红细胞指数鉴别轻型β地中海贫血和缺铁性贫血的作用评价

Reliability evaluation of red blood cell indices in discriminating between β-thalassemia trait and iron deficiency anemia

目的评估红细胞指数对鉴别婚育年龄轻型β地中海贫血和缺铁性贫血患者的作用，并利用这些指数建立基于反向传播（BP）人工神经网络的诊断模型以提高预测准确率。方法选择310例轻型β地中海贫血患者，其中男性131例．女性179例．年龄23～37岁，中位年龄25岁：125例缺铁性贫血患者作为对照，其中男性52例，女性73例，年龄20～34岁，中位年龄24岁。采用7种公式计算红细胞指数，分析它们对轻型β地中海贫血的预测灵敏度、特异度和约登指数，以此7种公式的计算值作为参数在MATLAB7．0程序上实现BP人工神经网络诊断模型．结果轻型β地贫血红细胞和血红蛋白（Hb）水平比缺铁性贫血高，平均红细胞体积（MCV）、平均红细胞血红蛋白（MCH）比后者低。7种指数中．Green and King指数（GKI）对轻型β地中海贫血的预测灵敏度（90．6％）及约登指数（78．6％）最高。BP人工神经网络诊断模型刈轻型β地中海贫血的诊断准确率为96．77％。结论GKI用于鉴别轻型β地中海贫血和缺铁性贫血可靠性相对较高。

Objective To evaluate the reliability of red blood cells（RBC） indices in discriminating between [3- thalassemia trait（13- TT） and iron deficiency anemia（IDA） for patients of fecundity age, and develop the back propagation（BP） artificial neural network- based diagnostic model using the indices for improving predictive accuracy. Methods A total of 310 patients with β-TT were enrolled, included 131 males and 179 females, which were aged 23 - 37 years old with a median age of 25. The 125 patients with IDA were set as control, included 52 males and 73 females, aged 20 - 34 years old with a median age 24. The RBC indices were calculated by 7 kinds of formulas. The sensitivity, specificity and Youden＇s index of RBC indices for predicting β-TT were also evaluated. The values of RBC indices were introduced to set up BP artifical neural network model with MATLAB 7.0 software. Results Compared RBC parameters between β-TT and IDA, the higher RBC and hemoglobin（Hb） levels were found in the former while higher mean corpuscular volume（MCV） and higher mean corpuscular hemoglobin（MCH） in the latter. Among 7 RBC indices, Green and King index （GKI） showed the highest sensitivity（90.6 %） and Youden＇s index was 78.6 %. The predicting accuracy of β-TT was 96.77 % with BP artificial neural network diagnostic model. Conclusion It is demonstrated that GKI appears reliable to discriminate between β-TT and IDA. The BP artificial neural network-based diagnostic model improves the predicting accuracy of β-TT.