A Machine Learning-Based Guide for Repeated Laboratory Testing in Pediatric Emergency Departments

Background/Objectives: Laboratory tests conducted in community settings are occasionally repeated within hours of presentation to pediatric emergency departments (PEDs). Reducing unnecessary repetitions can ease child discomfort and alleviate the healthcare burden without compromising the diagnostic process or quality of care. The aim of this study was to develop a decision tree (DT) model to guide physicians in minimizing unnecessary repeat blood tests in PEDs. The minimal decision tree (MDT) algorithm was selected for its interpretability and capacity to generate optimally pruned classification trees. Methods: Children aged 3 months to 18 years with community-based complete blood count (CBC), electrolyte (ELE), and C-reactive protein (CRP) measurements obtained between 2016 and 2023 were included. Repeat tests performed in the pediatric emergency department within 12 h were evaluated by comparing paired measurements, with tests considered justified when values transitioned from normal to abnormal ranges or changed by ≥20%. Additionally, sensitivity analyses were conducted for absolute change thresholds of 10% and 30% and for repeat intervals of 6, 18, and 24 h. Results: Among 7813 children visits in this study, 6044, 1941, and 2771 underwent repeated CBC, ELE, and CRP tests, respectively. The mean ages of patients undergoing CRP, ELE, and CBC testing were 6.33 ± 5.38, 7.91 ± 5.71, and 5.08 ± 5.28 years, respectively. The majority were of middle socio-economic class, with 66.61-71.24% living in urban areas. Pain was the predominant presented complaint (83.69-85.99%), and in most cases (83.69-85.99%), the examination was conducted by a pediatrician. The DT model was developed and evaluated on training and validation cohorts, and it demonstrated high accuracy in predicting the need for repeat CBC and ELE tests but not CRP. Performance of the DT model significantly exceeded that of the logistic regression model. Conclusions: The data-driven guide derived from the DT model provides clinicians with a practical, interpretable tool to minimize unnecessary repeat laboratory testing, thereby enhancing patient care and optimizing healthcare resource utilization.