The Phoenix Sepsis Score (PSS) was recently proposed by an international pediatric sepsis task force as a standardized framework for identifying organ dysfunction and stratifying risk in children with suspected infection. Although the score represents an important step toward harmonizing pediatric sepsis assessment, its prognostic performance—particularly for mortality prediction—has been evaluated primarily within the healthcare systems in which it was developed. Evidence regarding its applicability and transportability across diverse clinical contexts remains limited. In this study, published in the Chinese Medical Journal on March 19, 2026, researchers assessed the performance of the PSS for predicting in-hospital mortality among pediatric intensive care unit (ICU) patients in China and explored whether a clinically grounded modification could improve risk discrimination while preserving interpretability.
Using electronic health records and registry data from five hospitals across four provinces in China, we assembled a large multicenter retrospective cohort of pediatric ICU encounters admitted between January 2012 and December 2023. After applying a standardized definition of suspected infection and prespecified exclusion criteria, 9,221 encounters were included for analysis, with an in-hospital mortality rate of 13.4%. We first evaluated the discriminative performance of the original PSS across multiple clinically relevant time windows and derived metrics, using AUROC as the primary measure and AUPRC as a complementary indicator. Across analyses, the PSS demonstrated only moderate discrimination for in-hospital mortality, with AUROC values consistently around 0.60. These findings indicate that, within this Chinese ICU population, the PSS alone may have limited ability to distinguish patients at high versus low risk of death.
Importantly, these results should not be interpreted as a refutation of the PSS as an international standard but rather as evidence that prognostic performance may vary substantially across healthcare systems. Differences in baseline patient characteristics, prevalence of chronic comorbidities, resource availability, and clinical practices—particularly in the recognition and operationalization of “suspected infection”—can meaningfully influence cohort composition and observed score performance. In this context, our findings underscore the need for caution when extrapolating the prognostic utility of the PSS to settings beyond those in which it was originally derived, especially when the score is used for mortality risk stratification rather than for its primary purpose of sepsis identification.
To address the observed limitations while remaining aligned with the conceptual framework of the PSS, we pursued a modification strategy that emphasized clinical interpretability, transparency of coefficients, and feasibility for bedside implementation. We applied extreme gradient boosting (XGBoost) to identify candidate predictors associated with in-hospital mortality and used SHapley additive exPlanations (SHAP) to rank their relative importance. Crucially, predictor selection was not driven solely by algorithmic output. Variables were incorporated into the PSS framework only if they were clinically meaningful, routinely available in practice, and interpretable to frontline clinicians. This approach yielded a modified score, PSS+, which integrates selected demographic factors, comorbidities, and vital signs with the original PSS components.
To ensure robustness and generalizability, we adopted a site-aware validation strategy. Encounters from four hospitals were used for model development and internal validation, with each site split into training and holdout subsets, while data from a fifth hospital in Fujian Province served as an independent external validation cohort. The final PSS+ was implemented as a parsimonious logistic regression model and presented as a nomogram to facilitate practical use. In both internal testing and external validation, PSS+ demonstrated substantially improved discrimination compared with the original PSS variants and pediatric Sequential Organ Failure Assessment (pSOFA), achieving AUROC values of 0.75 and 0.71, respectively. Additional analyses showed no evidence of substantial multicollinearity among predictors, supporting the stability of the simplified model.
In summary, this multicenter study provides one of the first large-scale evaluations of the PSS for mortality prediction in pediatric ICU patients in China. While the PSS—an internationally proposed standard—showed only moderate prognostic performance in this setting, a context-aware and interpretable modification substantially improved discrimination without sacrificing clinical usability. These findings highlight the importance of local validation and cautious implementation of international sepsis scoring systems and suggest that thoughtfully adapted extensions such as PSS+ may offer more reliable support for early risk stratification and decision-making in diverse healthcare environments.
***
Reference
DOI: 10.1097/CM9.0000000000003988
About the Authors
Dr. Bin Du is Vice President of Peking Union Medical College Hospital and Director of the Critical Care Medicine Center, with extensive leadership in critical care medicine and sepsis research.
Dr. Xiang Zhou is an intensivist at Peking Union Medical College Hospital, Director of the National Critical Care Quality Control Center, with expertise in clinical epidemiology, sepsis, and big data research.
Dr. Ting Shu is Director of the Department of Healthcare Informatization Research at the National Institute of Hospital Administration, National Health Commission, contributing expertise in health system management and data governance.
Funding information
This work was supported by grants from the CAMS Innovation Fund for Medical Sciences (CIFMS) (No. 2024-I2M- C&T-C-002), National Key R&D Program of China (No.2024YFF1207104), National High Level Hospital Clinical Research Funding (Nos.2022-PUMCH-B-115 & 2022-PUMCH-D-005), and Beijing Municipal Natural Science Foundation (L222019).
Chinese Medical Journal
Observational study
People
Validation and modification of the phoenix sepsis score for predicting in-hospital mortality in children with suspected infection admitted to the intensive care unit
19-Mar-2026