Abstract:

Objective To construct a risk prediction model for bone metastasis in neuroblastoma（NB） through machine learning algorithms，aimed at assisting clinical diagnosis and treatment. Methods Totally，138 patients with NB who were first diagnosed in Xinhua Hospital of Shanghai Jiao Tong University School of Medicine from January 2019 to May 2024 were enrolled and randomly divided into training set（97 cases） and validation set（41 cases） in a 7∶3 ratio. The clinical data and laboratory determination results of all the NB patients at the first hospital admission were collected. LASSO regression analysis was used to screen variables. Nine machine learning algorithms were used to establish models for predicting the risk of NB bone metastasis. Receiver operating characteristic（ROC） curve was used to evaluate the diagnostic efficacy of each model. The optimal model was screened out based on the area under curve（AUC），and its efficacy was evaluated in the validation set. Results The variables screened by LASSO regression analysis included tumor bone marrow metastasis status，pretreatment risk classification of NB，Myc gene amplification status，bone marrow blast cell ratio，bone marrow megakaryocyte count，hemoglobin，lactate dehydrogenase（LDH），CD8⁺ T cell percentage and CD8⁺ T cell absolute value. The screened variables were included in 9 machine learning algorithms for training. The ExtraTrees model had optimal performance in predicting the bone metastasis risk of NB，with AUC of 1.000 and 0.927 in the training set and validation set. The ExtraTrees model achieved a 100% accuracy in prediction results in the training set，and a 90.2% accuracy in the validation set. Conclusions The construction of an ExtraTrees model based on machine learning algorithms has demonstrated significant potential for assessing the risk of bone metastasis in NB.

Key words: Neuroblastoma, Machine learning, Predictive model, Bone metastasis