Introduction

Illness severity scores and prognostic models serve multiple purposes in critical care.1 These include describing baseline illness severity in research participants, benchmarking intensive care unit (ICU) performance and evaluating quality improvement initiatives.2

Many challenges faced by ICUs have international scope, including pandemics, natural disasters and sociodemographic change. As a result, there is increasing interest in strengthening international collaboration in the field.[3], [4], [5] These efforts would be aided by the ability to apply scoring systems internationally. For example, in collaborative international research, this would allow for harmonised description of illness severity across registry datasets and trial populations. It would also enable the synthesis of international illness severity-stratified results by systematic reviews and meta-analyses.5

Unfortunately, many widely used models vary in their applicability to the full range of geographical and socio-economic contexts.2,6 The use of many such scores entails a significant data collection burden, and their performance is particularly inconsistent in lower-income settings.7,8 There are multiple explanations for this, including limited data availability for certain variables and outcomes.7,9 Because the mechanisms underlying data missingness differ between settings, this problem cannot easily be solved by a universal imputation strategy.6

One potential solution to this problem is the use of ‘simplified’ scoring systems restricted to variables for which data are widely available. Examples include the severity assessment score (SEVERITAS), Rwanda Mortality Probability Model (R-MPM), Tropical Intensive Care (TropICS) score and the Simplified Mortality Score for the ICU (SMS-ICU).8,[10], [11], [12]

Users of ‘simplified’ scores accept reduced predictive performance in exchange for improved utility across diverse settings. Consequently, when applied within individual countries, predictive performance will be inferior to that of locally calibrated models, e.g. the Australian and New Zealand Risk of Death (ANZROD) model.13 Therefore, they should not replace locally calibrated models for the evaluation of individual ICU performance.

SMS-ICU is calculated in the first 24 h of ICU admission using seven easily available variables. It was originally derived in clinical trial populations to predict 90-day mortality among acutely ill adults.11 The score has since been customised to predict in-hospital mortality in a large Brazilian cohort.14 Notably, it has not been validated for use in elective ICU admissions. This study aims to evaluate the suitability of this customised SMS-ICU score for use in critical care registries across multiple continents.