慢性乙型肝炎患者肝胆湿热证和肝郁脾虚证的决策树诊断模型初探

目的 建立基于临床数据的慢性乙肝患者典型证候的诊断模型，探索与慢性乙肝典型证候相关的症状、体征和实验室指标组合。 方法 对1 004例慢性乙肝患者进行证候学调查和实验室检测，包括 88项症状、舌脉象、20项体征和 14项实验室指标。然后选取典型证候的临床数据，联合采用Logistic回归、决策树、贝叶斯网络方法进行数据分析，构建能够有效区分不同证候的诊断模型。 结果 1 004例慢性乙肝患者中最为典型的证候为肝胆湿热证（340例，占33.86%）与肝郁脾虚证（366例，占36.45%）。首先采用Logistic回归分析得到具有统计学意义的16项症状与体征及3项实验室指标，然后采用决策树方法构建能够区分肝胆湿热证、肝郁脾虚证的诊断模型（辨证诊断准确率=74.36%），发现苔白、淡红舌、巩膜黄染、两目干涩、丙氨酸转氨酶（ALT）和 HBeAg是鉴别两证型的有效指标组合。采用贝叶斯网络分析的结果也验证了上述指标的重要性。 结论 本文建立了一种能够有效判别慢性乙肝患者肝胆湿热证、肝郁脾虚证的决策树诊断模型，为将基于经验的中医辨证转向基于数据模型的辨证提供了方法，有助于促进证候的客观化诊断。

Decision Tree Diagnostic Model of Gan-dan Damp-heat Syndrome and Gan-stangnancy Pi-deficiency Syndrome in Patients with Chronic Hepatitis B

Objective To establish a clinical data based practical diagnostic model for Chinese medicine syndrome differentiation in patients with chronic hepatitis B （CHB）, by way of seeking a combination of symp- toms, signs and laboratory criteria associated with the typical syndromes of the disease. Methods A syndroma- tologic and laboratory investigation on 88 symptoms, 20 signs, and 14 laboratory indexes, was conducted in 1 004 CHB patients. The clinical data of typical syndromes were selected and analyzed using Logistic regression analysis, decision tree and Bayesian network analysis in combination to establish a diagnostic model for effective- ly identifying the typical syndromes in CHB. Results The most typical syndromes revealed in the 1 004 CHB pa- tients were the Gan-dan damp-heat syndrome （340 patients, accounting for 33.86%） and the Gan-stagnancy with Pi-deficiency syndrome （366 patients, 36.45% ）. Based on the clinical data from patients with the two syn- dromes, 16 symptoms/signs and 3 laboratory indicators, provided with statistical significance, were selected u- sing Logistic regression analysis. Then, a diagnostic model for differential the two syndromes, which was proved to have an accurate diagnosic rate of 74.36%, was formed by decision tree method. It was found that white tongue coating, light red tongue, yellow sclera, eye dryness, blood levels of alanine aminotransferase and HBeAg appeared to be the effective combination of indexes that may be helpful to differentiate the two syn- dromes. And the significance of the above-mentioned indexes was also verified by a Bayesian network approach. Conclusion A decision tree model for diagnosing the two typical syndromes in CHB patients, the Gan-dan dampheat syndrome and the Gan-stagnancy with Pi-deficiency syndrome, was established, which could be helpful for shifting the diagnosis of syndrome from experience-based to the data-model-based form, to make the syndrome diagnosis more objectively.