Calculator Guides

APACHE II: Why a Severity Score Is Not a Mortality Percentage

Audit APACHE II component mapping, collection windows and missing observations, and distinguish a severity total from an individual hospital-mortality probability.

Quick Medical Calculator Editorial Team · Reading label · 3 min read

Retrospective study sequence, not publication date

Editorial record

This source-linked educational article has been updated through AI-assisted evidence checks and editorial review. Verify information against current guidance and local protocols before applying it to patient care.

Citation methodology

An APACHE II total describes illness severity. It does not independently supply a reliable mortality probability for one patient. The original study related a 0-71 score to hospital death, not simply death before ICU discharge, in 5,815 ICU admissions. It emphasized combining the score with the disease being treated. Knaus and colleagues, original validation.

Separate Three Things in the Record

  1. The physiological observations and their collection window.
  2. The calculated APACHE II severity total.
  3. Any separately estimated probability, including the model, diagnostic category, endpoint, and population to which it applies.

These are not interchangeable. The APACHE framework incorporates physiological abnormalities, age, and qualifying chronic health status. A handover that reports only "APACHE II 20, mortality 40%" conceals assumptions about case mix and calibration. ICNARC distinguishes the severity score from the mortality model and describes the physiological collection window as the first 24 hours in critical care. ICNARC definitions.

Worked Arithmetic Audit

Suppose a fictional teaching worksheet lists eight contributions as 1, 2, 3, 1, 3, 3, 2, 2 and calls their sum 17. The arithmetic is correct:

1 + 2 + 3 + 1 + 3 + 3 + 2 + 2 = 17

But that is only an addition check. It does not demonstrate that the physiological values were assigned the right points, that the remaining required variables were normal, or that an acute renal failure adjustment was handled correctly. An apparently plausible total can survive multiple component errors.

For example, core temperature 39.2 C, heart rate 124/min, and arterial pH 7.26 contribute 3, 2, and 2 points, respectively. Mistakenly assigning 1, 3, and 3 still gives the same subtotal of 7. Check each mapping, not only the sum. The scoring table in an NIHR trial-data analysis documents these bands. NIHR report, table 1.

Chronic-health points also require qualifying pre-existing severe disease or immunocompromise. Elective surgery alone is not worth two points; the cardiovascular criterion refers to NYHA class IV, not class III. Public trial protocol, Appendix K.

For teaching or audit, require a complete worksheet and the versioned scoring specification before accepting a reconstructed score. Missing measurements and treatment-confounded observations must be visible, not silently converted to normal values or worst possible values.

Sedation Is Not a Neurological Diagnosis

An intubated or pharmacologically paralyzed patient cannot simply be declared GCS 3 on that basis. The GCS developers recommend marking an untestable component as NT, rather than 1, and avoiding a total when a component cannot be tested. Any APACHE data-handling convention must be documented separately from the observed neurological examination. GCS developer guidance.

Check the Study Behind the Claim

A study developing HACOR to predict noninvasive ventilation failure is not interchangeable with an APACHE II validation study. The endpoint, inputs, and clinical setting differ. Duan and colleagues, HACOR study.

Use severity scores to support structured communication and appropriately designed comparisons. Do not use a generic score-to-mortality lookup to determine whether life-sustaining treatment is offered or to give a family a falsely precise individual prognosis.