Severity-of-illness scoring systems are used in the intensive care unit (ICU) setting as tools to stratify risk for therapeutic trials or as predictive instruments to evaluate the performance of a specific ICU over time or in comparison to another unit. It must be noted at the outset that they have limited utility with regard to individual patient outcomes.

Intensive care unit scoring systems fall into two broad categories. Risk prediction (also called “prognostic”) models are designed to assess prognosis at the time of admission to the ICU or within the first 24 hours. Organ dysfunction scores are used to quantify the burden of organ dysfunction and can be performed repeatedly throughout the ICU stay. Several generalizations can be made when comparing and contrasting the application of prognostic and organ dysfunction scoring systems. Prognostic scoring systems are concerned with predicting mortality, whereas organ dysfunction scores describe morbidity. Prognostic scores are obtained at time of admission or within the first 24 hours of the ICU stay and use a relatively complex set of calculations. Organ dysfunction measures are usually easier to use and can be obtained repeatedly. Prognostic scores reveal no information on individual organ dysfunction, whereas organ dysfunction scores are primarily concerned with individual organ function.

Risk prediction (prognostic) models are constructed from large databases obtained from patients admitted to hospitals throughout the United States and Europe. Logistic regression methods are applied to estimate the probability of mortality based on data obtained at or near the time of ICU admission. The accuracy of the system is influenced by how the criteria were derived for a specific system and how they are applied to a given patient. Patient factors, issues with data collection, and innovations in treatment over time all have an impact. The major outcomes measured by the most widely used systems are mortality and, in some cases, length of stay. The models assume that mortality is affected by physiologic derangements that occur early in the course of illness. The Acute Physiology and Chronic Health Evaluation (APACHE) was the first risk prediction model described. APACHE I included 34 physiologic variables derived from the medical
literature and expert opinion. It suffered from complexity and difficulty in application. APACHE II was designed to be more user-friendly and included 12 physiologic variables. The score was validated in 5,815 ICU admissions at 13 different centers. Limitations include the age of the database (outcomes are based on treatment from 1979 to 1982), lack of applicability to individual patients, and selection bias. APACHE III is an updated system that addresses some of these issues but is more difficult to use and utilizes proprietary software. The Simplified Acute Physiology Score (SAPS) was introduced in 1984 and has been updated as SAPS II. SAPS II uses data from 1991 to 1992 collected in ICUs in the United States and Europe. Seventeen variables representing physiology, type of admission, and underlying disease comprise the model.