ICU severity scores: SOFA, APACHE II and what they can predict
- Concept
ICU scores turn a very sick patient into a number. They are good at tracking organs and comparing groups, and poor at deciding one person's fate.
What it is
Severity scores summarise how ill a critically ill patient is, in two families. Prognostic scores such as APACHE II (Acute Physiology and Chronic Health Evaluation II, 1985) combine the worst values of 12 routine physiological measurements in the first 24 hours of ICU care with age and chronic health status; together with the admission diagnosis they estimate the hospital mortality of patients like this one. Organ dysfunction scores such as SOFA (Sequential Organ Failure Assessment, 1996) grade each of six organ systems from 0 to 4, using the worst value in each 24-hour period, for a total of 0 to 24, and are repeated to show the trajectory. SOFA underpins the Sepsis-3 definition: in a patient with infection, an acute rise of 2 or more points marks organ dysfunction. A revised SOFA-2 was published in 2025, keeping six systems with updated variables. Bedside screening tools such as qSOFA and NEWS2 are a different thing: they flag patients outside the ICU who may be deteriorating, so that a clinician looks closer. All these scores describe and predict for groups; none was built to decide the care of one patient.
Key points
- SOFA respiratory: PaO2/FiO2 ratio and whether the patient needs respiratory support.
- SOFA coagulation: platelet count. SOFA liver: serum bilirubin.
- SOFA cardiovascular: mean arterial pressure and the need for vasopressors, graded by drug and amount.
- SOFA central nervous system: Glasgow Coma Scale. SOFA kidney: creatinine or urine output.
- APACHE II variables: core temperature, mean arterial pressure, heart rate, respiratory rate, oxygenation, arterial pH, sodium, potassium, creatinine, haematocrit, white cell count and Glasgow Coma Scale.
- APACHE II also adds age points and chronic health points for prior severe organ insufficiency or immunocompromise, weighted by whether the admission is after elective surgery, after emergency surgery or non-operative.
- SOFA-2 (2025) renames the systems (brain, respiratory, cardiovascular, liver, kidney, haemostasis), counts delirium treated with drugs in the brain score, counts non-invasive support and ECMO, combines noradrenaline and adrenaline, and scores kidney replacement therapy.
How it works
- Discrimination asks whether a score ranks those who die above those who survive; it is measured by the area under the ROC curve.
- Calibration asks whether predicted deaths match observed deaths across risk groups; it drifts as care improves and populations change, so old models need local recalibration.
- The standardised mortality ratio (observed over predicted deaths) compares an ICU's results with what the model expected for how sick its patients were.
- SOFA was designed to describe organ dysfunction over time, not to predict death, although higher and rising scores go with higher mortality.
- Care given before ICU admission, such as fluids and vasopressors in the emergency department, changes the values a score sees and can distort comparisons between units.
Three kinds of score, three different questions
| Feature | SOFA | APACHE II | qSOFA and NEWS2 |
|---|---|---|---|
| Type | Organ dysfunction score | Prognostic (mortality) score | Screening, track and trigger |
| Where used | ICU, and to define sepsis | ICU, first day | Ward, emergency department, ambulance |
| Data | Six organ systems, 0 to 4 each | 12 physiological variables, age, chronic health | Bedside vital signs and level of consciousness |
| When | Each day, worst value | Once, worst values of the first 24 hours | With every set of observations |
| Question it answers | Which organs are failing, and is it getting better or worse? | How many such patients die? (benchmarking, research) | Should someone review this patient now? |
| Main pitfall | Not built to predict death; sedation and chronic disease distort it | Calibration drifts; not for one patient's decisions | qSOFA is insensitive; a trigger is not a diagnosis |
In practice
- A SOFA that rises over the first days signals that treatment is failing; a falling score supports the current plan.
- A ward patient with pneumonia triggers a NEWS2 review: the team assesses, and if organ dysfunction is found, sepsis care starts; the trigger itself is not the diagnosis.
- An APACHE II predicted mortality is a statement about a group: an ICU with fewer deaths than predicted may be performing well, but one patient's figure does not justify withholding treatment.
- Scores can misestimate mortality in populations unlike the one they were built in, such as COVID-19, transplant recipients or cardiogenic shock on mechanical support.
- In research, baseline severity scores show whether randomised groups were comparable and adjust for case mix in observational studies.
Next step
- Calculate SOFA on admission and each day from the worst values, and read the trend.
- With suspected infection, use a SOFA rise of 2 or more points to recognise sepsis and start the sepsis bundle.
- Record APACHE II from the first 24 hours for audit, benchmarking and research, not for individual triage.
- On the ward, act on NEWS2 triggers with a bedside review, and do not use qSOFA alone to rule sepsis out.
- Discuss prognosis from the whole clinical picture, the trajectory and the patient's goals, not from a single number.
Common traps
- An APACHE II predicted mortality of 80% means this patient will probably die: it is the proportion of similar patients who died, it cannot say which ones, and alone it is no basis for withdrawing care
- qSOFA is a sepsis severity score: it is a bedside prompt from three signs outside the ICU, and it misses many patients with sepsis, so it is not used alone to screen
- SOFA is calculated once on admission: it is repeated each day using the worst values, and its trend is what matters
- APACHE II can be recalculated each ICU day to follow progress: it is defined on the worst values of the first 24 hours, and following the course is the job of SOFA
- A sedated patient's low Glasgow score means brain failure: sedation distorts the Glasgow Coma Scale, and scoring rules use the assumed value before sedation
Test yourself
Think of the answer, then tap the question to see it.
A woman with pneumonia had a baseline SOFA of 0; her platelets have fallen and her creatinine has risen, and her SOFA is now 3. Does she meet the Sepsis-3 definition?
Yes: suspected infection plus an acute SOFA rise of 2 or more points marks organ dysfunction, which defines sepsis; it carries an in-hospital mortality above 10%.
Your ICU's observed mortality is lower than APACHE II predicts. What can and cannot be concluded?
The unit may do better than expected for its case mix, but the model's calibration may also have drifted or not fit this population; the ratio says nothing about any one patient.
Why should qSOFA not be used alone to screen ward patients for sepsis?
It is specific but insensitive: many patients with sepsis, or who later die, never reach a positive qSOFA; guidelines advise against it as a single screening tool compared with SIRS, NEWS or MEWS.
Which six organ systems make up SOFA, and what is measured in each?
Respiratory (PaO2/FiO2 and respiratory support), coagulation (platelets), liver (bilirubin), cardiovascular (mean arterial pressure and vasopressors), nervous system (Glasgow Coma Scale), kidney (creatinine or urine output).
What three parts make up APACHE II, and from which period are the data taken?
An acute physiology score from 12 variables, age points and chronic health points; the worst values of the first 24 hours in the ICU are used, and the admission diagnosis is added to estimate hospital mortality.
Elsewhere in DurgiPedia
- Early warning scores: NEWS2 and qSOFA
- Fever, infection and sepsis: the bedside signs
- Sepsis code: the first hour
- Early warning scores (NEWS2) and when to escalate
- Sepsis and septic shock: screening, the first-hour bundle from the nursing side and watching organ dysfunction
- ROC curve
- Standardisation of rates
Connections
In Pharmacology2
Educational content; it does not replace supervised training, your institution's protocols or the anaesthetist's or intensivist's decision. It contains no doses and no ventilator or infusion settings.
Sources: The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3) (JAMA (PubMed Central), 2016); Rationale and methodological approach underlying the development of the Sequential Organ Failure Assessment (SOFA)-2 score: a consensus statement (JAMA Network Open, 2025); Updated Sequential Organ Failure Assessment (SOFA)-2 score: major changes and emergency department perspectives (Clinical and Experimental Emergency Medicine (PubMed Central), 2025); Clinical review: scoring systems in the critically ill (Critical Care (PubMed Central), 2010); Surviving Sepsis Campaign: international guidelines for management of sepsis and septic shock 2021 (Intensive Care Medicine (PubMed Central), 2021); Predictive value of the APACHE II score in cardiogenic shock patients treated with a percutaneous left ventricular assist device (International Journal of Cardiology. Heart & Vasculature (PubMed Central), 2022).
Spotted a mistake?We are always improving for the community. Tell us what is wrong and our team will review it.Report a mistake
Representative model, for educational purposes only.
