kind that holds up under payer scrutiny.

12. Daily Patterns That Work Well

A few patterns experienced users have settled into:

• Start the day with the mismatch list. Patients where CareAssist's recommendation differs from the current status are usually the highest-leverage starting point.

• Read the narrative before the evidence on cases where the trajectory matters. For continued-stay reviews, the day-by-day story tells you more than the static snapshot.

• Use the PDF as the handoff artifact. Don't paraphrase the analysis when handing a case to a PA or appeals team — share the PDF. It's the durable, citable record.

• Re-check on long stays. As a stay progresses, the recommendation will evolve as new findings land. A patient who started as observation-eligible may have crossed into clear inpatient by day two — and the reverse is also true.

• Pay attention to the MAR-driven findings. The medications dimension is a strong signal of inpatient-level intensity, particularly for IV therapies. When the MAR data is rich, the case for inpatient is usually stronger; when the patient is on routine PO medications only, it's usually weaker.

Getting Help

• Application bugs or issues — contact your hospital's CareAssist administrator or designated support contact.

• Clinical disagreements with the model's reasoning — flag through your administrator; feedback feeds the ongoing improvements to the level-of-care model.

• Workflow questions or training — your UM leadership or internal CareAssist champion can usually answer or point you to someone who can.

Glossary

Acuity Evaluation — The assessment of severity of illness and intensity of service used to determine appropriate level of care. CareAssist performs this evaluation across multiple clinical domains.

ADT Unit — A facility, campus, or unit identifier sourced from the hospital's ADT (Admission, Discharge, Transfer) feed. Used in CareAssist to scope the census view and automation rules to specific parts of a multi-facility health system.

Classification Recommendation — CareAssist's recommendation on whether a patient should be classified as inpatient, outpatient, or observation, based on the acuity evaluation.

Concurrent Review — Evaluating a patient's status while they are still admitted, to ensure ongoing care meets medical necessity requirements. CareAssist supports concurrent review by continuously re-evaluating the encounter as new data arrives.

Criteria Review — The object in Epic where the result of a clinical-criteria evaluation (MCG, InterQual, or now CareAssist) is posted. The criteria review is the standard surface for status determinations in the Epic UM workflow.

Encounter — The time window during which a patient is admitted for one specific reason. CareAssist scopes its evaluation to the current encounter.

Intensity of Services (IS) — One of the two axes CareAssist evaluates. Reflects how resource-intensive the care being provided is.

InterQual / MCG — The two dominant clinical-criteria tools used in UM today. CareAssist provides a clinically deeper, more explainable alternative; future releases will fold MCG/InterQual outputs in alongside CareAssist's analysis where useful.

MAR (Medication Administration Record) — The record of medications actually administered to the patient in the hospital setting, distinct from prescriptions a patient takes themselves. A key input to the medications-intensity rubric.

Medical Necessity — The payer-defined standard a service must meet to be considered reimbursable. The standard CareAssist's classification recommendation is built to defend.

Observation / Outpatient Status — A billing classification (Medicare Part B) for a patient who, while occupying a hospital bed and receiving hospital care, is technically an outpatient. Distinct from inpatient (Part A) in how the stay is billed.

Peer-to-Peer (P2P) — A direct clinical conversation between the treating physician (or Physician Advisor) and a payer's medical reviewer, used to overturn a status denial before formal appeals.

Physician Advisor (PA) — A physician who supports the UM program by reviewing complex cases, conducting peer-to-peers, and providing clinical oversight of status determinations.

Rubric — A discrete clinical question scoped to one dimension of acuity (e.g., respiratory criteria, hemodynamic criteria, monitoring). Each rubric runs independently against the encounter's chart and produces a finding with supporting evidence.

Severity of Illness (SI) — One of the two axes CareAssist evaluates. Reflects how medically unwell the patient is.

UR Nurse (Utilization Review Nurse) — A registered nurse working in utilization management, typically conducting the first-pass clinical review on admitted patients and managing payer communication. The primary daily user of CareAssist.