Comparison of serum lactate and lactate-derived ratios as prognostic biomarkers in pediatric dengue shock syndrome using supervised machine learning models

TL;DR

This study compares serum lactate and lactate-derived ratios as predictors of severe outcomes in children with dengue shock syndrome, finding that the lactate-to-albumin ratio performs best.

Contribution

The study introduces and validates the lactate-to-albumin ratio (LAR) as a more effective prognostic biomarker than serum lactate in pediatric dengue shock syndrome.

Findings

LAR showed superior predictive performance (AUC: 0.82) compared to serum lactate (AUC: 0.72) and LB ratio (AUC: 0.68).

Random forest, XGBoost, and SVM models achieved the highest predictive accuracy in the study.

SHAP analysis confirmed LAR as the most influential predictor among lactate-based variables.

Abstract

Dengue shock syndrome (DSS), with critical complications encompassing mechanical ventilation (MV), dengue-associated acute liver failure (PALF), and encephalitis, is associated with high mortality in children. Although serum lactate is a recognized prognostic biomarker, it may not fully reflect the complex metabolic disturbances in DSS. Recent evidence suggests that lactate-derived indices, including lactate-to-albumin ratio (LAR) and lactate-to-bicarbonate ratio (LB), may enhance prognostic accuracy. This study aimed to evaluate and compare the predictive performance of the LAR, LB ratio, and serum lactate levels in pediatric DSS using machine learning approaches.

We conducted a secondary analysis of a retrospective cohort study involving children with DSS at a tertiary pediatric center in Vietnam (2013–2022). The primary composite endpoint included in-hospital mortality, MV, dengue-associated PALF and encephalitis. Predictors were selected based on clinical expertise, literature review, Akaike Information Criterion and Least Absolute Shrinkage and Selection Operator. Multiple supervised machine-learning algorithms – logistic regression, random forest (RF), support vector machine (SVM), k-nearest neighbor, naïve Bayes, AdaBoost, and XGBoost - were applied. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC) and feature importance was assessed using Shapley Additive Explanations (SHAP).

Of the 524 eligible patients (median age: 8.7 years), 17% met the composite endpoint. At admission, LAR demonstrated superior discriminatory ability (AUC: 0.82; 95% CI: 0.76–0.87) compared to serum lactate (AUC: 0.72; 95% CI: 0.65–0.78) and LB ratio (AUC: 0.68; 95% CI: 0.62–0.74) (all p < 0.001). The Youden-index based optimal LAR cutoff was 1.25, whereas that for the LB ratio was 0.20. The RF, XGBoost and SVM models achieved the highest performance. SHAP analysis revealed that LAR was the most influential predictor among the lactate-based variables.

LAR exceeded serum lactate and the LB ratio in predicting critical outcomes in pediatric DSS. These findings support its utility as a practical and accessible tool for early risk stratification in DSS patients. These results support the use of LAR as a practical and accessible tool for risk stratification in pediatric dengue care.

Figures