In recent years, the stratification of adult critical illness into phenotypes with shared clinical features, and endotypes with shared biological mechanisms of disease, has gained traction (1). These developments fore-shadowed and followed an emphasis on sepsis in adults as a dysregulated host response to infection (2) in the Sepsis-3 definition (2), increased recognition of inflammation in noninfectious diseases (3), and co-opted “big” gene expression and clinical datasets (1). The COVID-19 pandemic further accelerated these efforts. With exceptions, childhood critical illness remains more amorphous: sepsis is still defined by measurement of the systemic inflammatory response syndrome (SIRS) that has little prognostic accuracy for predicting severe disease (4,5). The etiology of childhood pneumonia is stubbornly opaque to even the most thorough research (6). Trauma, malignancies, and postoperative critical illness are less numerous (7) and have until recently been considered as separate from critical illness caused by infection. In addition, less proximate causes of childhood mortality, including malnutrition and complications of prematurity have been placed within the rubric of public health. Despite these challenges, severity scoring for childhood illness has progressed with the development of age-adjusted sequential (sepsis) organ failure assessment (SOFA) scoring, pediatric logistic organ dysfunction score-2 (PELOD-2), pediatric early warning system (PEWS), and iterations of the pediatric index of mortality (currently PIM3, used to bench-mark pediatric ICU performance) among others (5). A new definition of sepsis in children is planned for early 2024 (8), and the updated pediatric Surviving Sepsis Campaign International Guidelines remain a robust source of recommendations for management (9), including specific guidance to limit fluid bolus administration in settings without critical care capacity following the fluid expansion as supportive therapy (FEAST) trial in east African children (10).

For adults with sepsis, a dysregulated host response to infection has been operationalized by a two-point increase in the full SOFA score. This is an ordinal score from 0 to 24 encompassing Pao2/Fio2 ratio, platelet count, serum bilirubin concentration, mean arterial pressure/vasoactive infusion rate, Glasgow coma score, and either serum creatinine concentration or urine output (2). Typically the area under the receiver-operating characteristic (AUROC) curve—a measure of sensitivity and specificity that varies from 0.5 (no discrimination) to 1.0 (perfect discrimination)—is used to summarize the accuracy of a linear continuous or ordinal score for a binary outcome (such as mortality) (5). In large (> 1 million cases) mainly North American adult intensive care cohorts, SOFA had an AUROC curve of 0.74 and SIRS an AUROC of 0.64 for intensive care mortality (2). In small (< 500 cases) southeast Asian adult cohorts SOFA had an AUROC of 0.68 (mixed ICU and non-ICU) (11). The reduced prognostic accuracy in the southeast Asian cohorts may be secondary to a different etiological mix of critical illness, or a marker of host susceptibility, or a marker of differing therapeutic interventions in these units.

In this issue of Pediatric Critical Care Medicine, Chandna et al (12) set out to evaluate nine existing pediatric severity scores on prospectively collected routine data from over 1,500 child admissions, 97 of whom died, to a pediatric critical care unit in semi-rural Cambodia. The unit is capable of mechanical ventilation, vasoactive therapy, peritoneal dialysis, and specialist nursing care. The majority had presumed infection, nearly two-thirds had respiratory distress, and more than one in five had reduced alertness. The most common diagnoses in the cohort were pneumonia, bronchiolitis, and dengue shock syndrome/hemorrhagic fever—reflective of the high infectious disease burden in rural Cambodia (13). Pneumonia, bronchiolitis, undifferentiated sepsis, and melioidosis (infection with the Gram-negative bacterium Burkholderia pseudomallei) were the most common causes of death. In the primary analysis, eight of the published severity scores including quick (q)SOFA, qPELOD-2, and PEWS had AUROC curve values of 0.71–0.76, as did scores derived from resource-limited settings, including the FEAST-pediatric emergency triage (FEAST-PET) score (10) and PEWS-RL (resource-limited). In contrast, SIRS performed less well with an AUROC curve of 0.59. Crucially, these scores used clinical markers of disease such as heart rate, systolic blood pressure and mental status (qSOFA), or presence of lung crepitations (included in FEAST-PET). As in many pediatric settings, accurate measurement of blood pressure in young infants was difficult (14). None of the scores included weight-for-age, despite malnutrition being both prevalent and associated with acute mortality in children in this setting (13). Nor, except for white cell count in SIRS, were laboratory markers included despite their inclusion in nonabbreviated severity scores, such as SOFA and PELOD-2.

The authors subsequently derived their own new prognostic model for in-PICU mortality in Cambodia using routine respiratory, cardiovascular, and neurologic data, weight-for-age, and travel time to hospital. As with the Livepool (L)qSOFA score (heart rate, respiratory rate, capillary refill time, and mental status), problematic measurement of blood pressure was replaced with capillary refill time (14). The relationship of each candidate predictor to the outcome was examined individually before derivation of the model using penalized (ridge) logistic regression. Ridge penalization counters over-optimism in a model by using cross-validation to estimate a penalty parameter that is then used to shrink model coefficients and therefore reduce over-fitting. This new model had an AUROC of 0.84, significantly better than other assessed models. Validation of this model based on larger internal and external cohorts is planned.

The lack of validation on a separate cohort of patients was an acknowledged limitation of this study. By extension, a further limitation is the heterogeneity of critical illness etiology across locations with emerging critical care capacity: any model of disease severity will need considerable external validation to be widely implemented. A further limitation of the work was the use of abbreviated severity scoring systems, such as qSOFA in contrast to full scoring systems, such as the six variables of a 30-point SOFA score. However, the use of simple abbreviated scores was both pragmatic (the SOFA score requires Pao2/Fio2 ratio which is often not available to critical care pediatricians, in any setting), and justified by prior data that suggest negligible difference in prognostic accuracy between qSOFA and SOFA (5,15).

So why is this article important? At an institutional level, application of the score to critical care admissions could be used to triage the approximate 13% of children with a predicted mortality of greater than or equal to 10% to a high acuity clinical area with increased resource allocation. More broadly, these high-quality data from rural southeast Asia will contribute to regionally and globally applicable markers of severity in childhood critical illness. Regional sentinel sites for the surveillance of invasive bacterial disease already exist and several of these could provide high-quality, case-level clinical data for the development of severity scores from regions with a high childhood mortality burden. The varied prevalence of specific critical illnesses by region, such as dengue shock syndrome/hemorrhagic fever in southeast Asia, or falciparum malaria in sub-Saharan Africa, may mean that different severity scores are needed by disease-prevalence region. Alternatively, abbreviated severity scores may be appropriate across settings, which would have the advantage of enabling inter-regional and global comparisons. A compromise may be to have an abbreviated core data set standardized across settings, with additional variables by disease-prevalence region. There is no doubt that further high-quality data from the majority world, across a diverse range of settings, are needed.

In contrast to other severity scores, the new model presented by Chandna and colleagues contains two under-appreciated but easily measured variables: weight-for-age z score and estimated travel time to hospital. Both reflect socioeconomic status (undernutrition is highly prevalent in rural Cambodia, and poorer families tend to live away from the hospital) (13). A longer travel time may also reflect a delay in instituting therapy for critical illness that may be associated with mortality (9). More fundamentally, these two variables account for the question, “what sort of child is this?”; or in Bayesian parlance, “what is the prior probability of mortality in this child?” The incorporation of socioeconomic variables into critical illness severity scores may be contentious, in part because it refocuses the attention of the critical care physician on the more profound, and political, speciality of public health. However, if the incorporation of these variables as prognostic markers improves model performance then there is no clear reason why they should not be used. Perhaps the final transformative effect of COVID-19 will be an appreciation that we critical care practitioners must also account for less proximate socioeconomic risk factors for mortality in our patients, and become more knowledgeable public health physicians at the same time?

1. Maslove DM, Tang B, Shankar-Hari M, et al.: Redefining critical illness. Nat Med. 2022; 28:1141–1148 2. Singer M, Deutschman CS, Seymour CW, et al.: The third international consensus definitions for sepsis and septic shock (sepsis-3). JAMA. 2016; 315:801–810 3. Xiao W, Mindrinos MN, Seok J, et al.; Inflammation and Host Response to Injury Large-Scale Collaborative Research Program: A genomic storm in critically injured humans. J Exp Med. 2011; 208:2581–2590 4. Goldstein B, Giroir B, Randolph A; International Consensus Conference on Pediatric Sepsis: International pediatric sepsis consensus conference: Definitions for sepsis and organ dysfunction in pediatrics. Pediatr Crit Care Med. 2005; 6:2–8 5. Schlapbach LJ, Straney L, Bellomo R, et al.: Prognostic accuracy of age-adapted SOFA, SIRS, PELOD-2, and qSOFA for in-hospital mortality among children with suspected infection admitted to the intensive care unit. Intensive Care Med. 2018; 44:179–188 6. O’Brien KL, Baggett HC, Brooks WA, et al.: Causes of severe pneumonia requiring hospital admission in children without HIV infection from Africa and Asia: The PERCH multi-country case-control study. Lancet. 2019; 394:757–779 7. Perin J, Mulick A, Yeung D, et al.: Global, regional, and national causes of under-5 mortality in 2000-19: An updated systematic analysis with implications for the sustainable development goals. Lancet Child Adolesc Health. 2022; 6:106–115 8. Carrol ED, Ranjit S, Menon K, et al.; Society of Critical Care Medicine’s Pediatric Sepsis Definition Taskforce: Operationalizing appropriate sepsis definitions in children worldwide: Considerations for the pediatric sepsis definition taskforce. Pediatr Crit Care Med. 2023; 24:e263–e271 9. Weiss SL, Peters MJ, Alhazzani W, et al.: Surviving sepsis campaign international guidelines for the management of septic shock and sepsis-associated organ dysfunction in children. Pediatr Crit Care Med. 2020; 21:e52–e106 10. Maitland K, Kiguli S, Opoka RO, et al.; FEAST Trial Group: Mortality after fluid bolus in African children with severe infection. N Engl J Med. 2011; 364:2483–2495 11. Lie KC, Lau C-Y, Van Vinh Chau N, et al.; for Southeast Asia Infectious Disease Clinical Research Network: Utility of SOFA score, management and outcomes of sepsis in Southeast Asia: A multinational multicenter prospective observational study. J Intensive Care. 2018; 6:9 12. Chandna A, Keang S, Vorlark M, et al.: A Prognostic Model for Critically Ill Children in Locations With Emerging Critical Care Capacity. Pediatr Crit Care Med. 2024; 25:189–200 13. Chheng K, Carter MJ, Emary K, et al.: A prospective study of the causes of febrile illness requiring hospitalization in children in Cambodia. PLoS One. 2013; 8:e60634 14. Romaine ST, Potter J, Khanijau A, et al.: Accuracy of a modified qSOFA score for predicting critical care admission in febrile children. Pediatrics. 2020; 146:e20200782 15. Chandna A, Tan R, Carter M, et al.: Predictors of disease severity in children presenting from the community with febrile illnesses: A systematic review of prognostic studies. BMJ Glob Health. 2021; 6:e003451