To explore the prognostic factors for patients with drug-induced liver failure (DILF) and construct a logistic regression model (LRM). Methods: A retrospective analysis of clinical data was performed in 183 hospitalized patients, who were diagnosed with DILF in Xiangya Hospital, the Second Xiangya Hospital and the Third Xiangya Hospital, Central South University from January 2009 to January 2018. The patients were divided into an improved group (n=67) and an ineffective group (n=116) according to their prognosis. Univariate analysis was performed to screen for possible prognostic factors such as age, Tbil, SCr, PT and complications. According to the results of univariate analysis, the multivariate analysis was performed to determine the independent prognostic factors and construct a LRM. The LRM was compared with the model for end-stage liver disease (MELD), the predictive value of LRM and MELD was evaluated by receiver operating characteristic curve (ROC), the parameters such as area under the ROC (AUC) and total accuracy were compared between the 2 models and verified by another independent sample. Results: According to univariate analysis, there was significant differences in age, clinical type, hepatic encephalopathy, hepatorenal syndrome, WBC count, the ratio of aspartic acid transaminase (AST) to glutamine transaminase (ALT) (AST/ALT), Tbil, SCr, PT and alpha-fetoprotein (AFP) between the 2 groups (all P<0.05). Multivariate analysis revealed that: AFP, PT, AST/ALT, hepatic encephalopathy and hepatorenal syndrome were independent prognostic factors for DILF, which could be applied to constructing a LRM. The AUC of LRM and MELD was 0.917 (95% CI 0.876 to 0.959) and 0.709 (95% CI 0.633 to 0.786) respectively, the total accuracy rate of prediction for the LRM and the MELD was 86.7% and 68.3% respectively, there was significant difference in AUC and total accuracy rate between the LRM and the MELD (P<0.05). LRM was superior to MELD. Conclusion: AFP, PT, AST/ALT, hepatic encephalopathy and hepatorenal syndrome were independent prognostic factors for DILF; the LRM can well predict the prognosis in the DILF patients, which is superior to the MELD.
目的：研究药物性肝衰竭(drug-induced liver failure，DILF)患者的预后影响因素，并构建其logistic回归模型(logistic regression model，LRM)。方法：回顾性分析2009年1月至2018年1月在中南大学湘雅医院、湘雅二医院及湘雅三医院感染病科收治的183例DILF患者的临床资料，根据其预后分为好转组(n=67)及无效组(n=116)。使用单因素分析对年龄，Tbil，SCr，PT及并发症等可能的预后影响因素进行初筛，基于单因素分析结果，进一步使用多因素分析筛选出独立的预后因素并构建LRM模型。将LRM模型与终末期肝病(model for end-stage liver disease，MELD)模型进行比较，采用受试者工作曲线(receiver operating characteristic curve，ROC)评估LRM模型和MELD模型的预测价值，比较两模型间ROC曲线下面积(area under the ROC curve，AUC)及总正确率等参数的优劣，并使用独立样本进行验证。结果：单因素分析提示：年龄，临床分型，肝性脑病，肝肾综合征，WBC计数，天门冬氨酸转氨酶/谷丙氨酸转氨酶(aspartic acid transaminase/glutarine transaminase，AST/ALT)比值，Tbil，SCr，PT和甲胎蛋白(alpha-fetoprotein，AFP)等指标在两组间的差异均有统计学意义(均P<0.05)。多因素分析提示：AFP，PT，AST/ALT比值，肝性脑病和肝肾综合征为DILF患者独立的预后影响因素，可用于构建LRM模型。LRM模型和MELD模型的AUC分别为0.917(95% CI：0.876~0.959)和0.709(95% CI：0.633~0.786)，总正确率分别为86.7%和68.3%，差异有统计学意义(P<0.05)。结论：AFP，PT，AST/ALT比值，肝性脑病和肝肾综合征为DILF患者独立的预后影响因素；LRM模型能较准确地预测DILF患者的短期预后，其应用价值优于MELD模型。.