The Application Value Of Machine Learning In The Diagnosis Of Systemic Lupus Erythematosu

Objective:In recent years,the application value of machine learning(ML)in the recognition of systemic lupus erythematosus(SLE)has gradually attracted people’s attention,but there is still a lack of relevant support from evidence-based medicine.Therefore,we carry out this systematic evaluation and meta-analysis to explore the accuracy and application prospects of ML for SLE.Methods:Through searching PubMed,Embase,Cochrane Library,Web of Science database,and assisted by manual search,literature review and other ways,the relevant research on machine learning recognition of SLE and neuropsychiatric systemic lupus Erythematosus(NPSLE)currently carried out in clinic was collected as comprehensively as possible.After screening by means of duplication removal,reading abstract and full text,the quality assessment of the final included literature was conducted using the Quality Assessment of Diagnostic Accuracy Studies(QUADAS)-2.Extract the specific model of machine learning included in the literature,summarize the data,evaluate the accuracy of the diagnosis of SLE and NPSLE models with the bivariate mixed effect model,draw the summary receiver operator characteristic curve(SROC),calculate the area under curve(AUC),and draw the likelihood ratio forest map,pre-test and post-test probability map for analysis of publication bias and heterogeneity.Results:A total of 18 original studies were included in this metaanalysis,of which 10 were related to SLE diagnosis and 8 were related to NPSLE.In the machine learning model related to SLE,the area under the SROC curve is 0.95(95%confidence interval(CI):0.93-0.97),the sensitivity is 0.90(95%CI:0.85-0.93),the specificity is 0.89(95%CI:0.86-0.92),the positive likelihood ratio is 8.4(95%CI:6.2-11.4),the negative likelihood ratio is 0.12(95%CI:0.08-0.17),and the diagnostic odds ratio is 73(95%CI:40-134).In NPSLE-related model recognition,the area under the SROC curve is 0.89(95%CI:0.86-0.92),the sensitivity is 0.83(95%CI:0.79-0.87),the specificity is 0.83(95%CI:0.76-0.88),the positive likelihood ratio is 5.0(95%CI:3.4-7.3),the negative likelihood ratio is 0.20(95%CI:0.15-0.27),and the diagnostic odds ratio is 25(95%CI:13-47).Conclusion:As the model is summarized,it shows a relatively ideal specificity and sensitivity.These data are strong proof of the excellent performance of machine learning in identifying SLE and NPSLE.Based on the convenience of the collection of indicators in the model and the minimally invasive nature of the examination involved,machine learning is expected to be widely used in clinical practice to assist clinicians in making decisions,identify and treat SLE and NPSLE at an early stage,and help patients gain more timely and long-term benefits.