探讨淋巴细胞/凝血酶原时间比值（LPR）、纤维蛋白原/白蛋白比值（FARI）等实验室指标对肝衰竭预后的影响，并构建新的肝衰竭预后模型。

回顾性分析2017年6月至2020年12月中山大学附属第三医院收治的1 114例肝衰竭患者临床资料。患者均签署知情同意书，符合医学伦理学规定。其中男899例，女215例；年龄18～90岁，中位年龄46岁。收集肝衰竭患者的首次入院时血液学指标、肝功能指标、凝血功能指标以及基本临床特征。以肝衰竭确诊后90 d内死亡为患者预后结局指标，按7∶3的比例将纳入的研究对象分为建模组与验证组，再采用Lasso回归分析筛选肝衰竭预后影响因子并用十折交叉法进行验证，将影响因子纳入Logistic回归构建预测模型。通过ROC曲线下面积（AUC）评估预测模型的区分度，Hosmer-Lemeshow（H-L）指数评价校准度。

1 114例肝衰竭中90 d内死亡317例，病死率28.46%（317/1 114）。Lasso回归筛选LPR、凝血酶原活动度（PTA）、Na、TB、FARI、PT、中性粒细胞绝对值（NEU）、年龄（Age）是肝衰竭预后的重要影响因素，构建预测模型LPTFA，Logit P =-1.75-6.57×LPR（10⁹/L·sec）-0.04×PTA（%）-0.006×Na（mmol/L）+0.001×TB（μmol/L）+0.08×FARI（%）+0.009×PT（sec）+0.03×NEU（10⁹/L）+0.04×Age。绘制列线图预测肝衰竭患者90 d内死亡率，该模型特异度为0.74，敏感度为0.56。该模型AUC为0.704（95%CI：0.660～0.740），明显高于MELD评分AUC的0.612（95%CI：0.570～0.650）（Z=4.207，P<0.001）。验证组中LPTFA模型AUC为0.686（95%CI：0.62～0.75），亦明显优于MELD评分AUC的0.563（95%CI：0.490～0.640） （Z=3.143，P=0.001 7）。运用H-L指数验证，建模组P=0.41，验证组P=0.19，两组H-L指数均大于0.05，说明模型校准度较高。

与MELD评分相比，基于LPR及FARI建立的肝衰竭预后预测模型具有更好的预测价值，是较为可靠的预测模型。

To evaluate the effects of laboratory indexes, such as lymphocyte-prothrombin time ratio (LPR) and fibrinogen-albumin ratio index (FARI), on clinical prognosis of liver failure, and to establish a novel prognostic model for liver failure.

Clinical data of 1 114 patients with liver failure admitted to the Third Affiliated Hospital of Sun Yat-sen University from June 2017 to December 2020 were retrospectively analyzed. The informed consents of all patients were obtained and the local ethical committee approval was received. Among them, 899 patients were male and 215 female, aged from 18 to90 years, with a median age of 46 years. Hematological indexes, liver function indexes, coagulation function indexes and baseline clinical characteristics of patients with liver failure upon the initial admission were collected. The 90-d mortality after the diagnosis of liver failure was considered as the prognostic outcome. All patients were divided into the modeling and validation groups according to the ratio of 7:3. The factors affecting clinical prognosis of liver failure were screened by Lasso regression analysis and subject to 10-fold cross-validation. The influencing factors were included in Logistic regression analysis to establish a prediction model. The area under the ROC curve (AUC) was used to evaluate the discrimination degree of this prediction model, and the Hosmer-Lemeshow (H-L) index was employed to evaluate the calibration degree.

Among1 114 liver failure patients, 317 cases died within postoperative 90 d, with a mortality rate of 28.46%(317/1 114). Lasso regression analysis showed that LPR, prothrombin time activity (PTA), Na, TB, FARI, PT, absolute value of neutrophil (NEU) and age (Age) were critical prognostic factors of liver failure. A prediction model of LPTFA was established. Logit P=-1.75-6.57×LPR(10⁹/L·sec)-0.04×PTA(%)-0.006×Na(mmol/L)+0.001×TB(μmol/L)+0.08×FARI(%)+0.009×PT(sec)+0.03×NEU(10⁹/L)+0.04×Age. A nomogram was delineated to predict the 90-d mortality of patients with liver failure. The specificity and sensitivity of this model were 0.74 and 0.56. The AUC of this model was 0.704 (95%CI: 0.660-0.740), significantly higher than 0.612 (95%CI: 0.570-0.650) of MELD score (Z=4.207, P<0.001). In the validation group, the AUC of LPTFA model was 0.686 (95%CI: 0.620-0.750), significantly higher than 0.563 (95%CI: 0.49-0.64) of MELD score (Z=3.143, P=0.001 7). H-L index validated that P=0.41 in the modeling group, and P=0.19 in the validation group, and the H-L indexes in two groups were both greater than 0.05, indicating that the model calibration degree was relatively high.

Compared with MELD score, the prediction model for liver failure established based on LPR and FARI has better prediction value and is more reliable.