Evaluation of APACHE IV Performance and Drift in Patients With Sepsis Across Diverse Ethnicities

Sunday, March 22, 2026

9:45 AM - 10:45 AM Central Time

Location: Connections Central - RST 14

HN

Hamza Nabulsi, MD

JOR

Disclosure(s): No relevant financial relationship(s) to disclose.
SH

Sicheng Hao, MS

Research Associate
Massachusetts Institute of Technology

Disclosure information not submitted.
XL

Xiaoli Lui, PhD

Research Associate
Massachusetts Institute of Technology

Disclosure information not submitted.
JR

Jesse Raffa, PhD

Research Scientist
Massachusetts Institute of Technology

Disclosure information not submitted.
MK

Mohammed Kashkool, n/a

Medical Student
Shiraz University of Medical Sciences

Disclosure information not submitted.
LN

Lama Nazer, BCPS, PharmD, FCCM

Clinical Affairs Manager/Pharmacy
King Hussein Cancer Center

Disclosure information not submitted.
JM

João Matos, PhD

Graduate Student
University of Oxford

Disclosure information not submitted.
AW

An-Kwok Wong, MD, PhD

Duke University Medical Center

Disclosure(s): Merck Sharp & Dohme Inc.: Consultant (Ongoing), Consulting fees or other remuneration (Ongoing); Synappse: Owner/Co-Owner Founder/Co-Founder (Ongoing)
SL

Somin Mindy Lee, BASc

Undergraduate Student
University of Toronto

Disclosure information not submitted.
SL

Soyoon Lee, BS Candidate

Biomedical Engineering Undergraduate Student
University of Waterloo, Vancouver, British Columbia, Canada

Disclosure information not submitted.
LC

Leo Anthony Celi, MD, MPH, MS

MD, MPH, MS
Beth Israel Deaconess Medical Center BIDMC

Disclosure information not submitted.

Introduction: The decline in model performance over time, known as model drift, is a well-recognized challenge for maintaining the reliability of predictive models. Evidence suggests that such drift may be more pronounced in underrepresented patient populations. In this study, we evaluated the performance of the APACHE IV model over time across various underrepresented ethnic groups.

Methods: This was a retrospective study on adult patients from versions 1.0 and 2.0 of the eICU Collaborative Research Database, which includes data from over 200 hospitals across the United States. We included patients admitted with sepsis and had an APACHE IV score recorded at admission, considering only their first ICU admission. Patient ethnicities were categorized as Caucasian, African American, Hispanic, or Asian. Patients with unrecorded ethnicities were excluded. Two time periods were analyzed: 2014-2015 and 2020-2021. To evaluate the predictive performance of APACHE IV for mortality, we calculated the Area Under the Receiver Operating Characteristic Curve (AUROC) curve for discrimination and the Standardized Mortality Ratio (SMR) Values for calibration. These values were determined for all subgroups across both time periods and differences were assessed.

Results: A total of 390,509 patients were included in the study, with mean age of 63.2 ±16.9 (SD) years and 54.9% were males. For the entire cohort, the AUROC was 0.805 in 2014–2015 compared to 0.759 in 2020–2021, while the SMR increased from 0.810 to 1.16 across the same periods. Model performance and drift varied by ethnicity. For Asians, the AUROC was 0.823 vs. 0.826; for Hispanics, 0.873 vs. 0.755; for Caucasians, 0.801 vs. 0.756; and for African Americans, 0.808 vs. 0.774 in 2014–2015 vs. 2020–2021, respectively. The SMR increased over time for all groups: 0.944 vs. 1.00 for Asians, 0.827 vs. 1.166 for Caucasians, 0.727 vs. 1.13 for Hispanics, and 0.699 vs. 1.149 for African Americans.

Conclusions: Although model drift was observed in the overall cohort, the degree of change varied across ethnic groups. Future studies should ensure ethnic representation and model validation should account for these differences.