【摘要】 目的应用决策树模型(CART),纳入总淋巴细胞(TLC)计数、血红蛋白(HGB)等多个血常规指标,研究未抗病毒治疗HIV感染者TLC与CD4~+T淋巴细胞(简称CD4细胞)计数的相关性,并与受试者工作特征曲线(简称ROC曲线)常规分类方法比较。方法选取昆山市未抗病毒治疗HIV感染者297例,采集433份血样标本,检测血常规指标和CD4细胞计数。将血常规指标与CD4细胞计数进行相关性分析,以CD4细胞计数≤350个/uL、≤500个/uL为临界点,分别计算CART、常规ROC曲线分类方法的灵敏度、特异度、阳性预测值(PPV)、阴性预测值(NPV)和约登指数。结果 TLC、HGB、白细胞(WBC)计数、红细胞(RBC)计数、血小板(PLT)计数、中性粒细胞(WLCC)计数与CD4细胞计数均显著相关(P<0.05),纳入TLC、HGB、RBC、WBC指标,CART模型分类CD4细胞计数≤350个/uL的灵敏度、特异度、PPV、NPV和约登指数分别为36.4%、95.3%、84.2%、68.6%和0.317;纳入TLC、HGB、RBC指标,CART模型分类CD4细胞计数≤500个/uL的灵敏度、特异度、PPV、NPV和约登指数分别为92.0%、40.9%、78.0%、69.2%和0.329;两个CART模型均略优于ROC曲线分类方法。结论应用CART模型纳入TLC、HGB等多个血常规指标能有效预测CD4细胞计数,可在资源有限地区用于未治疗HIV感染者疾病进展监测。更多还原

【Abstract】 Objective To explore the relationship between TLC and CD4+T cell count among non-antiretroviral therapy HIV-1infected patients,and compare the results with those from traditional ROC method by using decision tree algorithm(CART)based on total lymphocyte count(TLC),hemoglobin(HGB)and other routine blood indexes.Methods A prospective study was conducted with 433 blood samples collected from 297 HIV-1positive individuals without receiving anti-retroviral therapy.TLC and other correlated routine blood cell count indexes were used to develop CART models based on CD4+T cell count≤350cells/!L,and ≤500cells/!L respectively.Sensitivity,specificity,PPV(positive predictive value),NPV(negative predictive value),Youden index were calculated and compared with those from ROC method.Results CD4+T cell count was significantly associated with TLC,HGB,white blood cell count(WBC),red blood cell count(RBC),blood platelet cell count(PLT),and neutrophil cell count(WLCC).The sensitivity,specificity,PPV,NPV and Youden index from CART Model for CD4+T count≤350cells/!L were 36.4%,95.3%,84.2%,68.6%and 0.317 respectively,and for CD4+T count≤500cells/!L were 92.0%,40.9%,78.0%,69.2% and 0.329 respectively.CART models appeared superior to traditional ROC method.Conclusion The study shows that CART model can effectively predict CD4+T cell count with reference to TLC,HGB and other routine blood indexes.CART model can be applied as a useful tool for CD4+T count prediction in non-antiretroviral therapy HIV-1 positive subjects in resource limited settings.更多还原