Journal Logo


A Palace With a Common Tongue or a Multivariate Tower of Babel?*

Chalfin, Donald B. MD, MS, MPH, FCCM1,2; Kramer, Andrew A. PhD, FCCM1

Author Information
doi: 10.1097/CCM.0000000000005549
  • Free
  • Editor's Choice

Predictive models of mortality have been used extensively in critical care, predominately for benchmarking and quality improvement purposes (1). These include, among others, Acute Physiology and Chronic Health Evaluation (APACHE) (2), Simplified Acute Physiology Score 3 (3), the Mortality Probability Model (4), and the Intensive Care National Audit & Research Centre (ICNARC) (5). Their popularity is evidenced by their inclusion within ICUs worldwide. Although the major scoring systems have undergone several upgrades and revisions, leading to more sophisticated models that facilitate wider use across more varied ICU settings, they still lack broad-based applicability across different countries.

In this issue of Critical Care Medicine, Raffa et al (6) present their findings from a new severity scoring system: the Global Open Source Severity of Illness Score (GOSSIS). GOSSIS is intended to create a usable model that can facilitate meaningful comparative assessments across healthcare systems and borders. Data for GOSSIS-1 (the first version of GOSSIS) currently comes from the merger of two extremely large multicenter datasets: the Australian and New Zealand Intensive Care Society database (7) (admissions in Australia and New Zealand) and the Phillips electronic ICU (eICU) database in the United States (8). GOSSIS-1 follows but does not duplicate the APACHE methodology and the findings from its resultant hospital mortality model shows very good comparative discrimination and calibration.

One could pose the question as to why would the GOSSIS model be so successful? For starters, the team that constructed GOSISS-1 has more than ample experience in creating a large open-source data repository (9) and the Philip’s eICU database (8). Second, the authors chose to select affluent countries with a common language and some important cultural similarities, albeit with major differences, between the United States and—collectively—the Australian and New Zealand healthcare systems and their organizational approaches to critical care. It is important to establish a shared foundation between somewhat similar societies before expanding to countries with different native tongues as well as underserved health systems. Third, the authors expanded upon the APACHE model, a benchmarking system that is well established and that has undergone several major upgrades and revisions since its inception, rather than attempting to create something new from a blank slate. In doing so, they tackled two of APACHE’s main problems: the selection of an individualized patient diagnosis and the inclusion of pertinent laboratory measurements. For the latter, GOSSIS-1 distinguishes itself by the addition of platelet count, international normalized ratio, and lactate, markers that could potentially help identify patients with nascent sepsis.

There is also the possibility that GOSSIS may be unsuccessful, and at the very least, may eventually require further upgrades and external validations, especially since the databases selected for GOSSIS-1 were streamlined and somewhat modified to facilitate a combined analysis. This raises the question as to whether future database additions and revisions to GOSSIS-1 will be able to yield the same quality of information and be equally successful in their discrimination and calibration. There is also the potential for bias in the dataset of U.S. ICU patients that were included in the GOSSIS-1 model. In particular, the ICUs in the Phillips eICU database, given the unique technology associated with telemedicine and tele-ICU care, may not be representative of the more general American critical care population. More broadly, one must remain cognizant of the fact that data that were used for the GOSSIS-1 model consist of populations with relatively low hospital mortality rates. Whether GOSSIS can ultimately incorporate countries that have much higher mortality and vastly different health delivery systems remains unknown. Further, one must also ask if it will even be possible for GOSSIS to penetrate an ICU benchmarking market that is already overcrowded, both scientifically and commercially.

Raffa et al (6), much to their wisdom and credit, explicitly stress that GOSSIS is an evolving analytical system, with GOSSIS-1 serving as the foundation for an eventual expansion to more countries. Given that future versions of GOSSIS will likely resolve the problems listed above, it is fair to ask how it will be beneficially used. The obvious answer is in comparing different national models for providing critical care services. An attempt to do such a study was previously carried out (10) but did not adjust for patient case-mix. Given GOSSIS’ extensive use of patient-level data, there is a good possibility that it may successfully facilitate such cross-national comparisons. Finally, another benefit of GOSSIS would be the ability to conduct observational research studies that include many countries, while trying to level the playing field by properly adjusting for patient case-mix and differing healthcare systems. The COVID-19 pandemic serves as an excellent exemplar of the type of research that would derive great benefit from a model such as GOSSIS.

Will future versions of GOSSIS veer toward a research tool of great significance or will it become a metaphorical Tower of Babel? What we have learned through myriad upgrades and additions to APACHE is that sustained quality requires a continued source of funding, a long-lasting and perhaps even passionate commitment of personnel, and an unbounding dedication to research and discovery. If GOSSIS can achieve that, then it could become a lingua franca for ICU research.


1. Kramer AA, Zimmerman JE, Knaus WA: Severity of illness and predictive models in Society of Critical Care Medicine’s first 50 years: A tale of concord and conflict. Crit Care Med. 2021; 49:728–740
2. Zimmerman JE, Kramer AA, McNair DS, et al.: Acute Physiology and Chronic Health Evaluation (APACHE) IV: Hospital mortality assessment for today’s critically ill patients. Crit Care Med. 2006; 34:1297–1310
3. Moreno RP, Metnitz PGH, Almeida E, et al.: SAPS 3–from evaluation of the patient to evaluation of the intensive care unit. Part 2: Development of a prognostic model for hospital mortality at ICU admission. Int Care Med. 2015; 31:1345–1355
4. Higgins TL, Teres D, Copes WS, et al.: Assessing contemporary intensive care unit outcome: An updated Mortality Probability Admission Model (MPM0-III). Crit Care Med. 2007; 35:827–835
5. Harrison DA, Parry GJ, Carpenter JR, et al.: A new risk prediction model for critical care: The Intensive Care National Audit & Research Centre (ICNARC) model. Crit Care Med. 2007; 35:1091–1098
6. Raffa JD, Johnson AEW, O’Brien Z, et al.: The Global Open Source Severity of Illness Score (GOSSIS). Crit Care Med. 2022; 50:1040–1050
7. Stow PJ, Hart GK, Higlett T, et al.; ANZICS Database Management Committee: Development and implementation of a high-quality clinical database: The Australian and New Zealand Intensive Care Society Adult Patient Database. J Crit Care. 2006; 21:133–141
8. Pollard TJ, Johnson AEW, Raffa JD, et al.: The eICU collaborative research database, a freely available multi-center database for critical care research. Sci Data. 2018; 5:180178
9. Johnson AE, Stone DJ, Celi LA, et al.: The MIMIC Code Repository: Enabling reproducibility in critical care research. J Am Med Inform Assoc. 2018; 25:32–39
10. Wunsch H, Angus DC, Harrison DA, et al.: Variation in critical care services across North America and Western Europe. Crit Care Med. 2008; 36:2787–2793, e1–e9

critical care; patient outcome; predictive models; scoring systems; severity of illness

Copyright © 2022 by the Society of Critical Care Medicine and Wolters Kluwer Health, Inc. All Rights Reserved.