Association between the albumin-to-lymphocyte ratio and short-term mortality in critically Ill pediatric patients: a retrospective cohort study and machine learning analysis.

Background: Albumin-to-lymphocyte ratio (ALR), integrating nutritional status and immune-inflammatory response, has been proposed as a potential prognostic biomarker in various clinical settings. However, evidence regarding its prognostic value in critically ill pediatric patients remains limited. This study aimed to investigate the association between early ALR and short-term all-cause mortality in pediatric intensive care unit (PICU) patients and to evaluate its predictive utility. Methods: Clinical data were extracted from a large single-center PICU database including critically ill children admitted between 2010 and 2019. Patients were stratified into tertiles according to ALR measured within the first 24 hours after PICU admission. The primary outcome was 28-day all-cause mortality. Cox proportional hazards models were applied to assess the association between ALR and mortality after adjusting for potential confounders. Restricted cubic spline analyses were conducted to explore potential non-linear relationships. Kaplan-Meier survival curves were used to compare survival across ALR groups. Subgroup analyses were performed to evaluate the robustness of the association. In addition, multiple machine learning models incorporating ALR and other clinical variables were developed to predict 28-day mortality, and model performance was assessed using the area under the receiver operating characteristic curve (AUC). Results: A total of 7,591 critically ill children were included, with an overall 28-day mortality rate of 4.53%. Across ALR tertiles, the intermediate group showed the lowest observed 28-day mortality and the highest survival probability. In the fully adjusted Cox model, both the intermediate and highest tertiles were associated with lower mortality risk than the lowest tertile, with the strongest association observed in the intermediate tertile (T2: HR 0.55, 95% CI 0.42-0.74; T3: HR 0.65, 95% CI 0.49-0.86). Restricted cubic spline analysis revealed a significant non-linear association between ALR and mortality, characterized by a rapid decline in risk at low ALR levels, a nadir in the intermediate range, and a slight increase at higher ALR levels. This pattern was consistent with the Kaplan-Meier analysis, which showed the highest survival probability in the middle ALR tertile. Subgroup analyses showed consistent associations across different clinical strata. Among the machine learning models, the XGBoost model achieved the best discriminative performance, with an AUC of 0.835. In the SHAP-based interpretation, ALR contributed to model prediction and ranked tenth among the selected features. Conclusion: As a simple, readily available, and cost-effective biomarker derived from routine laboratory tests, ALR may serve as a complementary tool for early risk stratification and prognostic assessment in the PICU setting.