Machine learning-based analysis of blood biomarker features in spastic cerebral palsy and their clinical significance.

Background: Cerebral palsy (CP) represents the most prevalent motor disability in childhood, with spastic cerebral palsy (SCP) constituting the predominant subtype. However, systematic characterization of differences in systemic inflammatory status and metabolic profiles between children with SCP and healthy peers remains limited. Here, we applied an interpretable machine-learning framework to evaluate and identify clinically informative inflammation- and metabolism-related biomarkers in children with SCP, thereby providing potential implications for disease monitoring and informing targeted intervention strategies. Methods: In this retrospective study, 330 children with spastic cerebral palsy (SCP) and 150 healthy controls were enrolled. Complete blood count and serum biochemical parameters were collected, from which 10 systemic immune-inflammation indices were derived. Feature preselection was performed using least absolute shrinkage and selection operator (LASSO) regression, followed by univariable and multivariable logistic regression to identify biomarkers independently associated with the outcome. Model interpretability was assessed using SHapley Additive exPlanations (SHAP), and feature importance was ranked according to SHAP values. Restricted cubic splines (RCS) were applied to evaluate potential nonlinear associations between key indicators and outcome risk, while receiver operating characteristic (ROC) curves were used to assess discriminative performance. Additionally, children with SCP were stratified into severe and mild subgroups according to the Gross Motor Function Classification System (GMFCS) levels, and inflammatory and biochemical differences across severity strata were analyzed. Data were split in a 7:3 ratio using outcome-stratified sampling, with the training set used for model development and the test set for independent performance validation. Results: Multivariable logistic regression identified 7 independently associated biomarkers: MPV, CHO, DBIL were protective factors, whereas PDW, BASO%, GLB, MCHC were risk factors. A nomogram constructed based on these biomarkers demonstrated favorable performance in discriminating SCP from controls; in the independent test set, the AUC was 0.972 (95% CI, 0.935-0.998). In the SCP subgroup analysis, 330 children were stratified by GMFCS into a severe group ( Conclusion: This study systematically characterized the inflammation- and metabolism-related profiles that distinguish children with spastic cerebral palsy (SCP) from healthy controls and identified biomarkers associated with disease severity. Indicators such as mean platelet volume (MPV) and platelet distribution width (PDW) may serve as potential biological correlates for monitoring disease status and evaluating intervention responses in SCP.