无创性预测门脉高压出血的Logistic回归模型随访比较

A following-up study of logistic regression models for non-invasive predicting risk of variceal bleeding

目的探讨利用多普勒等测定的参数,建立回归方程预测门脉高压出血危险性的最佳组合。方法肝硬化63例,首期30例采用多普勒测定门静脉(PV)、脾静脉(SPV)及肠系膜上静脉(SMV)血流量参数,计算脾/门静脉流量比(Qspv/Qpv)并测定血小板平均体积(MPV),多参数不同组合建立回归方程,即方程I(Qspv、Qsmv及MPV)。方程Ⅱ(Qspv及MPV)、方程Ⅲ(Qspv/Qpv)。以建立的方程前瞻性对另33例患者行半年的随访,分析各支方程的预测价值。结果方程I预测出血的灵敏度85·71%,特异性81·25%,阳性及阴性预测值分别为77·78%及80·00%。方程Ⅱ灵敏度73·33%,特异性73·33%,阳性及阴性预测值分别为73·68%及78·57%。方程Ⅲ灵敏度66·67%,特异性61·11%,阳性及阴性预测值分别为76·92%及65·00%。剔除病程5年以上者,方程I阳性及阴性预测值分别为90%和84·62%。结论以Qspv、Qsmv及MPV建立的回归模型对出血预测效率较高,如Qsmv测定不成功,以Qspv及MPV建立的方程可作参考。

Objective To explore an optimal regression model based on doppler-measured parameters for non-invasive predicting variceal bleeding. Methods Sixty-three patients of hepatic cirrhosis were enrolled. Qpv, Qspv and Qsmv were determined hy doppler. Parameters from the first 30 patients were collected to create logistic regression models. Model Ⅰ was based on Qspv, Qsmv and MPV, model Ⅱ based on Qspv, and MPV, and model Ⅲ based on Qspv,/Qpv. These models were used to prospectively predict bleeding in another 33 patients with following-up for 6 months. Results Model I had a sensitivity of 85.71%, specificity of 81.25 % and a positive predicting value （PPV） of 77.78 % , negative predicting value （NPV） of 80.00% in predicting bleeding： model Ⅱ had a sensitivity of 73.33%, specificity of 73.33%, PPV of 73.68%, NPV of 78.57%; model Ⅲ had a sensitivity of 66.67%, specificity of 61.11%, PPV of 76.92%, NPV of 65.00%. Excluding those with cirrhosis over 5 years, model Ⅰ had a PPV up to 90% and NPV 84.62% . Conclusion Regression model based on Qspv, Qsmv and MPV had a higher predicting value. If Qsmv fails to be determined in patients, model based on Qspv and MPV could be an alternative.