PB1137 - Identification of autoimmune diseases based on activated partial thromboplastin time,white blood cell, mean platelet volume and Complement3 test results: A machine learning practice of screening model

Background: Blood routine and coagulation routine tests are common screening tests which can often help to indicate and identify some hidden systemic diseases. Autoimmune diseases (AIDs) are a series of chronic, inflammatory and reactive diseases caused by over activation of the immune system.At present, screening for antinuclear antibodies (ANA) tests have been used for diagnosis and evaluation. There is currently no literature reporting the screening and identification of AIDs by screening meaningful biomarkers from routine hematological testing indicators and establishing disease diagnostic models. Thus, in this study, we report our work using machine learning technology to establish an AIDs screening model based on coagulation and blood routine tests results, in order to improve the screening and recognition ability of AIDs.

Aims: To explore a screening model for AIDs based on coagulation and blood routine tests results using machine learning tools in order to achieve early identification of AIDs.

Methods: In this cross-sectional study, the AIDs screening model establishment group included 338 patients who underwent coagulation tests and LA test from 2023.01-03 in our hospital for training and internal validation.A total of 52 candidate conditions including age, gender, diagnosis, test results such as blood routine test, coagulation test, LA, etc. Taking the diagnosis of AIDs as the standard, nine Python-based machine learning tools logistic regression, XGBoost, random forest, LightGBM, AdaBoost, KNN, decision tree, GBDT and SVM were used to establish AIDs screening models,and preformance was assessed.

Results: 9 AIDs screening models involving APTT, C3, WBC, and MPV were established, among which the model from SVM showed best efficiency, with an AUC of 0.783 in the establishment group and 0.745 in the validation group.

Conclusion(s): The AIDs screening model established via the SVM machine learning tool benefit the identification of AIDs, which may be used as an indicator for AIDs to forward the time window of diagnosis.