Predictive accuracy of severity scoring system: a prospective cohort study using APACHE III in a Korean intensive care unit

Aim: To evaluate the discrimination, calibration, and uniformity of fit by age group, operative status, and location before ICU admission of APACHE III in a single-center ICU population. Design and setting: Prospective data collection in a 25 bed mixed (surgical and medical) ICU of 850-bed teaching hospital in Pusan, South Korea. Subjects and methods: The worst values on APACHE III variables during 24 h following ICU admission were collected from the patientʹs charts and clinical flow sheets of 284 consecutively admitted subjects. Results: The mortality rate was 31.0%, and showed a strong positive correlation between APACHE III score (r=0.97, p<0.0001 for entire population, r=0.97, p<0.0001 for medical patients, r=0.91, p<0.0001 for surgical patients). Hospital mortality was significantly higher for medical patients than surgical patients (OR=7.23, 95% CI=3.76–13.88), and for patients located in the operating room than at ward before admitting ICU (OR=0.09, 95% CI=0.04–0.23). At the predicted risk of 0.5 (66 of APACHE III score), sensitivity was 0.72, specificity 0.91, and correct classification rate 0.85. Area under the ROC curve was 0.905 (95% CI=0.867–0.943). Correlation coefficient (r) between observed and expected mortality rate was 0.99. The value (chi-square) of Lemeshow–Hosmer (L–H) goodness-of-fit statistic was 6.54 (p=0.59). In patients stratified according to age groups, operative status, and location in the hospital before ICU admission, discrimination was generally good in all subgroups (area under the ROC curve >0.85), and the chi-squared of L–H goodness-of -fit statistic showed a good fit for all subgroup, especially for operative status. Conclusions: The predictive accuracy of the APACHE III scoring system showed better discrimination, as well as uniformity of fit. So, it was thought that could be utilized for the subject hospital.