Analysis Of Influencing Factors And Construction Of Neural Network Prediction Model In Patients With Systemic Lupus Erythematosus Complicated With Lupus Nephritis

BackgroundLupus nephritis(LN)is one of the most common manifestations of organ damage in patients with Systemic Lupus Erythematosus(SLE),and the severity of the disease is relatively high.The epidemiological characteristics and clinical manifestations of LN do not remain stable,but are constantly changing.As the treatment of the disease has become more scientific and sophisticated,the death rate of LN has decreased relative to the past.There has been a lack of diagnostic biomarkers strong enough to help clinicians accurately diagnose LN and assess the severity of disease activity and kidney damage in patients.ObjectivesThe main purpose of this study is to understand the clinical characteristics of SLE patients after the occurrence of LN,explore the risk factors related to the occurrence of LN,and predict LN through the neural network model.In order to prevent the further development of LN in SLE patients,certain scientific basis can be provided for the diagnosis and treatment of the already formed LN.MethodsThis study was a hospital-based study,and the subjects were divided into LN group and non-LN group according to the clinical diagnosis of SLE patients.The general demographic data,clinical manifestations,disease activity,laboratory indicators and other data of the study subjects were collected and compared between groups to explore the influencing factors related to the clinical characteristics of LN in SLE patients.First,by t test andχ~~2 inspection found differences between two groups of patients was statistically significant variables;Then,binary Logistic regression model analysis was carried out to explore the influencing factors related to LN occurrence.;Finally,LN is predicted by neural network model.P<0.05,the difference was considered statistically significant.ResultsAccording to the inclusion and exclusion criteria,a total of 621 patients with SLE were included in this study,including 268 patients with LN and 353 patients without LN.Univariate analysis of general demographic data showed that the age of LN patients was lower than that of non-LN patients(39.2±14.6 vs 42.1±15.0,P<0.05),LN disease activity in patients with higher(χ~~2=87.426,P<0.05),other variables,male to female ratio,family history,education level,BMI and drug use,were not significantly different between the two groups(P>0.05).Univariate analysis of clinical manifestations and laboratory indicators showed the following:compared to the non-LN patients,variables in the LN group were higher in incidence of serositis(13.1%vs 7.9%,P<0.05),white blood cell count(Z=2.61,P<0.01),neutrophil count(Z=2.59,P=0.01),positive rate of urinary erythrocyte(50.7%vs19.0%,P<0.01),positive rate of urinary leukocyte(42.9%vs 30.6%,P<0.01),urinary duct type positive rate(11.6%vs 5.1%,P<0.01)and urinary protein content at 24h(Z=8.04,P<0.01),urea(Z=6.69,P<0.01),creatinine(Z=7.22,P<0.01),uric acid(Z=8.51,P<0.01),anti-resistive protein positive rate(34.3%vs 24.6%,P<0.05),anti-nucleosome antibody positive rate(35.8%vs 25.5%,P<0.05),anti-DSDNA antibody positive rate(43.7%vs 28.6,P<0.05),variables in the LN group were lower in red blood cell count(Z=3.90,P<0.01),hemoglobin concentration(Z=2.54,P<0.05),estimated glomerular filtration rate(Z=5.72,P<0.01)and complement C3 level(Z=4.67,P<0.01).Multivariate Logistic regression model analysis showed that higher disease activity,positive urine erythrocyte,increased 24h albuminuria,increased urea,increased uric acid,decreased e GFR and low level of complement C3 were correlated with LN occurrence in SLE patients.The prediction accuracy of multilayer perceptron neural network model for LN and non-LN in verification set was 68.2%and 89.8%,respectively.The overall prediction accuracy was 80.3%and AUC was 0.841.The RBF neural network model predicted the correct rates of LN and non-LN in the verification set were 60.0%and 87.2%,respectively.The overall prediction accuracy was 74.7%and the AUC was 0.771.ConclusionsThrough the comparative study on the occurrence of LN in SLE patients,it was found that higher disease activity,positive urine erythrocyte,24h urine protein excess,elevated blood urea and uric acid,and decreased glomerular filtration rate and complement C3 level were associated with the incidence of LN,which could be used as indicators to predict LN.Compared with radial basis neural network model,multi-layer perceptron neural network model has higher accuracy in predicting LN and has more advantages.