利用可解释的人工智能方法研制一种算法，用于预测食管异物患者发生穿孔的风险。

收集浙江大学医学院附属第一医院2013年1月至2018年6月期间食管异物患者的临床数据。将427例食管异物患者分为训练组（298例）和验证组（129例），使用随机森林（RF）进行特征降维。使用7种机器学习方法来建立模型，包括极端梯度提升法（XGBoost）、RF、Logistic回归（LR）、支持向量机（SVM）、K近邻（KNN）、决策树（DT）和贝叶斯优化后的极端梯度提升法（BSXGBoost）。根据曲线下面积（AUC）选择最佳模型。使用决策曲线评估临床收益。采用沙普利加和解释（SHAP）方法来解释最佳模型。

BSXGBoost模型取得了最佳性能（AUC=0.837）。决策曲线显示，BSXGBoost模型的阈值概率在0.040至0.649（或0.69至0.85）时，其净收益优于其他模型。SHAP分析显示，主轴长度、嵌顿持续时间、年龄和主轴长度与小轴长度之比是对BSXGBoost模型贡献最大的4个变量。此外，SHAP力图展示构建的模型是如何将穿孔结局的个体化预测可视化的。

BSXGBoost模型对预测食管异物患者发生穿孔的风险有较高价值，SHAP方法可以提高机器学习模型的可解释性，帮助临床医生更好地理解结果背后的原因。

To investigate the risk factors of intracerebral hemorrhage (ICH) in patients with extracorporeal membrane oxygenation (ECMO).

A total of 105 patients received ECMO treatment at the Emergency Medicine Center of the First Affiliated Hospital with Nanjing Medical University from January 2017 to August 2020. Patients were divided into an ICH group (n = 14) and a non-ICH group (n = 91) according to whether there was cerebral hemorrhage during ECMO. The difference of baseline data and prognosis between these two groups was analyzed, and then the risk factors of ICH in ECMO patients were analyzed by multivariate Logistic regression and the receiver operating characteristic curve.

There were significant differences in the platelet count (U = 382.000, P = 0.016), creatinine (U = 419.000, P = 0.040), alanine aminotransferase (U = 396.000, P = 0.023), prothrombin time (U = 379.000, P = 0.015), activated partial thromboplastin time (U = 394.500, P = 0.022), continuous renal replacement therapy (χ² = 3.883, P = 0.049), platelet transfusion volume (U = 372.000, P = 0.010), plasma transfusion volume (U = 399.000, P = 0.015), minimum platelet count (U = 291.000, P = 0.001) and the longest value of prothrombin time during treatment (U = 341.500, P = 0.005) between the ICH group and the non-ICH group. Logistic regression showed that the minimum platelet count during ECMO [odds ratio = 0.614, 95% confidence interval (CI) (0.408, 0.923), P = 0.019] was a protective factor for ICH in patients with ECMO. The area under the curve for the minimum platelet count to predict ICH in ECMO patients was 0.799 [95%CI (0.667, 0.913), P < 0.001], and the optimal cut-off value was 49.5 × 10⁹/L, with sensitivity of 58.2% and specificity of 92.9%.

The minimum platelet count during ECMO is associated with ICH in patients with ECMO, and the lower the platelet count, the higher the risk of bleeding.

检测血清微小RNA-126（miR-126）在冠状动脉慢性完全闭塞（CTO）患者循环中的改变，并探讨其与CTO患者冠状动脉侧枝循环以及预后的关系。

回顾性分析2015年1月至2020年12月湖州市第一人民医院收治的造影结果为CTO病变的冠状动脉粥样硬化性心脏病患者。根据Rentrop分级，将65例患者分为侧枝循环良好组（30例）及侧枝循环不良组（35例），收集所有患者的临床一般资料、血常规、血生化结果。提取血清RNA，通过荧光定量PCR检测血清miR-126的表达水平。采用Pearson相关分析探讨血清miR-126与低密度脂蛋白胆固醇（LDL-C）、B型脑钠肽（BNP）、肌钙蛋白I（TnI）、左心室射血分数（LVEF）的关系，并绘制受试者工作特征（ROC）曲线分析血清miR-126对CTO侧枝循环形成的诊断价值。比较血清miR-126高表达者与低表达者的主要不良心血管事件和死亡情况。

侧枝循环良好组患者血清miR-126表达水平显著高于侧枝循环不良组[（25 ± 8）vs.（6 ± 4），t=8.745，P=0.001]，并且miR-126与LDL-C、BNP、TnI均呈负相关（r=-0.645、-0.589、-0.625，P=0.041、0.012、0.021），与LVEF呈正相关（r=0.618，P=0.038）。miR-126诊断CTO侧枝循环形成良好的曲线下面积为0.876[95%置信区间（0.820，0.939），P<0.001]，在最佳截断值15.17处敏感度为79.4%、特异度为88.7%。根据miR-126最佳截断值15.17将所有CTO患者分为高表达组（31例，miR-126 ≥ 15.17）和低表达组（34例，miR-126<15.17）。所有患者随访1年发现miR-126高表达组患者的主要心血管事件发生率[19.35%（6/31）vs. 35.29%（12/34）]和病死率[6.45%（2/31）vs. 14.71%（5/34）]均明显低于低表达组（χ²=2.057、1.150，P=0.015、0.028）。

miR-126参与CTO患者侧枝循环的形成，其降低提示预后不良。

To analyze risk factors of sepsis secondary to severe acute pancreatitis (SAP), and to provide lessons for its early clinical diagnosis and treatment.

From January 2013 to August 2020, a total of 362 patients diagnosed with SAP at the First Affiliated Hospital of Chongqing Medical University were enrolled. The patients were divided into a sepsis group (157 cases) and a non-sepsis group (205 cases) according to whether they had secondary sepsis. Univariate analysis and multivariate Logistic regression were used to analyze independent risk factors of sepsis secondary to SAP, and receiver operating characteristic curve analysis was performed to evaluate their value in predicting SAP with sepsis.

A total of 331 strains of pathogenic bacteria were detected in the sepsis group, with 258 strains of bacteria (77.95%). The leucocyte count, platelet count, platelet-to-lymphocyte ratio, albumin, aspartate aminotransferase (AST), creatinine, D-dimer, procalcitonin, acute physiology and chronic health evaluation (APACHE) Ⅱ score, sequential organ failure assessment score, length of hospital stay, length of ICU stay, and duration of mechanical ventilation were statistically significantly different in the sepsis and non-sepsis groups (all P < 0.05). The multivariate Logistic regression analysis showed that the AST [odds ratio (OR) = 1.004, 95% confidence interval (CI) (1.001, 1.008), P = 0.017], APACHEⅡ score [OR = 1.102, 95%CI (1.040, 1.166), P = 0.001], length of hospital stay [OR = 1.017, 95%CI (1.000, 1.033), P = 0.042], and duration of mechanical ventilation [OR = 1.080, 95%CI (1.027, 1.136), P = 0.003] were independent risk factors for SAP with sepsis. The area under the curve of AST, APACHEⅡ score, length of hospital stay, and duration of mechanical ventilation was 0.678, 0.723, 0.596, and 0.726, and their cut-off value at 60.5 U/L, 15.5, 28.5 d, and 4.5 d was most effective in predicting sepsis secondary to SAP. In addition, patients in the sepsis group had a significantly higher mortality rate compared to the non-sepsis group [9.55% (15/157) vs. 2.44% (5/205), χ² = 13.205, P = 0.001].

SAP patients with AST > 60.5 U/L, APACHEⅡ score > 15.5, hospitalization time > 28.5 d, and mechanical ventilation time > 4.5 d were prone to develop sepsis in the later period.

To compare the resuscitation effect and prognosis of patients with septic shock between using conventional hemodynamic index-oriented fluid resuscitation and using peripheral perfusion index (PPi)-oriented fluid resuscitation guided by conventional hemodynamic indexes, and to investigate the clinical value of using PPi to guide fluid resuscitation.

From October 2018 to June 2020, 40 adult patients with septic shock admitted to the Department of Critical Care Medicine of the Affiliated Hospital of Hangzhou Normal University were divided into a conventional group and an observation group (PPi < 1.4), with 20 cases in each group. All enrolled patients were monitored for pulse-indicated continuous cardiac output and PPi. Patients in both groups were treated according to clinical guidelines. Patients in the conventional group should achieve the following hemodynamic goals to terminate fluid resuscitation: central venous pressure 8 ~ 12 mmHg, urine output > 0.5 mL·kg^-1·h^-1, central venous oxygen saturation (ScvO₂) ≥ 70% and mean arterial pressure ≥ 65 mmHg. On the basis of achieving the above goals, patients in the observation group continued fluid resuscitation by the 2-5 rule with PPi < 1.4, and vasoactive drugs were adjusted to increase the mean arterial pressure until the target PPi ≥ 1.4 if it was not suitable to continue fluid replacement. The general information, the acute physiology and chronic health evaluation Ⅱ (APACHEⅡ) score after 0 and 6 h of resuscitation, and the heart rate, mean arterial pressure, cardiac output, ScvO₂, venous-to-arterial carbon dioxide difference (Pcv-aCO₂) and blood lactate after 0, 6, 24 and 48 h of resuscitation were recorded. The lactate clearance rate and ICU mortality rate at 6 and 24 h of resuscitation were calculated.

Before and after the two goal-oriented fluid resuscitation methods, there were no significant differences in the heart rate, mean arterial pressure, cardiac output, ScvO₂ and Pcv-aCO₂ at 0, 6, 24, and 48 h between the two groups (F = 0.197, 0.070, 2.658, 2.424, 1.687; P = 0.846, 0.958, 0.080, 0.119, 0.182), while the difference in blood lactate levels was statistically significant (F = 18.792, P < 0.001). Further pairwise comparison showed that the blood lactate levels at 6 and 24 h in the observation group were lower than those in the conventional group, and the lactate clearance rates at 6 and 24 h were higher (all P < 0.05). After 6 h of resuscitation, the APACHEⅡ score in the observation group was significantly lower than that in the conventional group [(21 ± 4) vs. (25 ± 5), t = 2.450, P = 0.019]. In addition, although the ICU deaths in the observation group were fewer than those in the conventional group, there was no significant difference in ICU mortality between the two groups (5/20 vs. 7/20, χ² = 0.476, P = 0.490).

Fluid resuscitation can improve the hemodynamic status of patients with septic shock, and the fluid resuscitation guided by conventional hemodynamic indexes combined with PPi can effectively reduce blood lactate, accelerate blood lactate clearance and decrease APACHEⅡ, which is expected to improve prognosis.

To screen key genes in sepsis-induced acute lung injury (ALI) by bioinformatics analysis.

The GSE10474 dataset was downloaded from the gene expression omnibus (GEO). The dataset included gene data of 13 patients with sepsis-induced ALI (ALI group) and 21 patients with sepsis (sepsis group). The limma package was used to screen differentially expressed genes between the two groups. Then gene ontology (GO) analysis and Kyoto encyclopedia of genes and genomes (KEEG) enrichment analysis were performed on these differentially expressed genes. The protein-protein interaction (PPI) network was established using the STRING database to identify the top 10 hub genes.

A total of 115 differentially expressed genes were identified from the GSE10474 dataset, including 65 up-regulated genes and 50 down-regulated genes. GO analysis showed that the differentially expressed genes of biological processes were mainly enriched in metal ion homeostasis, oxidative stress, and ionizing radiation. The cellular components were mainly concentrated in biofilms such as vacuolar membrane, Golgi apparatus, endoplasmic reticulum membrane, and lysosomal membrane. These genes were mainly associated with molecular functions such as biological transmembrane, ubiquitin-conjugating enzyme activity, protein tyrosine, serine and threonine kinases binding protein activity, and protein kinase inhibitory activity. KEGG enrichment analysis showed that the differentially expressed genes were mainly concentrated in the phospholipase signaling pathway, insulin signaling pathway, and T cell-mediated immune response and immune-related signaling pathways. The PPI network screened out the top 10 hub genes, including CD4, CD74, myeloid cell nuclear differentiation antigen (MNDA), triggering receptor expressed on myeloid cells 1 (TREM1), human leukocyte antigen DRA (HLA-DRA), cytohesin 1 interacting protein (CYTIP), coagulation factor XⅢA chain (F13A1), cystatin F (CST7), mitogen-activated protein kinase 1 (MAPK1), and cyclin dependent kinase inhibitor 1A (CDKN1A).

CD4, CD74, MNDA, TREM1, HLA-DRA, CYTIP, F13A1, CST7, MAPK1, and CDKN1A are key genes for sepsis-induced ALI, which can be used as new targets for clinical treatment and drug development.

To investigate the effect of right heart protective ventilation strategy on right heart function and prognosis in patients with acute respiratory distress syndrome (ARDS).

Using a prospective randomized controlled trial (RCT) method, 83 patients with ARDS who were hospitalized in Shaanxi Provincial People's Hospital from January 2021 to September 2022 were selected and randomly divided into a control group (n = 41) and a observation group (n = 42). The control group was treated with conventional lung protective ventilation, and the observation group was treated with right heart protective ventilation. Echocardiography was performed to evaluate right heart systolic and diastolic function. The ventilator parameters, blood gas analysis results, and prognostic indicators such as 28-day mortality, duration of mechanical ventilation, extracorporeal membrane oxygenation (ECMO), and prone position ratio and time were recorded respectively.

After 72 hours of mechanical ventilation, the respiratory compliance, tidal volume, respiratory rate, peak airway pressure, plateau pressure, driving pressure, Tei index, right ventricular fractional area change (RVFAC), right/left ventricular end-diastolic diameter (R/L) in the observation group were all improved as compared with the basal values (all P < 0.05). The plateau pressure, driving pressure, Tei index, and tissue Doppler velocities of the tricuspid annulus (E/e') were significantly lower, while the tricuspid annular plane systolic excursion (TAPSE), RVFAC, tricuspid annular motion (TAM), and Doppler velocities of the transtricuspid flow (E/A) were higher in the observation group than in the control group (all P < 0.05). The duration of mechanical ventilation in the observation group was shorter than that in the control group [(10 ± 3) d vs. (14 ± 6) d, t = 2.306, P = 0.026]. The 28-day mortality [38.1% (16/42) vs. 63.4% (26/41), χ² = 2.805, P = 0.099], ICU stay [(17 ± 6) d vs. (17 ± 7) d, t = 0.152, P = 0.873], amount of fluid [2 876.8 (1 062.9, 3 399.6) mL vs. 1 458.7 (-150.2, 3 821.4) mL, Z = 0.364, P = 0.750], proportion of prone position [47.6% (20/42) vs. 41.5% (17/41), χ² = 0.660, P = 0.360], ECMO [40.5% (17/42) vs. 31.7% (13/41), χ² = 1.850, P = 0.177], and proportion of vasoactive drug use [40.5% (17/42) vs. 48.8% (20/41), χ² = 1.122, P = 0.289] in the observation group and control group showed no statistically significant difference.

The right heart protective ventilation strategy given to ARDS patients is consistent with the effectiveness of the conventional lung protective ventilation strategy. It can reduce the incidence of right heart disorders, and has a protective effect on right heart function with less impact on both systolic and diastolic function.

To explore the clinical characteristics and prognosis-related factors in patients with sepsis-induced myocardial injury.

This study enrolled 69 patients with sepsis induced myocardial injury who were treated at the First Affiliated Hospital of Zhejiang University School of Medicine and Quzhou Kecheng District People's Hospital from December 2020 to December 2021. These patients were divided into the survival group (n = 45) and the death group (n = 24) according to the 28-day survival after discharge. The general data, laboratory-related indicators, length of hospital stay and sequential organ failure assessment (SOFA) score between the two groups were compared. The multivariate Logistic regression model was used to identify the independent risk factors for death in patients with sepsis-induced myocardial injury, and the predictive efficacies of relevant indicators for death in patients with sepsis-induced myocardial injury were evaluated by drawing receiver operating characteristic (ROC) curves.

The infection site of patients with sepsis-induced myocardial injury was mainly located at the lungs. The proportions of previous hypertension [62.50% (15/24) vs. 33.33% (15/45), χ² = 5.418, P = 0.024] and chronic lung disease [20.83% (5/24) vs. 4.44% (2/45), χ² = 4.612, P = 0.032], the width of red blood cell distribution[(15.0 ± 1.6)% vs. (14.0 ± 1.5)%, t = 2.426, P = 0.018], SOFA score [(9 ± 4) vs. (6 ± 4), t = 3.457, P = 0.001] and serum interleukin-6 (IL-6) level [763.00 (234.34, 2 656.94) ng/L vs. 94.28 (37.03, 262.98) ng/L, t = 4.158, P < 0.001] at admission in the death group were significantly higher than those in the survival group. However, the body temperature at admission of the patients in the death group was significantly lower than that in the survival group[(37.2 ± 1.2) ℃ vs. (37.8 ± 1.0) ℃, t = 2.312, P = 0.024]. The multivariate Logistic regression analysis revealed that SOFA score, body temperature, serum IL-6 at admission and previous hypertension were all important factors for predicting death in patients with sepsis-induced myocardial injury (all P< 0.05). The ROC curve analysis showed that the combined detection of the above four indicators [area under the cure (AUC) = 0.868, 95% confidence interval (CI) (0.783, 0.952), P < 0.001] and serum IL-6 [AUC = 0.806, 95%CI (0.694, 0.917), P< 0.001] both had good predictive efficacies for death in patients with sepsis-induced myocardial injury.

SOFA score, body temperature, serum IL-6 at admission and previous hypertension can predict the clinical outcomes of patients with sepsis-induced myocardial injury.