20.1 Documentation Impact on Mortality Index (SOI/ROM)

20.1 Documentation Impact on Mortality Index (SOI/ROM)

Quick Answer: All Patient Refined Diagnosis Related Groups (APR-DRGs) assign severity of illness (SOI) and risk of mortality (ROM) as independent subclasses 1–4 (minor, moderate, major, extreme). Valid secondary diagnoses that meet reporting criteria raise the expected risk the model assigns. Under-documentation leaves expected risk too low, so the same observed deaths make the mortality index look worse.

Domain VIII asks a different question than Domain IV. Chapter 11 treated SOI and ROM as program metrics—averages, subclass 3–4 shares, and mix caveats. This section treats them as the risk story behind a mortality index: how the coded record tells a severity engine whether a death was statistically surprising. Medicare Severity Diagnosis Related Groups (MS-DRGs) still drive most Inpatient Prospective Payment System (IPPS) payment. APR-DRG grouping often runs in parallel for quality profiling, some Medicaid or commercial contracts, and vendor mortality dashboards. Confusing the two engines is a classic exam miss: a CC or MCC can move payment and still leave ROM too low for the death the clinicians actually managed.

Independent 1–4 subclasses, not a single “severity” digit

APR-DRG assignment (3M methodology, used widely in inpatient quality work) produces three descriptors for each stay: a base APR-DRG, an SOI subclass, and an ROM subclass.

SOI is the extent of physiologic decompensation or organ-system loss of function. ROM is the likelihood of dying. They are related clinical ideas and they are not the same number. Each subclass is 1 minor, 2 moderate, 3 major, or 4 extreme. A stay can be SOI 4 and ROM 2, or SOI 2 and ROM 4. A dashboard that reports “severity” as one digit has already hidden the mortality problem.

CDI specialists do not key “ROM 4” onto the claim or the abstract. The grouper infers subclasses from principal diagnosis, secondary diagnoses and their combinations, age, sex, and certain procedures. Unspecified, unlinked, or missing secondaries produce subclasses that understate how sick the patient was. Manufactured secondaries produce subclasses that will not survive audit or clinical validation.

MS-DRG families commonly split with major complication or comorbidity (MCC) / with complication or comorbidity (CC) / without CC or MCC. That is a two- or three-tier payment engine plus a CC exclusion list that can neutralize a CC or MCC. It is not an independent four-level SOI/ROM pair. Do not treat “we captured an MCC” as proof that expected mortality is correct.

How SOI/ROM become a mortality index

Hospitals and vendors convert risk-adjusted mortality into a ratio. The form used throughout CDI teaching is:

Mortality index ≈ observed deaths ÷ expected deaths

Some tools label that ratio the mortality index; some invert it; a few report observed-minus-expected counts. Read the local definition. The interpretation CCDS items expect for the usual observed-to-expected (O/E) form is directional: when expected deaths are understated, the ratio rises even if the death count does not change.

• Observed deaths are actual deaths in the cohort the model defines. In-hospital expiration is common for APR-DRG ROM profiling. Centers for Medicare & Medicaid Services (CMS) public mortality measures often use a 30-day window and a different claims model (Chapter 21). Those are not interchangeable numbers.
• Expected deaths are the sum of predicted death probabilities for the stays in that cohort. When APR-DRG ROM (or another coded-risk engine such as a vendor model that consumes ICD-10-CM) is the risk adjuster, documentation that becomes coded diagnoses is a major input to expected value.

The numerator is largely a fact of who died. The denominator is a function of how completely the record described how sick they were. That split is the documentation impact Domain VIII tests.

The 7.5 and 12.5 figures are teaching arithmetic only. CMS and vendors do not publish a single national “correct” expected-death table for you to memorize, and you should not invent relative weights. Memorize the direction: under-capture of valid risk inflates the index; accurate capture raises expected and can move the index toward or below 1.0 without changing who died.

Valid secondaries raise expected risk—when they are reportable

A secondary diagnosis affects APR-DRG SOI/ROM only if it is coded. It should be coded only if it meets Uniform Hospital Discharge Data Set (UHDDS) reporting logic used with the ICD-10-CM Official Guidelines for Coding and Reporting: clinical evaluation, therapeutic treatment, diagnostic procedures, extended length of stay, or increased nursing care and/or monitoring—or if the guidelines otherwise require reporting. Dumping the entire problem list into the discharge summary does not automatically raise expected risk. Vague labels (“history of cancer,” “renal insufficiency,” “altered mental status”) often fail to carry the ROM weight of metastatic solid tumor, acute kidney injury with a specified stage, or metabolic encephalopathy when those are the conditions actually evaluated and treated.

Conditions that frequently move ROM when they are clinically present and documented include shock, acute respiratory failure, acute organ dysfunction linked to the underlying disease, anoxic brain injury, metastatic malignancy, and severe protein-calorie malnutrition. Conditions that more often move SOI include extensive multi-system decompensation that consumes resources even when immediate death risk is not the dominant story. Because the subclasses are independent, documenting only the resource-intensive complications can lift SOI while ROM still understates lethality. Mortality indices that use ROM-weighted expected deaths follow death risk, not “how busy the floor was.”

Age and procedures still feed the grouper. A review that ignores a young adult with the same shock physiology, or that ignores an operating-room procedure the logic uses, will misread the score. CDI leverage remains reportable diagnoses that are true, not a target subclass.

Under-documentation makes outcomes look worse

Picture two clinically identical deaths from septic shock with acute hypoxemic respiratory failure and acute kidney injury. Hospital A’s record supports those diagnoses as reportable secondaries (or as principal diagnosis when UHDDS “after study” logic points there). Hospital B’s record says “sepsis,” “shortness of breath,” and “elevated creatinine” without linking organ dysfunction or naming acute respiratory failure. Hospital B’s expected mortality is lower. Both hospitals count one observed death. Hospital B’s ratio looks worse.

A high mortality index can mean true excess mortality, a sicker mix the model cannot see, coding error, or a mixture. CDI’s ethical job is to reveal true risk, not to manufacture it. The August 2026 ACDIS/AHIMA Guidelines for Achieving a Compliant Query Practice still require a nonleading query with sourced clinical indicators and no reimbursement or quality-outcome language. Do not write “please document acute respiratory failure to improve our mortality index.” Place the indicators (arterial blood gas, ventilator support, documented hypoxemia, vasopressors, treatment) and let the provider name the diagnosis. Provider judgment stays independent. If the chart cannot support shock, do not chase ROM 4.

Worked documentation path

An 82-year-old is admitted from home with pneumonia and dies on hospital day 4. The attending wrote “pneumonia, DNR, family meeting.” The record also contains norepinephrine for a mean arterial pressure of 55 mm Hg after fluids, invasive mechanical ventilation, lactate 6.2 mmol/L, creatinine rising from 0.9 to 2.4 mg/dL, and bilateral infiltrates on imaging. Without clarification, coding may report pneumonia plus “hypotension” or unspecified kidney disease. APR-DRG ROM may sit at 2. The death looks unexpected relative to coded risk.

If the provider documents septic shock, acute hypoxemic respiratory failure, and acute kidney injury with clinical support, ROM can rise (often toward 3 or 4 depending on the rest of the cluster—do not memorize a guaranteed subclass). Expected mortality for that stay increases. Observed death is still one. The cohort index moves because the denominator moved.

Present on admission (POA) does not rewrite APR-DRG SOI/ROM the way it rewrites some Patient Safety Indicator and hospital-acquired condition logic, but POA still belongs in the completeness conversation: a high-lethality condition that started after admission is a different quality story than the same condition present at the inpatient order. Do not collapse those stories. Do not invent an unpublished CMS cut score for “acceptable” ROM among decedents.

Traps that invert the mortality-index story

• Treating SOI and ROM as the same subclass, or as CC versus MCC.
• Believing a higher MS-DRG relative weight automatically repairs the mortality index.
• Putting mortality-index or reimbursement language in the query.
• Adding unsupported diagnoses to “help quality.”
• Assuming every problem-list secondary raises ROM.
• Confusing in-hospital APR-DRG expected deaths with CMS 30-day mortality models.
• Trying to change observed deaths by recoding discharge status from expired to something else—that is not CDI work; it is a compliance failure.