Answers are grounded in this paper and cite their sources.
Background: Pediatric patients with hematological and oncological (HO) diseases who are admitted to the pediatric intensive care unit (PICU) are at particularly high risk of mortality.Prior studies have predominantly relied on conventional regression methods to identify prognostic factors in this population; however, machine learning (ML) approaches may offer incremental value by enhancing prognostic modeling beyond traditional techniques.Methods: This retrospective cohort study included children with HO diseases admitted to the PICU between September 2020 and August 2025.Clinical and laboratory predictors obtained within 24 hours of PICU admission were analyzed.Univariate and multivariable logistic regression analyses were conducted to identify significant predictors and to construct a multivariable regression model.In parallel, benchmark ML models were developed, and their performance was compared with that of the regression model using the area under the receiver operating characteristic curve (AUROC).Results: A total of 180 admissions involving 124 patients were included, with an overall mortality rate of 16.1%.In multivariable logistic regression analysis, lactate (odds ratio [OR], 2.058; 95% CI, 1.440-2.941),C-reactive protein (OR, 1.095; 95% CI, 1.034-1.160),and invasive mechanical ventilation (OR, 7.498; 95% CI,) were independently associated with mortality.The final regression model demonstrated an AUROC of 0.883 (95% CI, 0.787-0.936).The ML models achieved comparable AUROCs (range, 0.882-0.919)and similarly identified lactate and C-reactive protein among the most influential predictors, with procalcitonin emerging as an additional variable of high prognostic relevance.Conclusions: Both regression and ML approaches consistently identified markers of physiological compromise as being strongly associated with mortality, outweighing baseline HO characteristics.Although performance gains over logistic regression were modest, ML approaches provided complementary perspectives that may improve risk stratification in this high-risk population.
No takes yet. Share an insight, caveat, or question.
Jang et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: