Data Mining And Interpretation Of Laboratory Test Results In Autoantibodies

Objective: A retrospective analysis of autoantibodies(AAB) test results, to investigate the positive distribution trends and clinical significance in disease of anti-nuclear antibody(ANA), anti-ribonucleoprotein antibody(n RNP/Sm), anti-Smith antibody(Sm), anti-Sjogren’s syndrome A antibody(SSA), anti-Sjogren’s syndrome Ro-52 antibody(Ro-52), anti-Sjogren’s syndrome B antibody(SSB), anti-topoisomerase Ⅰ antibody(Scl-70), anti-PM-Scl antibody(PM-Scl), anti-cytoplasmic aminoacyl-t RNA synthetase antibody(Jo-1), anti-centromere antibody B(CENP B), anti-proliferating cell nuclear antigen antibody(PCNA), anti-double stranded DNA antibody(ds-DNA), anti-nucleosome antibody(Nuclesome), anti-histone antibody(Histone), anti-ribose body P protein antibody(Rib ? P) and the M2 anti-mitochondrial antibody(AMA-M2). To evaluate the diagnostic efficiency of these autoantibodies in autoimmune diseases(AID).Through data mining, to explore strategies to detect autoantibodies,to establish the diagnostic model of autoimmune disease, to provide reference frame for the standardized interpretation of the autoantibodies test report and the clinical diagnosis decision for improving the efficiency of diagnosis of autoimmune diseases.Method: Through the hospital information retrieval platform,we made statistics on autoantibodies test results totaling 9585 copies of our hospital since January 2013 to September 2014, 8904 copies included in the study, which diagnosed with systemic lupus erythematosus(SLE) 668, accounting for 7.5%; diagnosed other AID without SLE 1279, accounting for 14.4%; non AID 6957, accounting for 78.1%. Eliminated a total of 681 copies. Combined with the clinical information of patients to analysis the autoantibodies in the positive distribution of different gender, age group(≤20 years of age group, 21 to 49 years old, ≥50 years old) and disease group(SLE group, AID control group, non AID control group), and the constituent ratio of various autoimmune diseases(ie, the incidence rate of treatment groups). Used ROC analysis of SPSS17.0 software to calculate the area under ROC curve(AUC) of each autoantibody, evaluated the efficiency of diagnosis for autoimmune diseases. Used the hierarchical clustering method to cluster the project variables and sample cases, analyzed the degree of similarity between the autoantibodies, compared the consistency with cluster grouping packet and clinical grouping. Maked use of decision tree, Logstic regression and artificial neural network(ANN)to establish diagnostic model of SLE and AID respectively, then carried out the blind test. compared the diagnostic efficacy of three kinds of models, selected the advantage model, extended ROC data set, used the posterior probability, the rate of misdiagnosis and the rate of missed diagnosis to interpreted the inspection report of autoantibodies.Results: 1.The positive rate of ANA, n RNP / Sm, Sm, SSA, Ro-52, SSB, Scl-70, PM-Scl, Jo-1, CENP B, PCNA, ds-DNA, Nuclesome, Histone, Rib ? P and AMA-M2 in 8904 cases patients were 24.4%, 8.4%, 3.0%, 16.1%, 12.1%, 3.5%, 2.0%, 2.1%, 1.7%, 2.3%, 4.6%, 4.3%, 3.6%, 7.6%, 4.6% and 3.4% respectively. In addition to PM-Scl, PCNA,Scl-70 and Jo-1, the positive rate of other autoantibodies in women was significantly higher than that of males(P<0.001). The positive rate of 12 kinds autoantibodies(except PM-Scl, Jo-1, PCNA and AMA-M2) between the three age groups was statistically significant(P<0.001). In the SLE group, in addition to Jo-1,CENP B and Scl-70, the positive rate of autoantibodies was significantly higher than that of the two diseases control group(P <0.05), the positive rate of all autoantibodies in AID control group were significantly higher than the non-AID control group. The morbidity of autoimmune diseases and SLE in treatment groups were respectively 20.3% and 6.96%. 2.The sequence of the area under ROC curve(AUC) for Sex, age and 16 autoantibodies in differential diagnosis of SLE: ANA(0.889)> SSA(0.821)> n RNP/Sm(0.774)> Ro-52(0.735)> Histone(0.704)> Nuclesome(0.692)> Rib ? P(0.685) > ds-DNA(0.662)> sex(0.640)> Sm(0.638)> SSB(0.579)> AMA-M2(0.537)(P<0.001), the ROC curve of age is under the diagonal reference line,AUC is 0.330(P <0.001),while the Scl-70, PM-Scl, Jo-1, CENP B and PCNA had no significant difference(P> 0.05). 3.In variable clustering, ANA, SSA, Ro-52 and SSB four clustered together; n RNP / Sm, Sm, ds-DNA, Nuclesome, Histone and Rib ? P clustered together; CENP B and AMA- 2 clustered together; sex, age, Scl-70, PM-Scl, Jo-1, CENP B, PCNA and AMA-M2 each category, were relatively far away from the other autoantibodies. The correlation coefficient between the Sample clustering and Clinical grouping was r=0.603, consistency measure Kappa = 0.476,(P<0.001), a general correlation between them. 4.By blind test validated, the sensitivity of Logstic model was 81.2%, specificity 97.6%, positive predictive value 73.5%,negative predictive value 98.4%,accuracy 96.4%. All these performance indicators are slightly higher than decision tree model and ANN model.AUC of Logstic models for differential diagnosis of SLE(AID) was 0.963(0.872), significantly higher than the other two(P<0.05).So chose the Logstic model as the best diagnostic model of SLE and AID. Used the index of posterior probability, the rate of missed diagnosis and misdiagnosis rate to extend ROC data set, for the clinical diagnosis and autoantibodies reports on reference, had good guiding significance.Conclusion: The data mining technology used for the data analysis and processing of autoantibodies detection may change the limited data into efficient diagnosis and treatment information, improve the diagnosis efficiency of autoimmune diseases, provide an effective technology way for diagnosis and treatment of clinical disease, laboratory physician participation in clinical diagnosis and to carry out inspection of information consultation.