International comparison of critically ill patients

Abstract

Severity of illness scores are useful in quality improvement for benchmarking care in the ICU and in research for risk-adjustment. [1] However, these scores lack generalizability beyond the patients on whom the scores were trained on. An international consortium, Global Open Source Severity of Illness Score (GOSSIS) Project, has put together various databases to create a definition of critical care severity independent of geographic and cultural data. [2] In order to combine the databases efficiently, 200 variables of interest were identified and extracted from the databases, when possible. Some variables showed different distributions throughout the databases that can be attributed to differences in medical practices and selection bias of hospitals. Thus, the predictive models that were trained in one database scored poorly when tested on a different database. Thus, a combination of the extracted variables from different databases with some necessary data manipulation should create a more generalizable severity of illness score.