Differentiating septic arthritis from non-infectious inflammatory causes of acute monoarticular arthritis in children: A machine learning approach based on routine laboratory tests.

OBJECTIVE: Acute monoarthritis in children poses a diagnostic challenge, particularly in distinguishing septic arthritis from non-infectious inflammatory causes. Delayed or incorrect diagnosis may lead to serious complications or inappropriate treatment. This study aims to develop and validate machine learning (ML) models for distinguishing septic arthritis from non-infectious inflammatory arthritis in children presenting with acute monoarthritis, using routinely available laboratory markers including body temperature, conventional inflammatory markers (CRP and ESR), and complete blood count (CBC) parameters along with derived hematologic ratios. METHODS: We retrospectively analyzed data from 129 pediatric patients, including 66 with non-septic arthritis and 63 with septic arthritis. Conventional inflammatory markers and CBC-derived ratios (e.g., neutrophil-to-lymphocyte ratio, platelet-to-lymphocyte ratio, red cell distribution width-to-platelet ratio) were collected. The dataset was split into training (80%) and test (20%) sets. Five supervised ML algorithms-Random Forest (RF), Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), AdaBoost, and Gradient Boosting-were trained using stratified tenfold cross-validation. Diagnostic performance was assessed using area under the ROC curve (AUC) and feature importance analysis. RESULTS: The Random Forest model achieved the highest diagnostic accuracy with an AUC of 0.97 (95% CI: 0.911-1.000), outperforming other models and individual inflammatory markers. Among the input features, neutrophil count, CRP, and several derived ratios were consistently ranked among the most important predictors. Multivariate analysis supported the predictive value of these hematologic indices. CONCLUSION: Machine learning models based on routine CBC parameters can aid in distinguishing septic arthritis from non-infectious causes of acute monoarthritis in children. These results support the integration of ML-based tools into emergency care workflows to facilitate earlier and more accurate clinical decision-making.