A first-level customization study of SAPS II with Norwegian Intensive Care and Pandemic Registry (NIPaR) data

Area of Science:

• Critical Care Medicine
• Health Informatics
• Biostatistics

Background:

• Mortality prediction models (MPMs) are essential for intensive care unit (ICU) benchmarking and stratification.
• These models require regular updates using local and contextual data for continued relevance.
• The Simplified Acute Physiology Score II (SAPS II) is a widely utilized MPM in European ICUs.

Purpose of the Study:

• To customize and evaluate an updated Simplified Acute Physiology Score II (SAPS II) model using recent Norwegian intensive care data.
• To compare the performance of the new customized model (Model C) against the original SAPS II (Model A) and a previously updated model (Model B).

Main Methods:

• A customized SAPS II model (Model C) was developed using data from the Norwegian Intensive Care and Pandemic Registry (NIPaR) from 2018-2020 (excluding COVID-19 patients).
• Model C's calibration, discrimination, and uniformity of fit were compared to the original SAPS II (Model A) and a 2008-2010 NIPaR-based model (Model B).
• Performance metrics included Brier score, Cox's calibration regression, and area under the receiver operating characteristic curve (AUC).

Main Results:

• Model C demonstrated superior calibration compared to Model A (Brier scores 0.132 vs. 0.143) and comparable calibration to Model B (0.133).
• Both Model C and Model B showed better uniformity of fit across various patient and admission characteristics than Model A.
• The updated models (B and C) exhibited acceptable discrimination with an AUC of 0.79.

Conclusions:

• Observed mortality and SAPS II scores have evolved, making updated MPMs superior to the original SAPS II.
• Regular customization of prediction models with local datasets is necessary to optimize their performance.
• External validation is recommended to confirm the findings of the updated SAPS II model.