【摘要】 目的应用决策树模型（CART）研究MSM人群HIV感染者总淋巴细胞计数（TLC）与CD₄⁺T细胞计数的相关性,并与常规ROC分类方法比较。方法选取137例MSM人群HIV感染者,所有病例均未接受抗病毒治疗,采集203例血样标本,检测血常规指标和CD₄⁺T细胞计数,把与CD₄⁺T细胞计数显著相关的血常规指标纳入CART模型,计算灵敏度、特异度、阳性预测值（PPV）、阴性预测值（NPV）和约登指数。结果 203例血样标本的总淋巴细胞计数（TLC）与CD₄⁺T细胞计数分别为（1 771.9±720.3）cells/μL、（410.9±201.4）cells/μL,两者呈显著正相关（r=0.603,P<0.001）。CART模型分类CD₄⁺T细胞计数≤350 cells/μL的灵敏度、特异度、PPV、NPV和约登指数分别为66.7%、70.4%、58.4%、77.2%和0.371,CART模型分类CD₄⁺T细胞计数≤500 cells/μL的灵敏度、特异度、PPV、NPV和约登指数分别为93.8%、53.5%、83.4%、77.5%和0.473;两个CART模型均优于ROC分类方法。结论 TLC与CD₄⁺T细胞计数显著相关,TLC分类预测MSM人群HIV感染者CD₄⁺T细胞计数具有良好的灵敏度和特异度,在资源有限地区可以用于MSM人群艾滋病疾病进展监测,应用CART模型纳入多个指标联合预测的效果更佳。更多还原

【Abstract】 Objective To develop a more effective method based on decision tree algorithm（ CART）,in order to explore the correlation between total lymphocyte count（ TLC） and CD₄⁺Tcount in MSM HIV-1 infected patients,and to compare the results from CART with that from traditional ROC method. Methods In a prospective study,203 blood samples were collected from 137 HIV-1 positive MSM individuals. All included subjects had no access to any anti-retroviral therapy. CD₄⁺Tcount and other correlated complete blood cell count index were included to develop CART models. Sensitivity,specificity,PPV（ positive predictive value）,NPV（ negative predictive value） and Youden index were calculated form the models and then were compared with those from ROC method. Results The TLC and CD₄⁺Tcount were 1771. 9 ± 720. 3 cells / L and 410. 9 ± 201. 4cells / L,respectively,and TLC was positively correlated with CD₄⁺Tcount（ r = 0. 603,P = 0. 000）. The sensitivity,specificity,PPV,NPV and Youden index in CART models for CD₄⁺Tcount（ ≤350 cells / L） were 66. 7%,70. 4%,58. 4%,77. 2% and0. 371,respectively. Similarly,the sensitivity,specificity,PPV,NPV and Youden index in CART models for CD₄⁺Tcount（ ≤500cells / L） were 93. 8%,53. 5%,83. 4%,77. 5% and 0. 473,respectively. CART models performed better than traditional ROC method. Conclusion The study shows that TLC is significantly correlated with CD₄⁺Tcount. Considering the sensitivity and specificity,TLC could be applied as a surrogate marker of CD₄⁺Tcount among MSM HIV-1 positive subjects in resource-limited settings. The CART models are superior to traditional ROC,especially in multiple factors joint models.更多还原