Skip to content

Medical Documentation Software, Clinical Documentation Improvement, CDI Clinical Documentation Improvement, CDI Program, Insight

Quality Through Risk Models · Clinical Documentation Reference

The risk models behind your quality rankings, captured in real time.

Elixhauser is the comorbidity risk model that shapes mortality risk-adjustment and US News & World Report quality rankings. This page is our working reference on how it functions, and how HITEKS captures the qualifying diagnoses inside Epic with Queryless CDI.

Maintained by
AHRQ / HCUP
Current version
ICD-10-CM v2025.1
Comorbidity measures
38
Data source
Inpatient claims only
Schedule a Demo
Background

What the Elixhauser Index actually measures

The Elixhauser Comorbidity Index was introduced by Anne Elixhauser and colleagues in a 1998 Medical Care paper as a way to predict hospital resource use and outcomes from administrative claims data, without needing chart review or clinical exam findings.

Unlike the older Charlson Index, which collapses 17 conditions into a single weighted score, Elixhauser uses 38 distinct comorbidity categories and, in most modern implementations, does not collapse them into one universal weight. Each condition is evaluated as its own risk-adjuster, and different downstream models (mortality, readmission, cost) apply their own coefficients to the same 38 flags. AHRQ maintains the index through HCUP, publishing an annually refreshed software and ICD-10-CM mapping. Because it runs entirely on data every hospital already reports, it scales across any inpatient population, which is exactly why AHRQ Patient Safety Indicators, CMS value-based payment models, and US News & World Report mortality rankings all lean on some variant of it.

Mechanics

How comorbidities are identified

Elixhauser comorbidities are not simply any diagnosis on the chart. Four rules govern whether a condition counts.

01Secondary diagnoses only

The principal diagnosis is excluded from Elixhauser scoring (US News is a notable exception, which may also weigh the principal diagnosis). Only secondary, coexisting conditions feed the 38 measures.

02Present on admission, for 18 of them

18 of the 38 measures are POA-sensitive. The underlying code must be flagged as present at the time of the inpatient order (AHRQ's PATIO concept) to trigger the comorbidity.

03Comorbidities are not chronic conditions

A condition qualifies because it plausibly affects length of stay, nursing intensity, or mortality risk during this encounter, not because it is lifelong.

04Hierarchies apply

Several measures (marked with an asterisk in AHRQ's spec) are superseded when a more severe, related diagnosis is also present, to avoid double-counting the same clinical problem.

38
38 comorbidity measures

Drawn only from secondary diagnoses, the conditions that raise complexity, length of stay, cost, readmission and mortality.

Dx
Secondary diagnoses only

The principal diagnosis is generally excluded. Risk adjustment rises or falls on the secondary diagnoses you capture.

POA
Present on admission matters

18 measures only count when the condition is present at the time of the inpatient order.

// the data elements behind the model

Data elementPurpose
DX1-DXnArray of ICD-10-CM diagnoses (secondary only) used to assign the 38 comorbidity measures
YEARDischarge year - determines whether a given code is exempt from POA reporting
DQTRDischarge quarter (1-4) - also used for POA-exemption logic
POA1-POAnPresent-on-admission indicator array, required to assign the 18 POA-sensitive comorbidity measures
Source: AHRQ HCUP, Elixhauser Comorbidity Software Refined for ICD-10-CM, User Guide v2025.1. PATIO values: Y triggers the comorbidity; N, U, W, X all suppress it.

// from a diagnosis to a ranking

Secondary Dx Present onadmission Comorbiditycaptured Riskadjustment Qualityranking
Illustrative flow. A captured secondary diagnosis, present on admission, becomes a comorbidity that adjusts expected risk and moves the ranking.
Reference

The 38 comorbidity measures

Each measure has a fixed SAS-style variable name (prefixed CMR_) used throughout AHRQ documentation and most vendor risk tools. Eighteen require POA validation; the rest can be assigned from any secondary diagnosis regardless of timing.

// 38 measures · amber = POA-sensitive

CMR_AIDSAIDS
CMR_ALCOHOLAlcohol abuse
CMR_ANEMDFDeficiency anemias
CMR_AUTOIMMUNEAutoimmune conditions
CMR_BLDLOSSChronic blood loss anemia
CMR_CANCER_LEUKLeukemia
CMR_CANCER_LYMPHLymphoma
CMR_CANCER_METSMetastatic cancer
CMR_CANCER_NSITUSolid tumor, in situ*
CMR_CANCER_SOLIDSolid tumor, malignant*
CMR_CBVDCerebrovascular disease
CMR_COAGCoagulopathy
CMR_DEMENTIADementia
CMR_DEPRESSDepression
CMR_DIAB_CXDiabetes, complicated
CMR_DIAB_UNCXDiabetes, uncomplicated*
CMR_DRUG_ABUSEDrug abuse
CMR_HFHeart failure
CMR_HTN_CXHypertension, complicated
CMR_HTN_UNCXHypertension, uncomplicated*
CMR_LIVER_MLDLiver disease, mild*
CMR_LIVER_SEVLiver disease, mod-severe
CMR_LUNG_CHRONICChronic pulmonary disease
CMR_NEURO_MOVTNeuro. movement disorders
CMR_NEURO_OTHOther neurological disorders
CMR_NEURO_SEIZSeizures / epilepsy
CMR_OBESEObesity
CMR_PARALYSISParalysis
CMR_PERIVASCPeripheral vascular disease
CMR_PSYCHOSESPsychoses
CMR_PULMCIRCPulmonary circulation disease
CMR_RENLFL_MODRenal failure, moderate*
CMR_RENLFL_SEVRenal failure, severe
CMR_THYROID_HYPOHypothyroidism
CMR_THYROID_OTHOther thyroid disorders
CMR_ULCER_PEPTICPeptic ulcer w/ bleeding
CMR_VALVEValvular disease
CMR_WGHTLOSSWeight loss

// 38 comorbidity measures

18 present-on-admission sensitive20 standard
18 of the 38 measures only count when the condition is present on admission. Counts only, no code names.
The mortality model

From comorbidity score to death probability

Comorbidity flags alone are just a checklist. Van Walraven and colleagues' 2009 Medical Care study converted the Elixhauser measures into a weighted point system, where each comorbidity contributes a positive or negative integer to a single summary score. That score then maps, non-linearly, onto a probability of in-hospital death. The weights are not symmetric and not universal: a comorbidity's weight in the mortality model can differ from its weight in a readmission model, a PSI model, or US News's own methodology. The same 38 flags feed multiple, independently calibrated models.

// comorbidity burden vs expected mortality

Comorbidity burden (Elixhauser score) Expected mortality risk
Schematic. Curve shape approximates the published relationship for illustration; not a reproduction of AHRQ's or Van Walraven's exact coefficient table.

// coefficient to probability

lowhigh expectedprobability
Each captured comorbidity adds coefficient weight that shifts the expected mortality probability. Schematic.
Downstream impact

Where Elixhauser powers national quality measurement

Because it is inexpensive to compute from claims alone, Elixhauser-derived risk adjustment is folded into programs hospitals are held accountable to. AHRQ Patient Safety Indicators use Elixhauser comorbidities as risk-adjusters, including the CMS Failure to Rescue measure introduced as PSI-4's replacement under Hospital Value-Based Purchasing. US News & World Report hospital rankings use a variant of the framework in their mortality and survival scoring. All of them consume the same signal: whether a secondary diagnosis was coded, correctly mapped, and, for 18 of 38 measures, correctly flagged as present on admission. Errors at the point of documentation propagate directly into public-facing mortality and safety metrics.

// observed vs expected (O:E)

expected baseline ExpectedObserved O : E < 1
Schematic. Observed below a fair expected baseline signals stronger risk-adjusted performance. Not hospital-specific.

// why the expected side matters

A quality measure compares what was observed against what was expected given how sick the population was. Better comorbidity capture sharpens the expected side, and that is where mortality-based rankings are won.

Where it gets hard

Coding nuances that silently change the score

Elixhauser mapping is not a keyword match against the diagnosis title. Two codes that look almost identical can land on different sides of the comorbidity line.

// two near-identical codes, different weight

Code A Code B near-identical high weight little to none risk weightrisk weight
Generic example. No real codes or values shown.

// encephalopathy: not all variants qualify

ICD-10-CMTitleElixhauser measure
F05Delirium due to known physiological conditionNEURO_OTH*
G928Other toxic encephalopathy (MS-DRG MCC)- not mapped -
G9340Encephalopathy, unspecified (MS-DRG CC)NEURO_OTH*
G9341Metabolic encephalopathy (MS-DRG MCC)NEURO_OTH*
I674Hypertensive encephalopathyHTN_CX
Toxic encephalopathy alone (G92.8/G92.9) is not an Elixhauser comorbidity, but metabolic encephalopathy (G93.41) is. The distinction lives entirely in how the etiology is worded.

// chronic kidney disease: stage must be present on admission

CKD stage 3, 4, 5 or ESRD codes (N18.3x-N18.5) map to a renal failure comorbidity, but only when present on admission. If the qualifying stage is not documented and flagged as POA, the comorbidity does not trigger, even though chronic kidney disease appears in the code title. Combination codes (diabetic or hypertensive CKD) route to their own families instead.

yesno Coded condition Present onadmission? Counts toward comorbidity Excluded from scoring
Timing, not just the code, decides whether a condition qualifies. Mappings: AHRQ HCUP FY2025, final 03/28/2025.
Why precision matters at the point of care

Queryless CDI, inside Epic, at the point of care

Every mapping quirk above is a place where clinically accurate documentation can still fail to register in a risk model, understating expected mortality or dropping an observed-to-expected ratio out of favor. Because POA timing, exact wording, and code specificity all matter independently, this is a documentation integrity problem, not just a coding one. HITEKS' Epic-embedded CD and CDI engines are built to catch exactly this class of gap in real time, inside the provider's native Epic Note Editor, covering up to 90% of diagnoses across the risk models this page describes, including Elixhauser risk adjustment, HCCs, CC/MCC capture, and SOI/ROM. Each CDI nudge pairs a compliant, specificity-based query, never a diagnosis suggestion, with evidence hyperlinked to the source chart, so a POA-qualifying CKD stage or a correctly linked toxic-metabolic encephalopathy is captured before the chart closes, not caught retrospectively in an audit.

Embedded in Epic

Epic Toolbox certified. Runs inside the workflow, not alongside it.

Specificity-based and compliant

Every prompt clarifies specificity, never suggests a diagnosis. AHIMA-aligned.

Real time at the point of care

Captured while the note is written and the patient is top of mind.

95-99%
documentation accuracy
80%
denial prevention
Up to 90%
of diagnoses across risk models
Figures are HITEKS product benchmarks and reflect materially higher accuracy and denial prevention than legacy AI-assisted CDI tools.
Go deeper

Authoritative references

AHRQ / HCUPElixhauser Comorbidity Software Refined for ICD-10-CM

Official AHRQ tool page with the SAS/software package and full user guide (v2025.1).

AHRQAHRQ Patient Safety Indicators (PSI) Resources

Technical specs for each PSI, including how Elixhauser comorbidities enter each measure's risk-adjustment.

CMSFY2025 ICD-10-CM Official Coding Guidelines

Appendix I (p.115) for the complete Present on Admission reporting guidelines.

Medical Care, 2009Van Walraven C, et al. - Elixhauser Measures Into a Point System for Hospital Death

The weighted point-system that maps the 38 measures to a probability of in-hospital death.

Elixhauser is one of several risk models HITEKS supports. Each model weights conditions differently, so each needs its own documentation focus. HITEKS tunes capture to the model that matters for your goal, whether that is mortality, patient safety, or public rankings.

See HITEKS capture the diagnoses that move your rankings.

Schedule a Demo