| Objective: Erythrocyte parameters were used to establish a mathematical model to screen and diagnose thalassemia trait(TT)and to distinguish α-TT,β-TT and iron deficiency anemia(IDA)in microcytic hypochromic anemia.Methods: Eight erythrocyte parameters of thalassemia screening population in Peking University Shenzhen Hospital from January 2019 to December 2020 were collected retrospectively,including RBC,HGB,HCT,MCV,MCH,MCHC,RDW-SD,RDW-CV.According to the gene results of thalassemia,the patients were divided into TT group and control group,and the differences of red blood cells between the two groups were compared.The genotypic distribution of thalassemia and the differences of red blood cells among different types were analyzed,and the efficacy of single parameters of red blood cells in the diagnosis of thalassemia was compared.A new formula was established by logistic regression and an AI model was established on the Dx AI intelligent research platform to screen thalassemia and distinguish α-TT from β-TT.The thalassemia screening population tracked in Peking University Shenzhen Hospital from January to December 2021 is an external verification set to verify the effectiveness of the new formula and AI model.The groups of α-TT,β-TT and IDA in microcytic hypochromic anemia were selected to analyze the differences of red blood cells among different groups.A new formula was established by logistic regression and anther AI model was established on the Dx AI intelligent research platform for differential diagnosis.The diagnostic efficacy was compared with 13 foreign formulas and verified externally.Results: A total of 3150 people were collected,including 1545 thalassemia as TT group and 1605 people as control group.Except for HCT,there were significant differences in eight parameters of red blood cells between the TT group and the control group.The highest diagnostic efficacy of single parameter of red blood cell was MCV(AUC,0.922).The new formula BDS=0.401*RBC+1.969*MCV-6.468*MCH-0.314*RDW-SD+0.468*RDW-CV+0.49*MCHC-144.635 was highly effective in the diagnosis of thalassemia by logistic regression.The AUC was 0.953 and Yoden’s index was 0.745.The sensitivity,specificity,positive likelihood ratio andnegative likelihood ratio were 0.895,0.850,0.852 and 0.894 respectively.Using the Dx AI intelligent research platform,four AI models are established to diagnose thalassemia with four machine learning methods.Among them,the Model-4 diagnosis based on support vector machine learning(SVM)has the best diagnostic efficacy,with an AUC of 0.970,accuracy of 0.913,sensitivity of 0.891 and specificity of 0.935.The AI model BDA was established to differentiate α-TT from β-TT.The AUC,sensitivity and accuracy of BDA in the training set were 0.936,0.939 and 0.874 respectively,and the AUC in the verification set was 0.906,sensitivity was 0.910 and accuracy was 0.843,which could well distinguishα-TT from β-TT and had higher clinical value.In the differential diagnosis of β-TT and IDA,the diagnostic efficiency of the new formula BDZ=8.059*HCT-8.744*RBC-0.185*MCV-2.279*HGB-1.501*RDW-SD+1.96*RDW-CV+0.875*MCHC-221.376 is higher than that of13 foreign formulas.AUC is 0.992,Yoden index is 0.917,sensitivity,specificity,positive likelihood ratio and negative likelihood ratio are0.958,0.959,0.957 and 0.959,respectively.The cut-off value is 0.157.The new AI model BDB can differentiate and diagnose α-TT,β-TT and IDA.The AUC of the model BDB is 0.993,the accuracy is 0.938,the sensitivity is 0.935 and the specificity is 0.969.Conclusion: Except for HCT,there are statistical differences in erythrocyte parameters between TT group and control group.When screening thalassemia with single parameter,the AUC of MCV was the highest.The new formula BDS and Modol-4 have high application value in the diagnosis of thalassemia.The AI model BDA can distinguish α-TT from β-TT.The new formula BDZ has high diagnostic value in the differential diagnosis ofβ-TT and IDA,and the model BDB is effective in the differential diagnosis of α-TT,β-TT and IDA. |
PDF Full Text Request