A physician-led clinical decision-support project in inpatient rehabilitation medicine

Inpatient Rehabilitation · 2026 – Present

Rehab Starts at Referral.

Every inpatient rehabilitation admission begins with a referral packet and a deadline. RehabScreen is an AI-driven clinical decision-support platform that reads the referral, drafts a standardized patient summary, and lays the evidence out against admission criteria, so the physiatrist can make a faster, better-documented decision.

The State of Referral Review

The rules are strict. The clock is short.

48 hrs

The preadmission screening must be completed in the 48 hours immediately preceding IRF admission.

3 hrs/day

Patients must generally be able to tolerate and benefit from about three hours of therapy a day, at least five days a week.

60%

At least 60% of an IRF's patients must have one or more of the conditions CMS specifies for inpatient rehabilitation.

13

Qualifying conditions, from stroke and spinal cord injury to major multiple trauma and specified arthritis presentations.

Every admission begins with a referral packet: history and physical, progress notes, therapy evaluations, nursing flowsheets — often dozens of unstructured pages. Someone has to read all of it, judge whether the patient can tolerate intensive therapy and is expected to improve measurably, and document why.

That work happens by hand, under the same discharge pressure acute care is under. Reviewers read packets differently, key details are buried, and the documentation that justifies an admission is rebuilt from scratch every time.

Requirements summarized from the IRF coverage criteria at 42 CFR 412.622 and the IRF classification criteria at 42 CFR 412.29. Simplified for a general audience; see the regulations for exceptions.

The Solution

Read once. Decide well.

Three layers, built in order. The foundation is a tool that reads referrals the same way every time. Prediction comes last, and only once the foundation is sound.

Automated Referral Review

Extracts diagnosis, comorbidities, medical stability, functional status and therapy tolerance from unstructured referral documents, and organizes them around IRF admission criteria.

In development

Standardized Patient Summary

A one-page summary with the same structure every time, so each referral is read the same way by every reviewer, on every shift.

In development

Predictive Analytics

Exploring models for expected length of stay and discharge disposition, to start discharge planning on day one instead of week two.

Exploratory

The physiatrist decides. Always.

RehabScreen drafts and organizes. A rehabilitation physician reviews the screening and makes the admission decision, as CMS requires.

How It Works

Packet in. Decision out.

  1. 01

    Referral received

    Acute-care documents arrive from the referring hospital or case manager.

  2. 02

    Structured extraction

    Diagnosis, comorbidities, function, therapy tolerance and social context are pulled into fixed fields.

  3. 03

    Criteria mapping

    Evidence is laid against IRF coverage criteria, with gaps and missing documentation flagged.

  4. 04

    Summary drafted

    A one-page standardized summary is generated for review.

  5. 05

    Physiatrist decides

    The rehabilitation physician reviews, edits, and accepts or declines the admission.

Sample Summary

One page. Same order. Every time.

Standardization is the point: the fields a physiatrist needs to make and defend an admission decision, in the same place on every referral, each traceable to the document it came from.

Illustration only. Synthetic patient; no real health information.

Preadmission SummarySynthetic example
Impairment group
Stroke, left MCA ischemic, hospital day 6 · CMS-qualifying condition [H&P p.1]
Medical stability
Afebrile, hemodynamically stable, no IV titration [Progress note, day 6]
Current function
Mod assist transfers; min assist gait 25 ft with rolling walker [PT note]
Therapy tolerance
Completed PT and OT sessions same day without rest breaks [OT note]
Disciplines
PT · OT · SLP
Prior level / home
Independent; lives with spouse; 3 steps to enter [Case mgmt note]
Flags for reviewer
Anticoagulation plan not documentedSLP eval pending
Predicted LOS
Model in development
Discharge disposition
Model in development

Roadmap

Built in the open. Step by step.

PHASE 1

Referral extraction and criteria mapping

Turn unstructured referral documents into structured fields aligned to IRF admission criteria.

In progress
PHASE 2

Standardized summary template

Define the one-page summary with physiatrists, liaisons and therapists who review referrals.

In progress
PHASE 3

Validation against physician review

Compare the tool's output with physiatrist screening decisions to measure accuracy before any clinical use.

Planned
PHASE 4

Length of stay and discharge prediction

Explore whether referral-time data can support earlier, more realistic discharge planning.

Exploratory

FAQs

Frequently asked questions

What is RehabScreen?

An AI-driven clinical decision-support platform for inpatient rehabilitation facilities. It reviews incoming referral documents, generates a standardized patient summary, and is exploring predictive analytics for length of stay and discharge planning.

Does the AI decide who gets admitted?

No. The tool drafts and organizes; it does not decide. A rehabilitation physician reviews the preadmission screening and makes the admission decision, as CMS requires for IRF admissions.

What does it read?

The referral packet sent from acute care: history and physical, progress notes, therapy evaluations, nursing documentation, and case management notes.

How does it relate to IRF coverage criteria?

It organizes the evidence around them: need for active and ongoing intervention from multiple therapy disciplines (one of which is PT or OT), ability to participate in an intensive rehabilitation program, a reasonable expectation of measurable functional improvement, and need for rehabilitation physician supervision and an interdisciplinary approach.

Has it been validated?

Not yet. The tool is in active development. Validation against physician review is planned, and results will be posted here when available.

How is patient privacy handled?

RehabScreen is being built to meet HIPAA requirements for protected health information, and patient data is encrypted.

Who is building it?

Roberto Andrade, MD, a resident physician in Physical Medicine and Rehabilitation.

Rehab starts at referral. Build it with us.

Physiatrists, admission liaisons, therapists, case managers and informaticists: if you review IRF referrals, we want your input on what the summary should contain.

Get involved