Pediatric Sepsis Score Stratification Faces Timing Ambiguity in Hydrocortisone Treatment Study

In a scientific correspondence published in Critical Care, researchers have raised critical questions regarding the temporal alignment between risk stratification and treatment in pediatric sepsis research. While an initial investigation by Sanchez-Pinto and colleagues identified that associations between early hydrocortisone administration and patient outcomes varied across Phoenix Sepsis Score strata in children with community-acquired sepsis, interpreting these findings as evidence for predictive enrichment requires demonstrating that patient risk classification strictly preceded treatment initiation.

The central methodological concern identified by Wenzhen Zhuo and co-authors centers on the timeframes used to assign patients to risk categories. In the original development cohort, a high-risk phenotype was established based on the worst organ dysfunction values documented during the initial 12 hours following pediatric intensive care unit admission. However, hydrocortisone therapy was initiated much earlier, at a median of 4.7 hours in high-risk patients and 4.1 hours in low-risk patients, meaning that the assessment window extended well past the start of therapy for many participants.

Because the phenotypic assessment period overlapped with active treatment, the resulting risk scores may partially reflect therapeutic response rather than true baseline illness severity. If hydrocortisone therapy mitigates organ dysfunction or reduces the need for supportive interventions, such as resolving septic shock within the first 12 hours, post-treatment physiological improvements could lead to patient reclassification. Comparisons conducted within these post-hoc categories may therefore fail to reflect the actual treatment effects in clinical subgroups defined prior to receiving the drug.

The authors also note that this temporal ambiguity extends to the study’s covariate adjustment strategy. The propensity score model incorporated both the Pediatric Risk of Mortality III score assessed at 12 hours and refractory shock defined by vasoactive therapy over that same 12-hour window. Because these variables blend pre-treatment clinical indicators with measurements taken after hydrocortisone was started, adjusting for them risks blocking part of the causal treatment effect or introducing selection bias, which conflicts with standard principles of target trial emulation.

Furthermore, while the primary study applied an explicit 24-hour landmark to avoid immortal time bias, the correspondence emphasizes that this design does not guarantee that pre-landmark measurements were obtained before therapy began. The hypothetical trial framework also introduced an eligibility mismatch by designing a protocol that excludes patients expected not to survive 24 hours at baseline, while the observational emulation excluded children who actually died within 24 hours, thereby conditioning analysis on observed survival that could itself be altered by early hydrocortisone.

To resolve these interpretive limitations, the authors outline two complementary target trial emulations. An admission-time analysis would establish eligibility, treatment strategies, and follow-up at a shared pre-treatment baseline while accommodating a defined initiation grace period. Alternatively, a 12-hour landmark analysis could evaluate children who remain eligible and untreated at that point, using pre-treatment scores to evaluate subsequent initiation. The authors conclude that while the Phoenix Sepsis Score remains valuable for prognostic enrichment, clarifying that apparent effect modification persists when classification strictly precedes drug administration is necessary before using the score to select hydrocortisone therapy.

In detailing how researchers can verify whether reclassification occurred, Zhuo and colleagues suggest comparing patient classifications derived from the full 12-hour observation window directly against classifications generated using only pretreatment measurements. Because the true extent and direction of potential classification shifts remain uncertain, reporting the precise timing of each score-determining physiological measurement relative to the initial dose of hydrocortisone is an essential step for evaluating whether clinical improvement altered risk category assignments.

The authors also caution that while reproducing findings across two separate patient cohorts demonstrates statistical reproducibility, it cannot overcome structural limitations inherent to the observational study design. If both cohorts share the identical 12-hour measurement window and early median treatment initiation, both datasets remain vulnerable to the exact same temporal confounding and post-treatment classification bias. Demonstrating that the apparent treatment effect modification persists when Phoenix Sepsis Score risk determination strictly precedes hydrocortisone administration is therefore critical before establishing predictive enrichment.

Addressing the specific operational details of target trial emulation, the correspondence stresses that merely removing early-treated children from the current analytical framework cannot substitute for a properly structured prospective design. A delayed 12-hour decision protocol answers an entirely different clinical question from an admission-time strategy, as evaluating patients who remain untreated at 12 hours cannot estimate the efficacy of therapy already administered during those initial hours. Aligning eligibility, treatment strategies, and follow-up with the precise clinical decision point ensures that observational analyses mirror genuine clinical dilemmas without confounding treatment timing.

Ultimately, the correspondence distinguishes between prognostic and predictive applications of the Phoenix Sepsis Score in pediatric intensive care. The score maintains its documented utility for prognostic enrichment by identifying children at higher risk of adverse outcomes from community-acquired sepsis. However, until prospective analyses verify that risk tiering precedes intervention, clinicians and trialists should avoid treating these observational associations as definitive proof that the score can guide personalized hydrocortisone treatment decisions.

Subject of Research: Temporal alignment and target trial emulation in pediatric sepsis risk stratification

Article Title: Temporal alignment of Phoenix Sepsis Score stratification and hydrocortisone treatment

Article References: Zhuo, W., Sun, M., Li, M., & Gong, X. (2026). Temporal alignment of Phoenix Sepsis Score stratification and hydrocortisone treatment. Critical Care, 30(1), Article 579. https://doi.org/10.1186/s13054-026-06462-3

Image Credits: AI Generated

DOI: 10.1186/s13054-026-06462-3

Keywords: Phoenix Sepsis Score, hydrocortisone, pediatric sepsis, community-acquired sepsis, predictive enrichment, prognostic enrichment, target trial emulation, organ dysfunction, immortal time bias, landmark analysis, PRISM III score, refractory shock, pediatric intensive care unit, covariate adjustment, treatment initiation, effect modification, critical care