Case Study · Skilled Nursing Facility

How AI Chart Review Reduced SNF Denial Rates by 38% and Recovered 2+ Hours Per MDS Coordinator Daily

A multi-site skilled nursing operator deploys AI-assisted admission chart review — and transforms how MDS coordinators and DONs prepare for every new resident.

SNF Regulatory Context

Skilled nursing facilities operate under CMS Conditions of Participation and are subject to Medicare Part A coverage criteria, MDS 3.0 assessment requirements, and RUG/PDPM payment classification. Admission chart review is a high-stakes workflow: errors in clinical documentation at intake directly affect reimbursement level, authorization approvals, and survey compliance. With SNF average denial rates running 12–18% industry-wide, accurate and timely chart review is a frontline financial defense.

The Challenge

For the MDS coordinators and Directors of Nursing at this four-site SNF operator, every new admission triggered a labor-intensive documentation sprint. Before a resident could be classified under PDPM, coordinators had to manually comb through hospital discharge summaries, therapy evaluations, physician orders, medication reconciliation records, and payer authorization letters — often arriving in fragmented fax packets or across multiple EHR portals.

A single admission review took an average of 55–70 minutes when performed manually. With daily census pressure and a lean staffing model typical of skilled nursing, this bottleneck pushed PDPM classification timelines past the 5-day assessment window on nearly one in five admissions — creating both compliance risk and revenue leakage.

Medicare Advantage and managed Medicaid payers were also intensifying prior authorization scrutiny. Denials citing “insufficient documentation of skilled need” or “lack of functional status detail” were increasing quarter-over-quarter, and the appeal process was consuming 4–6 additional hours of coordinator time per case.

“By the time I finished reviewing one chart, two more had come in. We were always a step behind before the resident even walked through the door.”
— MDS Coordinator, 9 years SNF experience

The Solution

The operator’s VP of Clinical Operations piloted an AI-assisted admission chart review tool integrated with their existing EHR and fax-intake workflow. The tool was configured specifically for SNF admission requirements, generating a structured clinical intake brief for each new resident that surfaced:

1. PDPM Classification Readiness

Auto-extracted ICD-10 codes, comorbidities, and functional status indicators from hospital records, pre-mapped to PDPM clinical categories (PT, OT, SLP, nursing, NTA) to accelerate Day 1 classification.

2. Skilled Need Justification Summary

Synthesized physician orders, therapy evaluations, and nursing notes into a concise skilled need narrative — the exact language payers require to approve and sustain authorization.

3. Medicare & MA Authorization Gap Flags

Identified missing elements required by Medicare Part A and managed care payers: three-qualifying-inpatient-days verification, prior authorization status, skilled level of care criteria alignment, and expected length of stay documentation.

4. MDS 3.0 Data Pre-Population Cues

Highlighted Section GG functional scores, cognitive assessment indicators, and NTA comorbidity items present in the hospital record, giving coordinators a head start on MDS completion before the resident arrived.

5. Denial Risk Scoring

Flagged admissions with high denial probability based on payer history, diagnosis-authorization mismatch patterns, and documentation incompleteness — enabling proactive outreach to the referring hospital before gaps became denials.

MDS coordinators and DONs retained full clinical and compliance authority. The AI brief was designed as a structured intake primer — reducing chart hunt time, not replacing the coordinator’s expertise in applying clinical judgment to each resident’s unique situation.

Results at 90 Days

What Changed on the Floor

Before: On admission notification, the MDS coordinator pulled hospital discharge summaries, tracked down therapy evals, and reconciled medication lists across up to three separate fax queues and EHR views — often taking over an hour before classification work could begin.

After Go-Live: A structured AI intake brief appeared in the coordinator’s dashboard within minutes of admission notification. Coordinators spent 15–20 minutes validating, flagging any discrepancies, and moving directly into PDPM classification and MA prior auth follow-up.

MDS Impact: Pre-populated MDS 3.0 cues reduced look-up time for Section GG and NTA items. Coordinators reported completing initial assessments 30–40% faster on average, with fewer correction cycles after physician co-signature.

Denial Impact: High-risk admission flags enabled same-day outreach to referring hospitals to obtain missing documentation before submission. The volume of “incomplete documentation” denials dropped by more than a third within the first 60 days.

Key Outcomes

• Faster PDPM Classification: 5-day assessment compliance improved from 79% to 91%, protecting PDPM payment accuracy and reducing late-classification revenue adjustments.
• Fewer MA Denials: Skilled need documentation gaps caught at admission — before submission — drove a 38% reduction in initial denial rate across Medicare Advantage plans.
• Reduced Appeal Burden: Fewer initial denials meant significantly less coordinator time spent on retrospective appeals — estimated at 18–20 hours recovered per month across the four-site group.
• Coordinator Retention Signal: Post-pilot staff surveys showed meaningful improvement in job satisfaction scores, with “administrative overload” dropping as a top cited stressor.
• Survey Readiness: Consistent, structured admission documentation created a more defensible clinical record — contributing to cleaner audit trails during state survey reviews.
• Compliance Maintained: All PDPM classifications and authorization submissions remained under MDS coordinator and DON authority. AI output was advisory and audit-logged.

Lessons for SNF Operators

Three implementation factors proved decisive: first, involving MDS coordinators in configuring the PDPM category mapping and skilled-need narrative templates, since they understood payer language nuances that generic configurations missed. Second, establishing a parallel-run period of 21 days where AI briefs were compared against manual reviews before coordinators relied on them as the primary intake summary.

Third — and particularly important for multi-site SNF operators — payer-specific tuning matters significantly. Medicare Advantage plan criteria for skilled need vary from plan to plan, and documentation language that satisfies one plan’s reviewers may not satisfy another’s. The AI configuration was adjusted per major payer contract within the first 30 days, which the team credited as a key driver of the denial rate improvement.

Operators with high Medicaid census should also evaluate whether their state’s managed Medicaid PA requirements are included in the AI’s criteria library — this varied by vendor and required custom configuration in this deployment.