Objective: To determine the predictive value of dynamic changes of neutrophil/lymphocyte ratio (NLR) combined with the model of end-stage liver disease (MELD) score in patients with acute-on-chronic hepatitis B liver failure. Methods: Patients with acute-on-chronic hepatitis B liver failure who were hospitalized in the Department of Hepatology of Qilu Hospital of Shandong University from January 2010 to July 2023 were retrospectively enrolled. According to the clinical outcomes of patients within 30 days of admission, they were divided into the survival group and the death group. The dynamic changes in NLR and initial values on day 3, 5, 8, and 12 in two groups were analyzed for the diagnostic value of 30-day prognosis in patients with acute-on-chronic hepatitis B liver failure. Logistic regression analysis and machine learning XGBoost algorithm were used to evaluate the risk factors influencing the prognosis of patients at 30 days. Receiver operating characteristic(ROC) curve was used to evaluate the diagnostic value of NLR and initial value change combined with MELD score on day 12 of admission in patients with chronic acute hepatitis B liver failure. Results: A total of 243 patients were enrolled in the study, including 145 patients in the survival group [115 males, 30 females, aged 25-74 (47±11)] and 98 patients in the death group [80 males, 18 females, aged 22-80 (49±13) ]. The median initial NLR of survival group and death group were 3.5 (2.1, 5.3) and 4.9 (2.9, 8.3), respectively, and the difference was statistically significant (P=0.003). The variation of NLR from the initial value on day 3, 5, 8, and 12 in the survival group [1.6 (0, 4.3), 1.9 (-0.2, 4.1), 2.0 (-0.1, 4.3) and 2.9 (0.3, 7.0), respectively] were lower than that in the death group [3.2 (0.9, 7.5), 5.1 (1.8, 7.6), 5.8 (2.0, 10.6) and 9.6 (3.5, 16.4), respectively] (all P<0.001). Logistic regression multivariate analysis showed that the changes in NLR on the 12th day and initial value (OR=1.07,95%CI:1.01-1.14, P=0.014), the changes in NLR on the 3rd day and initial value (OR=2.71, 95%CI: 1.32-5.55, P=0.007), the initial value of NLR (OR=1.18,95%CI:1.01-1.37,P=0.035) and fibrinogen (OR=0.21,95%CI:0.05-0.96,P=0.044) were related factors for death within 30 days. Machine learning XGBoost algorithm showed that the weight of the change between the NLR on the 12th day and the initial value was the highest. The area under the ROC curve of the combined MELD score was 0.812 (95%CI: 0.728-0.895), the specificity was 67.78%, and the sensitivity was 82.35%. Conclusion: Dynamic change of NLR combined with MELD score has high predictive value for the short-term prognosis of patients with acute-on-chronic hepatitis B liver failure.
目的： 研究中性粒细胞与淋巴细胞比值（NLR）的动态变化联合终末期肝病模型（MELD）评分对慢加急性乙型肝炎肝衰竭患者短期预后的预测价值。 方法： 回顾性纳入2010年1月至2023年7月在山东大学齐鲁医院肝病科住院的慢加急性乙型肝炎肝衰竭患者，根据入院30 d内患者的临床转归，分为生存组和死亡组。分析两组患者入院后第3、5、8和12天NLR与初始值（入院后第1天）的变化量的动态变化对慢加急性乙型肝炎肝衰竭患者30 d内预后的预测价值。采用logistic回归分析、机器学习XGBoost算法评估患者30 d内预后的影响因素。采用受试者工作特征（ROC）曲线评估入院第12天NLR与初始值变化量联合MELD评分对慢加急性乙型肝炎肝衰竭患者短期预后的预测价值。 结果： 共243例患者纳入研究，生存组145例，男115例，女30例，年龄25~74（47±11）岁；死亡组98例，男80例，女18例，年龄22~80（49±13）岁。生存组与死亡组的初始NLR的中位值分别为3.5（2.1，5.3）和4.9（2.9，8.3），差异有统计学意义（P=0.003）；生存组NLR第3、5、8和12天与初始值的变化量［分别为1.6（0，4.3）、1.9（-0.2，4.1）、2.0（-0.1，4.3）、2.9（0.3，7.0）］均低于死亡组［分别为3.2（0.9，7.5）、5.1（1.8，7.6）、5.8（2.0，10.6）、9.6（3.5，16.4）］（均P<0.001）。多因素logistic回归分析结果显示NLR第12天与初始值的变化量（OR=1.07，95%CI：1.01~1.14，P=0.014）、NLR第3天与初始值的变化量（OR=2.71，95%CI：1.32~5.55，P=0.007）、NLR初始值（OR=1.18，95%CI：1.01~1.37，P=0.035）和纤维蛋白原（OR=0.21，95%CI：0.05~0.96，P=0.044）是影响患者30 d内死亡的相关因素。机器学习XGBoost算法发现，NLR第12天与初始值的变化量联合MELD评分对慢加急性乙型肝炎肝衰竭患者短期预后有预测价值，ROC曲线下面积为0.812（95%CI：0.728~0.895），特异度为67.78%，灵敏度为82.35%。 结论： NLR动态变化联合MELD评分对慢加急性乙型肝炎肝衰竭患者的短期预后具有较高的预测价值。.