Related Experiment Video
Updated: May 11, 2026
Observational Study Protocol for Repeated Clinical Examination and Critical Care Ultrasonography Within the Simple Intensive Care Studies
Published on: January 16, 2019
Validation of SAPS3 admission score and its customization for use in Korean intensive care unit patients: a
So Yeon Lim1, Shin Ok Koh, Kyeongman Jeon
1Division of Pulmonary and Critical Care Medicine, Department of Medicine, Samsung Medical Center, Sungkyunkwan University School of Medicine, Seoul, Korea.
Background And Objective:
To externally validate the simplified acute physiology score 3 (SAPS3) and to customize it for use in Korean intensive care unit (ICU) patients.
Methods:
This is a prospective multicentre cohort study involving 22 ICUs from 15 centres throughout Korea. The study population comprised patients who were consecutively admitted to participating ICUs from 1 July 2010 to 31 January 2011.
Results:
A total of 4617 patients were enrolled. ICU mortality was 14.3%, and hospital mortality was 20.6%. The patients were randomly assigned into one of two cohorts: a development (n = 2309) or validation (n = 2308) cohort. In the development cohort, the general SAPS3 had good discrimination (area under the receiver operating characteristics curve = 0.829), but poor calibration (Hosmer-Lemeshow goodness-of-fit test H = 123.06, P < 0.001, C = 118.45, P < 0.001). The Australasia SAPS3 did not improve calibration (H = 73.53, P < 0.001, C = 70.52, P < 0.001). Customization was achieved by altering the logit of the original SAPS3 equation. The new equation for Korean ICU patients was validated in the validation cohort, and demonstrated both good discrimination (area under the receiver operating characteristics curve = 0.835) and good calibration (H = 4.61, P = 0.799, C = 5.67, P = 0.684).
Conclusions:
General and regional Australasia SAPS3 admission scores showed poor calibration for use in Korean ICU patients, but the prognostic power of the SAPS3 was significantly improved by customization. Prediction models should be customized before being used to predict mortality in different regions of the world.
