Autonomous
Software signs
- Assigns the final codes; the claim drops without coder review for charts that meet confidence and chart-type rules
Guide · Hospital coding and automation
Autonomous inpatient coding: what it covers, and where the coder still owns the chart
Short answer: on inpatient claims autonomous coding is a scope line, not a yes or no. Short, structured, single-problem stays can be finished by software at measured accuracy. Long stays, high-value MCCs, charts touched by the FY2027 grouper changes and POA ambiguity stay assisted or coder-owned, and a 300-chart blind test decides which is which.
768 MS-DRGs · Version 44MCC · CC · neitherOct 1, 2026 · FY2027 grouper300 charts · blind re-abstraction
Edition
Oct 2026
Focus
Inpatient coding automation
Audience
HIM · CDI · CFO
Codes
768 MS-DRGs · V44
Effective date
Oct 1, 2026
Length
8 min read
What "autonomous" means on the inpatient side
A professional claim files from the note. An inpatient stay has a principal diagnosis, dozens of secondaries and procedures, a POA flag on each, and a grouper that turns all of it into one of 768 MS-DRGs.
Assisted
Coder signs
- Software proposes codes, evidence and queries; a coder accepts, edits or rejects every chart
Coder-owned
Coder abstracts
- Software only runs edits and flags omissions after the fact
Takeaway
The question to ask a vendor is not whether it can code inpatient. It is which chart types it finishes alone, at what measured MS-DRG accuracy, and what happens to the rest.
Chart characteristics that predict autonomous accuracy
Short stays with one dominant problem, templated operative reports and discharge summaries, and secondary diagnoses corroborated by the problem list, assessment and orders: software does what a strong coder does on an easy chart.
| Characteristic | Favors automation | Favors a coder |
|---|---|---|
| Length of stay | Short, single problem | Long, multiple acute problems |
| Documentation | Templated, structured, corroborated by orders and results | Narrative, conflicting, late addenda |
| Severity drivers | Few, well-documented conditions | High-value MCCs that need clinical criteria |
| Grouper stability | Logic unchanged in Version 44 | Touched by an FY2027 MS-DRG change |
| Present on admission | Clear from the admission record | Ambiguous or undocumented |
| Chart type | Recommended mode | Why |
|---|---|---|
| Long stays with multiple acute problems | Coder-owned | Principal diagnosis selection depends on sequencing rules and clinical judgment about the reason for admission |
| Sepsis, respiratory failure, malnutrition, encephalopathy | Coder-owned + CDI query | High-value MCCs that auditors test against clinical criteria; documentation often needs a query before coding |
| Charts affected by FY2027 grouper changes | Assisted until validated | New MS-DRGs 210, 211, 361, 362, 403, 404, 449 and 523 to 525 change how cases group; software trained on prior years needs re-validation |
| Procedures with device and level detail (spinal fusion, revisions) | Assisted | MS-DRGs 523 to 525 depend on extent and device; operative notes vary in how they state both |
| Present-on-admission uncertainty | Coder-owned | POA indicators drive hospital-acquired condition logic and quality measures; ambiguity needs a query |
| Conflicting or late documentation | Coder-owned | Addenda, disagreement between consultants and the attending, or missing discharge summaries |
Takeaway
Any chart whose correct code depends on a question only the physician can answer is a CDI chart, not an automation chart.
FY2027 changes any inpatient coding model must be re-validated against
Inpatient codes interact through the grouper: the principal diagnosis chooses the base MS-DRG, secondaries decide MCC, CC or neither, and procedures move the case between surgical hierarchies. One wrong secondary changes the payment [1]. A vendor whose accuracy figures predate October 1, 2026 is quoting a different grouper.
| Change | Effect on coding logic |
|---|---|
| 768 MS-DRGs under Version 44 | Grouper tables, edits and any model trained on Version 43 output need the new tables |
| MS-DRGs 210 and 211 replace 258 to 262 | Two-way MCC split for pacemaker revision; CC capture no longer changes the group |
| MS-DRGs 361 and 362 replace base 264 | Same two-way logic for other circulatory O.R. procedures |
| MS-DRG 449 replaces 466 to 468 | Hip or knee prosthesis revision collapses to one group |
| MS-DRGs 403 and 404 | Periprosthetic joint infection as principal diagnosis, with antibiotic-eluting filler logic |
| MS-DRGs 523 to 525 | Extensive or complex spinal fusion except cervical, three severity levels, device-specific logic |
| Severity designation changes | Z59 housing codes revised; E40 to E43 remain MCCs; new FY2027 codes assigned individually |
The 300-chart test, step by step
Three hundred charts across service lines is enough to see patterns; thirty is not.
Sample by service line and length of stay
Step 1
Stratify so long stays and high-value MCC conditions are represented, not just the easy charts.
Blind re-abstraction
Step 2
Coders code from the record without the software's output; disagreements go to a third reviewer with the auditor as ground truth.
Group both sets under Version 44
Step 3
- MS-DRG match
- Severity level match
- Principal diagnosis match, with sequencing
- POA agreement
Price the variance
Step 4
Translate mismatches into relative weight and dollars; separate over-coding from under-coding.
Set the scope line
Step 5
Allow autonomous completion only for the chart types that passed, with a standing monthly sample behind them.
| Metric | Definition | Why it matters |
|---|---|---|
| MS-DRG match rate | Share of charts where the software's final MS-DRG equals the blind coder's | The payment-level test |
| Severity level match | Share where the MCC, CC or no CC/MCC level matches | Where most dollar variance and audit exposure sits |
| Principal diagnosis match | Share where the principal diagnosis matches, with sequencing | Drives the base group and most denials |
| Secondary diagnosis recall and precision | Conditions captured and conditions over-coded, by severity level | Over-coding is the compliance risk; under-coding is the CMI risk |
| POA accuracy | Agreement on present-on-admission indicators | Quality measures and hospital-acquired condition logic |
| Query opportunity rate | Charts where the coder would have queried and the software coded through | The CDI gap automation can hide |
| Deferral rate | Charts the software declines to finish autonomously | The real automation rate is the share it finishes correctly, not the share it attempts |
Attempted versus finished correctly, an illustration
Two systems on the same 300 charts (illustrative, not a vendor benchmark)
Publish the deferral rate with the accuracy rate. The real automation rate is the share finished correctly.
System A: claims 90% coverageattempted
90%coverage claimed, accuracy unverified
System B: finishes 40%, defers the restfinished correctly at high accuracy
40%deferral rate published
Coder-owned by designlong stays, MCC criteria, POA ambiguity
35%CDI charts, not automation charts
0%25%50%75%100%
Unverified claimVerified completionCoder-owned by scope
Takeaway
A system that finishes 40 percent of charts at high accuracy and defers the rest is more useful than one that claims 90 percent coverage you cannot verify.
Six guardrails written down before the first unreviewed chart
Write the scope policyChart types and confidence thresholds eligible for autonomous completion, approved by compliance and HIM.
Stand up the monthly blind samplePer chart type, reported alongside DNFB and denial rates.
Auto-defer FY2027-touched chartsUntil that chart type is re-validated under Version 44.
Route indicators without a diagnosis to CDINever coded through.
Keep the audit trail per accountThe evidence each code rested on, retrievable when a payer or auditor asks.
Keep a rollback planA chart type returns to assisted mode when the sample falls below threshold. Our hospital coding services run this mode model; results feed the same command center as DNFB and denials. See the ASP AI Engine edits note and the AI suite; read with The CMI on FY2027 shifts, what to outsource first and the OPPS 2027 prior authorization codes.
Frequently asked questions
Can autonomous coding handle inpatient charts?
Some of them. Short, single-problem stays with structured documentation can be finished by software at measurable accuracy. Long stays, charts hinging on high-value MCC conditions such as sepsis or malnutrition, charts affected by the FY2027 grouper changes, and charts with present-on-admission ambiguity should stay assisted or coder-owned.
How should we measure MS-DRG accuracy before trusting a vendor?
Run a blind re-abstraction on a stratified sample of about 300 recent discharges. Compare MS-DRG match, severity level match, principal diagnosis match, secondary diagnosis recall and precision, POA accuracy and the query opportunity rate, grouped under the current Version 44 tables, and price the variance in relative weights.
What changed in the FY2027 MS-DRGs that affects coding automation?
The final rule set 768 MS-DRGs under Version 44 and created new groups including 210 and 211 for pacemaker revision, 361 and 362 for other circulatory O.R. procedures, 449 for hip or knee prosthesis revision, 403 and 404 for periprosthetic joint infection and 523 to 525 for extensive spinal fusion, while revising several severity designations. Any model validated on prior-year output needs re-validation.
What is a reasonable autonomous completion rate on inpatient?
Whatever your own blind sample supports for the chart types you allow. Judge a system by the share of charts it finishes correctly and the deferral rate it reports, not by the share it attempts.
- Federal Register: FY 2027 IPPS and LTCH PPS final rule (CMS-1849-F), August 4, 2026
- CMS fact sheet: FY 2027 Hospital IPPS and LTCH PPS Final Rule (CMS-1849-F), July 31, 2026
- Federal Register: FY 2027 IPPS and LTCH PPS proposed rule (CMS-1848-P), April 14, 2026
Publish the deferral rate with the accuracy rate. The real automation rate is the share finished correctly.
Want the 300-chart test run on your discharges?
Send us a de-identified sample and we return MS-DRG, severity and principal diagnosis match rates by service line, the query opportunities, and a scope line you can hold any vendor to.