Importance: Although the Phoenix sepsis criteria were developed from international cohorts, their performance in mortality prediction and feasibility for routine electronic health record (EHR) applications have been understudied outside of pediatric intensive care units and high-income settings outside of the US. Objective: To validate the Phoenix criteria's mortality prediction performance against other severity scores, and to evaluate the feasibility of using routine EHR data for identifying pediatric sepsis across the full spectrum of hospital care. Design, Setting, and Participants: This cohort study analyzed pediatric hospitalizations between April 1, 2009, and March 31, 2024, at 36 publicly funded hospitals in Hong Kong, China. The cohort included children 28 days to younger than 18 years of age with presumed infection. Hong Kong's population-based EHR-Clinical Data Analysis and Reporting System (CDARS)-was the source of inpatient and outpatient clinical data, including demographic characteristics, medication records, and laboratory and microbiological results. Exposure: The Phoenix-8 criteria were used primarily, defining sepsis as a score of 2 or higher and uncomplicated infection as a score lower than 2. Other severity scores-Phoenix-4, pediatric Sequential Organ Failure Assessment (pSOFA), and Pediatric Logistic Organ Dysfunction version 2 (PELOD-2)-were also calculated. Main Outcomes and Measures: Performance of Phoenix-8 criteria and comparator severity scores for all-cause hospital mortality was assessed by area under the precision recall curve (AUPRC), area under the receiver operating characteristic curve (AUROC), sensitivity, and specificity. Availability of EHR data within the CDARS was determined for score calculation. Results: Among 140 633 patients with presumed infection (median IQR age, 4.0 1.0-8.0 years; 77 665 males 55.2%), 7547 (5.4%; 95% CI, 5.2%-5.5%) met the Phoenix-8 criteria for sepsis; 133 086 patients (94.6%; 95% CI, 94.5%-94.8%) were classified as having uncomplicated infection. Patients who met the Phoenix-8 criteria had higher mortality (4.5% 95% CI, 4.0%-5.0%; n = 337 vs 0.1% 95% CI, 0.1%-0.1%; n = 114; P < .001) than patients with uncomplicated infection. Phoenix-8 demonstrated an AUPRC of 0.27 (95% CI, 0.23-0.32) and an AUROC of 0.91 (95% CI, 0.89-0.93) for mortality, outperforming pSOFA (AUPRC: 0.15 95% CI, 0.12-0.19; AUROC: 0.80 95% CI, 0.78-0.83), but not Phoenix-4 (AUPRC: 0.23 95% CI, 0.19-0.27; AUROC: 0.90 95% CI, 0.88-0.91) or PELOD-2 (AUPRC: 0.29 95% CI, 0.24-0.33; AUROC: 0.89 95% CI, 0.87-0.91). Mean data availability for variables required in score calculation was 50.7% (95% CI, 50.6%-50.7%) for pSOFA, 43.3% (95% CI, 43.2%-43.4%) for Phoenix-8, 38.7% (95% CI, 38.6%-38.7%) for PELOD-2, and 28.6% (95% CI, 28.5%-28.7%) for Phoenix-4. Conclusions and Relevance: In this cohort study of hospitalized patients with presumed infection in Hong Kong, the Phoenix criteria effectively identified patients at high risk of death across the full spectrum of inpatient care and offered the most favorable balance of mortality discrimination and data availability among the severity scores evaluated. These findings support the generalizability of the Phoenix-8 criteria for pediatric sepsis identification across international settings.
He et al. (Tue,) studied this question.