The research team, led by Wenxia Ma of the infectious disease department along with Chao Du of the pediatric intensive care unit and Xue Li of Shanxi Bethune Hospital, conducted a single-center retrospective observational study. They reviewed the records of children aged 0 to 16 years who were admitted with septic shock to the pediatric intensive care unit (PICU) of Shanxi Children’s Hospital between January 2020 and December 2024. From the hospital and laboratory information systems, they gathered clinical variables—including the use of mechanical ventilation and vasoactive medications—and a broad panel of laboratory parameters measured within the first 24 hours after PICU admission. The primary outcome they sought to predict was mortality within 28 days.

The dataset ultimately comprised 123 children with septic shock, and the outcome figures underscore how lethal the condition remains in this setting: 83 of the 123 patients, or 67.5 percent, died within 28 days. That high event rate, grim as it is, provided the statistical backbone for model building. It also highlights why clinicians urgently need better early warning instruments. Existing general severity frameworks, such as PRISM, PELOD, and SOFA, were designed for broader populations and can be cumbersome to compute at the bedside, and the authors note that their restricted practicality motivated the search for a septic-shock-specific alternative.

To sift through the many candidate predictors without overfitting the model to their modest sample, the researchers employed a two-stage variable selection strategy. First, univariable logistic regression screened each candidate variable for an association with 28-day mortality. Then, they applied least absolute shrinkage and selection operator (LASSO) regression, a statistical technique that penalizes model complexity and shrinks the coefficients of weak predictors toward zero, effectively eliminating variables that add little predictive information. Ten-fold cross-validation, reported in the supplementary material, guided the choice of the LASSO penalty, and the shrinking coefficient profiles show how the field of candidates was narrowed to a compact set of four.

Those four surviving predictors were each independently associated with death, and some of the effect sizes are striking. Children who required mechanical ventilation had roughly 24-fold higher odds of dying within 28 days (odds ratio 23.95, 95% confidence interval 6.97 to 100.54). Receiving two or more vasoactive agents—drugs such as vasopressors that prop up failing blood pressure—carried an odds ratio of 6.95 (95% CI 2.08 to 26.59). Elevated troponin T, a protein that leaks from injured heart muscle, above 40 ng/L conferred an odds ratio of 6.70 (95% CI 1.93 to 28.69), and a prolonged prothrombin time greater than 18 seconds, a marker of disordered blood clotting, carried an odds ratio of 5.33 (95% CI 1.82 to 17.05).

From the fitted multivariable logistic regression model, the team derived the point-based SCSS. The four dichotomized variables—mechanical ventilation, use of at least two vasoactive agents, prothrombin time above 18 seconds, and troponin T above 40 ng/L—translate into integer points, producing a total score that ranges from 0 to 5. Higher scores correspond to higher predicted risk of death. The appeal of such a tool lies in its accessibility: every input variable is either an obvious feature of the child’s care or a routine laboratory value available within the first day of PICU admission, meaning no specialized assays or complex equations are needed to compute it at the bedside.

How well did it perform? Within the study population, the model showed strong apparent discrimination, quantified by an area under the receiver operating characteristic curve (AUC) of 0.902, with a 95% confidence interval of 0.839 to 0.965. An AUC of that magnitude means that in nine out of ten random pairings of one survivor and one nonsurvivor, the score ranked the child who died as being at higher risk. Calibration—the agreement between predicted probabilities and observed death rates—was assessed with calibration plots and evaluated through internal bootstrap validation using 1,000 resamples, which the authors describe as showing favorable apparent calibration.

The score also proved capable of sorting patients into clinically meaningful risk tiers. When children were stratified into low-, intermediate-, and high-risk groups according to their SCSS values, the observed 28-day mortality rates were 23.8 percent, 90.0 percent, and 90.9 percent, respectively—a gradient the authors report as statistically significant (log-rank P < 0.001). Decision curve analysis, a technique that evaluates the net benefit of using a model to guide clinical decisions across a range of risk thresholds, indicated that acting on the SCSS would provide considerable clinical utility across a wide span of threshold probabilities. In practical terms, the analysis suggests the score could support decisions about escalation of care without imposing net harm across most plausible decision scenarios.

Despite these encouraging numbers, the authors are explicit about the limitations, and their caution is worth amplifying. This was a single-center retrospective study of 123 patients, and the internal validation—bootstrap resampling of 1,000 iterations—was performed on the final fitted model after predictor selection had already been completed. That sequencing, the team acknowledges, can yield optimistic performance estimates, because the same data used to choose the predictors also helped assess the result. Retrospective design introduces further constraints, including dependence on the completeness of documentation in the hospital and laboratory information systems and the possibility of referral patterns specific to one institution. The findings should therefore be interpreted with care rather than taken as a proven bedside standard.

What the SCSS does offer, pending confirmation, is a template for pragmatic early risk stratification in pediatric septic shock: a handful of variables captured in the first 24 hours, a five-point scale any clinician can compute, and a risk gradient sharp enough to distinguish a quarter-mortality cohort from cohorts where nine of ten children died. The authors conclude that external validation in larger, multicenter studies is required to confirm the model’s performance and to optimize clinically meaningful risk stratification before the score can be recommended for broader clinical application. Until such studies arrive, the SCSS stands as a promising candidate—one whose simplicity may prove to be its greatest strength if its accuracy survives the journey beyond a single hospital’s walls.

Subject of Research: Development of a clinical score for predicting 28-day mortality in pediatric septic shock

Article Title: Development and internal validation of a Septic Shock Clinical Score System (SCSS) for predicting 28-day mortality in pediatric patients: a single-center retrospective study

Article References: Ma, W., Du, C., & Li, X. (2026). Development and internal validation of a Septic Shock Clinical Score System (SCSS) for predicting 28-day mortality in pediatric patients: a single-center retrospective study. BMC Pediatrics. https://doi.org/10.1186/s12887-026-07639-z

Image Credits: AI Generated

DOI: 10.1186/s12887-026-07639-z

Keywords: pediatric septic shock, mortality prediction, risk stratification, prognostic scoring system, SCSS, PICU, logistic regression, LASSO regression, troponin T, prothrombin time, mechanical ventilation, vasoactive agents