Machine Learning-Based Prediction Of Treatment-Related Thyroid Adverse Events In Patients With Tumors Using PD-1/PD/L1

Objective: The aim of this study is to investigate the relationship between clinical parameters and laboratory parameters of tumor patients receiving immune checkpoint inhibitors and the occurrence of thyroid adverse events as part of endocrine-related adverse reactions.Furthermore,the study aims to establish a clinically valuable predictive method for thyroid-related adverse reactions occurring in tumor patients undergoing treatment with immune checkpoint inhibitors.Method: This study aims to collect data from patients who received immunosuppressive agents at Shengjing Hospital,Affiliated with China Medical University,between January2019 and December 2022.The collected clinical parameters include: patient age,gender,smoking status,alcohol consumption,weight,medical history,time of tumor diagnosis,tumor type,presence of distant metastasis,surgical history,prior treatments before PD-1/PD-L1 therapy,duration of treatment,type of PD-1/PD-L1 inhibitor used,and medication dosage.If patients experience immune-related adverse events,additional data will be collected,including the timing,type,duration of the adverse event,and whether medication was continued thereafter.The collected laboratory parameters include: white blood cell count,lymphocyte count,neutrophil count,monocyte count,eosinophil count,basophil count,hemoglobin level,platelet count,albumin level,prealbumin level,alanine aminotransferase level,aspartate aminotransferase level,gamma-glutamyl transferase level,creatine kinase level,creatine kinase isoenzyme level,calcium ion concentration,low-density lipoprotein level,high-density lipoprotein level,triglyceride level,total cholesterol level,creatinine level,and fasting blood glucose level.Machine learning techniques will be employed to predict whether patients will experience thyroid adverse reactions based on the clinical parameters and laboratory parameters separately.Additionally,data fusion of pre-treatment clinical parameters and laboratory parameters will be performed to predict the occurrence of thyroid adverse reactions using machine learning methods.Result: During the diagnosis and treatment of patients,the differences of laboratory parameters between patients with adverse reactions and those without adverse reactions showed various patterns over the treatment time.Only a small number of clinical and laboratory parameters were directly associated with thyroid events in tumor patients following the use of immunospot screening inhibitors.Clinical data and laboratory data were fused,and support vector machine was used to predict the clinical and laboratory information before medication.The accuracy of the final model was 0.8182.Conclusion: Specific data distribution patterns of clinical and laboratory parameters have complex relationships with thyroid events.The clinical and laboratory information of tumor patients before medication can effectively predict the occurrence of thyroid events in subsequent diagnosis and treatment.