Long-Term Outcome Prediction After Immunosuppressive Therapy for Severe Aplastic Anemia in Childhood by Machine Learning Methods

Abstract

Background: Severe Aplastic Anemia (SAA) in children is a rare severe disease, and the prognosis after immunosuppressive therapy (IST) is heterogeneous. Few models could accurately predict the long-term outcomes of immunosuppressive therapy for children SAA patients. Previous researches mainly focused on a few pretreatment factors,because SAA patients need to test the bone marrow at 3, 6, 9, and 12 months after IST treatment, and response-based surrogates of outcome have been shown to be highly predictive in other diseases. Based on these, we collected clinical electronic medical records (EMR) from 203 children with SAA, including comprehensive clinical tests from blood routine to bone marrow examination throughout the entire process, to establish a model with more effective prognostic factors to predict the prognosis of patients. Our study will provides a novel and more effective prognosis time node for reducing the times of bone marrow puncture and to guide the next treatment. Methods: Based on the machine learning methods, we established a model with the AUC 0.962 to predict the long-term outcomes in the early stage of SAA patients with IST. Findings: By analyzing the indicators related to long-term efficacy, we found that some of the indicators such as white blood cell count, lymphocyte count and absolute reticulocyte count (ARC) which are consistent with previous studies; but the age is not a suitable predictor for children with SAA, the lymphocyte ratio of bone marrow smear is more effective than lymphocyte count in blood, the C-reactive protein, level of vitamin B12, IL-6 and IL-8 in the early stage of the disease is highly correlated with long-term efficacy (P<0.05). Three months after IST can be used as a time node to guide the next treatment. Interpretation: Herein, we established a novel model regarding the prognosis analysis of children's SAA patients. We found that Third month after IST treatment can be considered as an essential time node of long-term prognosis. In addition, we further identified several new predictors (e.g. level of vitamin B12, IL-6 and IL-8, et al), but not including the factor of age. In summary, the utilization of our prediction model and identification of the effective and suitable prognostic factors are of great significance for the prognosis of children's SAA patients and the guiding of the relevant clinical treatment. Funding Statement: This work was supported by the National Key Research and Development Program of China (2016YFC0901503), the National Natural Science Foundation of China (81500156, 81170470). Declaration of Interests: The authors declare no competing financial interests. Ethics Approval Statement: Our study has been were reviewed and approved by the Clinical Research Ethics Committee of Blood Diseases Hospital & Institute of Hematology, Chinese Academy of Medical Sciences.