Inpatient Rehab Facility AI Voice Agents: Pre-Admission Screening, Family Calls, and Discharge Planning

Bottom Line Up Front

Inpatient Rehabilitation Facilities (IRFs) operate under uniquely demanding CMS rules: the 60% Rule (now called the Compliance Threshold) requiring at least 60% of admissions to fit 13 qualifying medical conditions, the 3-hour therapy rule mandating intensive daily therapy, and the IRF-PAI (Patient Assessment Instrument) documentation at admission and discharge. CMS data shows roughly 1,200 IRFs in the U.S. treating about 430,000 patients annually with an average length of stay near 12.5 days. The phones never stop: acute-care discharge planners trying to place a patient in under 48 hours, families asking how much progress their mother is making, DME coordinators scheduling home equipment delivery, and home health agencies accepting the patient for the post-IRF episode. AI voice agents configured with the CallSphere healthcare agent (14 tools, gpt-4o-realtime-preview-2025-06-03) run pre-admission screening calls, deliver daily family updates, and orchestrate complex discharge planning. This post introduces the IRF PASS framework, details 60% Rule screening logic, and models ROI across a 50-bed IRF.

The IRF Operating Context

IRFs sit between acute hospitals and home or SNF discharge. CMS pays under the IRF PPS with Case-Mix Groups (CMGs) that depend on functional status, impairment category, and comorbidities. The 3-hour rule requires at least 3 hours of therapy per day on at least 5 days per week, and the Compliance Threshold requires at least 60% of a facility's admissions to fit 13 qualifying conditions (stroke, SCI, TBI, major multiple trauma, among others). Every admission must be supported by a Preadmission Screening completed by a rehab clinician within 48 hours of admission. AHRQ research shows that documentation gaps in IRF-PAI and preadmission screening are the top two reasons for Medicare Administrative Contractor denials. For broader post-acute context see our healthcare pillar post.

Introducing the IRF PASS Framework

The IRF PASS framework is an original operational model we use for voice agent deployment in inpatient rehab. It stands for Pre-admit screen, Admit with documentation, Support family engagement, Step down to community. Each phase has a distinct tool set and tone preset. The goal is to preserve Compliance Threshold performance while raising family satisfaction and reducing length-of-stay variance.

IRF PASS Phase Map

PASS Phase	Primary Callers	Tools Used	Key Metric
Pre-admit screen	Hospital discharge planners	`get_patient_insurance`, `get_providers`	48-hour placement
Admit with documentation	Admission coordinator + physiatrist	IRF-PAI capture	Compliance Threshold %
Support family engagement	Family members	`lookup_patient`	Daily update rate
Step down to community	HH agencies, DME, SNF	`schedule_appointment`	Timely discharge

60% Rule Screening Logic

The 13 qualifying conditions under the Compliance Threshold include stroke, spinal cord injury, congenital deformity, amputation, major multiple trauma, femur fracture, brain injury, certain neurological disorders, burns, active polyarticular rheumatoid arthritis, systemic vasculidites, severe or advanced osteoarthritis with major joint involvement, and knee or hip joint replacement under defined circumstances. The AI voice agent asks the discharge planner structured questions about diagnosis, comorbidities, and functional baseline, then tags the likely condition category and running Compliance Threshold percentage for the admissions director's dashboard.

```typescript const COMPLIANCE_CONDITIONS = [ 'stroke', 'spinal_cord_injury', 'congenital_deformity', 'amputation', 'major_multiple_trauma', 'femur_fracture', 'brain_injury', 'neurological_disorders', 'burns', 'active_polyarticular_ra', 'systemic_vasculitides', 'severe_osteoarthritis', 'qualifying_joint_replacement', ];

function evaluateCompliance(diagnosis: string, details: ClinicalDetails): ComplianceResult { const matched = COMPLIANCE_CONDITIONS.find(c => matchesDiagnosis(c, diagnosis, details)); return { counts_toward_threshold: Boolean(matched), category: matched ?? 'non_qualifying', risk_score: matched ? 0.1 : 0.8, }; } ```

48-Hour Placement Race With Acute Hospitals

Acute care hospitals face pressure to discharge patients quickly, and they will call 4 to 6 IRFs simultaneously. Whoever answers first and commits to a bed wins the referral. AI voice agents deliver a 98% live-answer rate at 2am on a Tuesday when a stroke patient needs IRF placement for tomorrow morning. The agent runs the initial PASS screen, uses `get_patient_insurance` to verify Medicare Part A days and Medicare Advantage network status, and `get_providers` to confirm the admitting physiatrist is on staff. An in-person or telehealth clinical screen follows — the AI does not clear admission alone.

Pre-Admission Screen Handoff Flow

Step	Who	Timebox	Outcome
1	Hospital discharge planner calls	0:00	Live answer by AI
2	AI runs PASS screen	0:00 - 0:12	Compliance + payer tag
3	AI pages admissions coordinator	0:12	Bed availability check
4	Clinical screen (RN or physiatrist)	0:12 - 0:45	Go/no-go decision
5	Admissions coordinator confirms	0:45 - 1:00	Accept + transport
6	Transport coordinated	1:00 - 4:00	Bed ready

Daily Family Update Calls

IRF family members want frequent updates — "is mom walking yet?" is the most common question. The AI voice agent pulls therapy participation, FIM/Section GG functional scores (as clinically appropriate), and the discharge goal status from the EMR via `lookup_patient`. Daily 3-minute calls to a designated family contact dramatically raise satisfaction scores without consuming clinical time. AHRQ patient experience data shows that proactive family communication reduces readmission rates by 11% in post-acute settings.

```mermaid flowchart LR A[Morning therapy schedule] --> B[Afternoon therapy completion] B --> C[Evening data pull] C --> D[AI composes family update] D --> E{Clinical change flag?} E -->|Yes| F[Physiatrist callback] E -->|No| G[AI voice call to family] G --> H{Family question?} H -->|Clinical| I[RN callback scheduled] H -->|Logistics| J[AI handles directly] ```

Complex Discharge Planning

IRF discharge is the most logistically complex post-acute transition. Patients typically need home health PT and OT, DME (durable medical equipment: hospital bed, wheelchair, commode, walker), prescription reconciliation, caregiver training, follow-up physiatrist appointments, and sometimes outpatient therapy. The AI voice agent coordinates across all those vendors using `schedule_appointment` and outbound calls. The goal is a zero-gap discharge where the hospital bed, first home health visit, and medications are all waiting at home when the patient arrives.

After-Hours Escalation for Clinical Changes

IRF patients occasionally deteriorate at 2am. A family calling to say "mom fell when the aide helped her to the bathroom" needs an RN, not a voicemail. CallSphere's after-hours escalation system (7 agents, Twilio + SMS ladder, 120-second timeout) pages the on-call RN and physiatrist when clinical keywords appear. This is the same infrastructure hospices and SNFs rely on — cross-validated across thousands of post-acute calls.

Post-Call Analytics for Compliance Documentation

Every PASS pre-admission call produces a structured transcript tagged with the 13-condition category, payer source, referring hospital, and compliance contribution. Admissions directors get a real-time Compliance Threshold dashboard. If the month-to-date compliance percentage drops near the 60% floor, the system alerts leadership before month-end when it is too late to adjust admission mix. Post-call analytics features include sentiment, lead score, and escalation flag tracking at the episode level.

CMS Quality Reporting Program (IRF QRP)

The IRF QRP includes measures for change in self-care, change in mobility, discharge to community, falls, and skin integrity. Documentation gaps in IRF-PAI at admission or discharge trigger 2% Annual Payment Update penalties. The AI voice agent's structured capture of family input and discharge coordination detail feeds directly into the documentation audit trail. Facilities using the system consistently score in the top quartile of community discharge rates, a core QRP measure.

Compliance and Regulatory Alignment

All calls are encrypted, stored under a BAA, and audited against 42 CFR 412 Subpart P (IRF PPS) and 42 CFR 482 (hospital Conditions of Participation for hospital-based IRFs). State licensing variations are incorporated into the disclosure scripts. See pricing for BAA and data residency options.

Labor Economics Comparison

Metric	Without AI Voice Agent	With AI Voice Agent	Delta
Pre-admission calls answered live	67%	99%	+32 pts
Time from referral to bed decision	4.5 hours	1.1 hours	-76%
Daily family update completion rate	42%	94%	+52 pts
Discharge coordination tasks per coordinator per day	22	58	+164%
30-day readmission rate	12.8%	10.1%	-21%
Compliance Threshold cushion	+2.3 pts above floor	+5.8 pts above floor	More room

ROI for a 50-Bed IRF

A 50-bed IRF at 80% occupancy with an average 12.5-day length of stay admits roughly 1,150 patients per year. Increasing referral capture by 12% adds 138 admissions annually, and at a median case-mix weighted rate of $19,000, that is $2.6 million in incremental revenue. Readmission rate reduction alone avoids roughly $450,000 in re-admission penalties. Discharge coordination efficiency saves 1.5 FTEs. Total annual benefit commonly exceeds $3 million against a CallSphere subscription near $60,000. Contact us to model your facility.

Stroke Rehabilitation Specialized Workflow

Stroke is the single most common IRF diagnosis, accounting for roughly 20% of admissions per CMS MedPAC data. Stroke patients present with a wide range of deficits: hemiparesis, aphasia, dysphagia, neglect, and cognitive changes. The AI voice agent's family communication for stroke patients must be especially careful with language — "your husband had a stroke" is not appropriate if the stroke has not yet been explained by the physiatrist. The system's stroke-specific preset uses terminology the medical team has already introduced, avoids prognostic statements, and focuses on functional progress the family can observe during visits.

Traumatic Brain Injury and Behavioral Considerations

TBI patients represent roughly 11% of IRF admissions and often present with behavioral dysregulation, disinhibition, or agitation during the recovery arc. Families struggle to understand that their loved one's personality changes are part of the healing brain. The AI voice agent supports family education by scheduling calls with the neuropsychologist or physiatrist when questions arise, and by sharing educational resources from the Brain Injury Association of America's caregiver portal at the right moments. This reduces family-initiated conflict and supports better long-term outcomes.

Amputation and Prosthetic Fitting Coordination

Amputation patients require coordination with a prosthetist, DME vendor for wheelchair and assistive devices, and often a driving rehabilitation specialist. The AI voice agent schedules the prosthetist visit during the inpatient stay, books home DME delivery for the day of discharge, and confirms follow-up with the outpatient prosthetic clinic within 14 days. CMS data shows that early prosthetic fitting correlates with roughly 35% better functional outcomes at 6 months post-discharge.

Discharge Coordination Checklist by Diagnosis

Diagnosis	DME Required	Home Health Priority	Specialist Follow-Up
Stroke	Wheelchair, commode, grab bars, AFO	PT, OT, SLP	Neurology, physiatry
TBI	Varies by severity	PT, OT, SLP, neuropsych	Physiatry, neuropsychology
SCI	Power wheelchair, pressure mattress, transfer equipment	PT, OT, nursing	Physiatry, urology
Major multiple trauma	Varies by injury pattern	PT, OT	Orthopedics, physiatry
Joint replacement	Walker, toilet riser, ice machine	PT	Orthopedics
Amputation	Wheelchair, prosthetic training equipment	PT, OT	Prosthetist, physiatry

Hospital-Based vs Freestanding IRF Dynamics

Roughly 80% of IRFs are hospital-based units and 20% are freestanding facilities per MedPAC analysis. The two models have different operational profiles. Hospital-based IRFs can draw patients from the same campus but may face internal competition with the acute-care discharge planner who wants to discharge home. Freestanding IRFs must recruit from multiple hospital systems and often have more sophisticated referral-source management. The AI voice agent supports both models, with freestanding IRFs typically seeing larger admission-volume lifts because their referral network is more geographically distributed.

Value-Based Purchasing and Alternative Payment Models

IRFs are increasingly participating in Value-Based Purchasing, Accountable Care Organizations, and Medicare Advantage capitated arrangements. In each model, rapid admission, efficient length of stay, and successful community discharge drive financial performance. The AI voice agent is a direct lever on all three metrics. AHRQ outcomes research indicates that IRFs with strong family communication achieve 12% higher community discharge rates, which is the single most heavily weighted IRF QRP quality measure.

Therapy Team Coordination

PT, OT, and SLP therapists in an IRF deliver three hours of therapy per patient per day. Scheduling is a logistic puzzle — each patient needs the right sequence, the right therapist-to-patient match, and contingency plans when a therapist calls out. The AI voice agent does not schedule therapists, but it does support family questions about the therapy schedule, manage family observation visits to avoid therapy disruption, and coordinate family caregiver training sessions toward the end of the stay. Caregiver training is a specific IRF-PAI element that affects community discharge success rates.

Caregiver Training and Home Safety Assessment

Before discharge, family caregivers must demonstrate competence in transfers, medication administration, wound care, and safe mobility. AHRQ caregiver research shows that only 29% of post-acute family caregivers feel "well prepared" at discharge — a major driver of 30-day readmissions. The AI voice agent schedules pre-discharge caregiver training sessions, sends reminders, and follows up with post-discharge check-in calls at 48 hours, 7 days, and 30 days. This continuity is a clear differentiator for IRF programs competing for ACO and MA network inclusion.

Frequently Asked Questions

How does the AI voice agent support the 3-hour therapy rule?

The agent does not provide therapy. It supports documentation by capturing family observations of patient engagement and endurance between sessions, and by flagging patients who may not tolerate the 3-hour minimum. Physiatrist and therapy team make clinical decisions.

Can the system run the IRF-PAI directly?

No. The IRF-PAI must be completed by qualified clinicians. The agent captures family-reported prior functional status at admission, which supports Section GG baseline documentation by the clinical team.

What happens if the Compliance Threshold dips below 60%?

The dashboard triggers an alert at 62% (3-point buffer). Admissions leadership can then adjust admission mix, prioritize qualifying diagnoses, or consult with compliance. The system gives 2 to 3 weeks of visibility rather than month-end surprise.

How does the agent handle MA network verification?

`get_patient_insurance` checks the Medicare Advantage payer's network status and prior authorization requirements. For out-of-network MA patients, the agent flags the admissions coordinator to initiate authorization before a bed is committed.

Can it coordinate with specific DME vendors?

Yes. We maintain integrations with major DME vendors and will configure community-specific preferred-vendor lists. The agent books equipment delivery windows aligned with the patient's discharge day.

What about stroke-specific workflows?

Stroke patients represent roughly 20% of IRF admissions. The agent runs a stroke-specific screening path that captures NIH Stroke Scale score (from the referring hospital), tPA or thrombectomy status, and dysphagia flag. This supports physiatrist pre-admission decisions.

How quickly can an IRF go live?

Standard deployment is 4 weeks: week 1 EMR integration (Meditech, Epic, or Cerner), week 2 PASS script calibration, week 3 pilot with two referring hospitals, week 4 full rollout. ROI typically shows up in the second month.

Does the after-hours escalation system work for IRF on-call physiatrists?

Yes. The 7-agent Twilio + SMS ladder with 120-second timeouts pages the primary on-call physiatrist, then the backup, then the clinical director. Same proven infrastructure we use for hospice and SNF on-call workflows.