OBJECTIVE: To study customized APACHE II and SAPS II models in predicting hospital death in patients with a prolonged length of stay in the ICU. DESIGN: Prospectively collected database. SETTING: Thirteen ICUs with 5-10 beds in Finnish secondary referral hospitals. INTERVENTIONS: None. MEASUREMENTS AND RESULTS: The database was collected between 1994 and 1999 and included 23,953 ICU admissions. In order to customize the original APACHE II and SAPS II models and to validate the models, the database was randomly divided into customization data ( n=12,064) and into validation data ( n=11,889). Logistic regression analysis was used for customization. As the length of the ICU stay was prolonged, the calibration and discrimination of both customized models worsened gradually in the validation data. Patients whose ICU stay lasted 7 days or longer (1,312 patients) consumed more than one half of all ICU days and TISS-points. Among these patients, goodness-of-fit statistics was 221.5 and 306.3 ( P<0.0001 for both) and the areas under ROC curve 0.65 and 0.62 for the customized APACHE and SAPS models, respectively. The models underestimated the risk of death in the low range and overestimated it in the high range of predicted mortality. On the other hand, both models discriminated well between survivors and non-survivors if the ICU stay was 2 days or less. CONCLUSIONS: Despite customization, the predictive models may not support clinical decision-making in those patients who require a high share of resources. More relevant instruments are needed for the prediction of outcome of patient groups who consume the major part of ICU resources.
OBJECTIVE: To study customized APACHE II and SAPS II models in predicting hospital death in patients with a prolonged length of stay in the ICU. DESIGN: Prospectively collected database. SETTING: Thirteen ICUs with 5-10 beds in Finnish secondary referral hospitals. INTERVENTIONS: None. MEASUREMENTS AND RESULTS: The database was collected between 1994 and 1999 and included 23,953 ICU admissions. In order to customize the original APACHE II and SAPS II models and to validate the models, the database was randomly divided into customization data ( n=12,064) and into validation data ( n=11,889). Logistic regression analysis was used for customization. As the length of the ICU stay was prolonged, the calibration and discrimination of both customized models worsened gradually in the validation data. Patients whose ICU stay lasted 7 days or longer (1,312 patients) consumed more than one half of all ICU days and TISS-points. Among these patients, goodness-of-fit statistics was 221.5 and 306.3 ( P<0.0001 for both) and the areas under ROC curve 0.65 and 0.62 for the customized APACHE and SAPS models, respectively. The models underestimated the risk of death in the low range and overestimated it in the high range of predicted mortality. On the other hand, both models discriminated well between survivors and non-survivors if the ICU stay was 2 days or less. CONCLUSIONS: Despite customization, the predictive models may not support clinical decision-making in those patients who require a high share of resources. More relevant instruments are needed for the prediction of outcome of patient groups who consume the major part of ICU resources.
Authors: Anna-Giulia Brunello; Matthias Haenggi; Oliver Wigger; Francesca Porta; Jukka Takala; Stephan M Jakob Journal: Intensive Care Med Date: 2009-09-16 Impact factor: 17.440
Authors: Kristel J M Janssen; Yvonne Vergouwe; Cor J Kalkman; Diederick E Grobbee; Karel G M Moons Journal: Can J Anaesth Date: 2009-02-07 Impact factor: 5.063
Authors: Murray M Pollack; Richard Holubkov; Ron Reeder; J Michael Dean; Kathleen L Meert; Robert A Berg; Christopher J L Newth; John T Berger; Rick E Harrison; Joseph Carcillo; Heidi Dalton; David L Wessel; Tammara L Jenkins; Robert Tamburro Journal: Pediatr Crit Care Med Date: 2018-03 Impact factor: 3.624
Authors: Benjamin Kasenda; Willi Sauerbrei; Patrick Royston; Alain Mercat; Arthur S Slutsky; Deborah Cook; Gordon H Guyatt; Laurent Brochard; Jean-Christophe M Richard; Thomas E Stewart; Maureen Meade; Matthias Briel Journal: BMJ Open Date: 2016-09-08 Impact factor: 2.692