Indices de gravité et applications en réanimation

Severity scores are commonly used in ICU since more than 25 years. In order to understand the usefulness but also the limitation of these scores, ICU physicians need to know how they are developed, validated and updated. The construction of such scores requires large databases and modern statistical methods in order to identify and weigh the included variables. Validation of a score is assessed through three sets of tests: first, the discrimination representing the ability to predict death or survival. This is measured with the Area under the ROC curves. The closer to 1 the ROC area is, the better the discrimination. Second, the calibration is the assessment of the quality of the score according to different grades of severities. A score might be very accurate for very severe patients but with huge discrepancies between predicted and observed mortality for less severe patients. This parameter is analysed by the C test developed by Hosmer and Lemeshow. Third, goodness-of-fit tests the performance of a score with a different case-mix. A score might have good discrimination and calibration for certain types of patients but poor results in a different ICU with different patients. Scores might be improved through customisation techniques. The most recent scores have different components: i.e. the SAPS 3 scores identify three boxes; box1: patients characteristics before ICU admission; box2: circumstances of ICU admission; box3: degree of physiologic derangement. There is a general international consensus to consider that individual score should not be used to help in deciding whether a patient should be admitted or not in ICU. Scores are useful tools to assess ICU performance and to characterize patients included in trials.