急性胃肠损伤分级联合qSOFA评分在脓毒症诊断中的价值

目的:构建急性胃肠损伤（acute gastrointestinal injury,AGI）分级联合qSOFA评分诊断脓毒症的预测模型，并评价其价值。方法:为前瞻性观察性研究，纳入2018年9月至2019年9月入住常熟市第一人民医院普通病房的感染或疑似感染患者，排除年龄小于18岁、妊娠、中途放弃治疗、入院后3 d内死亡患者。记录感染前48 h内至发病后24 h的基本生命体征、实验室化验结果、AGI分级情况，根据患者是否诊断为脓毒症分为脓毒症组和非脓毒症组。将研究对象按7∶3比例随机分为建模组和验证组，在建模组中对单因素分析有统计学差异的因素进行多因素logistic回归分析。分别建立诊断模型A（qSOFA）、模型B（AGI分级联合qSOFA评分）以及模型C（多因素分析中有统计学差异的变量为参数）。在验证组进行验证，通过ROC曲线评价模型的诊断效果，绘制校准曲线评价一致性，使用决策曲线分析评价净效益，并绘制各预测模型诊断的列线图。结果:共2 553例患者纳入研究，建模组1 789例，验证组764例，两组患者基本情况差异无统计学意义，其中共326例患者发生了脓毒症。单因素分析显示年龄、性别、感染源、体温、心率、呼吸急促、意识改变、严重水肿、高血糖、白细胞数量、CRP、PCT、低血压、低氧、急性少尿、凝血功能异常、高乳酸血症、毛细血管充盈受损或皮肤花斑、AGI分级及qSOFA对脓毒症的发生有显著影响（均 P<0.01）。多因素Logistic回归分析，显示预测脓毒症的危险因素为年龄（ OR=1.027, P<0.01），感染源（ OR=2.809, P=0.03），低血压（ OR=35.449, P<0.01）、低氧血症（ OR=57.018, P<0.01），AGI分级（ OR=19.313, P<0.01）。ROC分析显示建模组中模型A、B、C的曲线下面积（AUC）分别为0.784，0.944，0.971，敏感度分别为63.9%，89.5%，97.5%，特异度分别为90.8%，90.3%，88.1%，验证组中，模型A、B、C的AUC为0.832，0.975，0.980，敏感度分别为72.7%，90.9%，96.6%，特异度分别为92.2%，94.5%，92.8%。模型B、模型C的AUC明显大于模型A（ P<0.01），而在验证组之间B、C（ P=0.684）。模型A在验证组中的校准度较差，精确性低，有漏诊脓毒症的风险（ P=0.044）。决策曲线分析显示模型B及模型C的净效益优于模型A。 结论:AGI分级联合qSOFA评分在脓毒症诊断中的具有较高的预测价值和准确性。

The value of acute gastrointestinal injury grading combined with qSOFA score in the diagnosis of sepsis

Objective:To develop a prediction model of acute gastrointestinal injury (AGI) grading combined with qSOFA score for the diagnosis of sepsis, and evaluate its value.Methods:This was a prospective observational study. The patients with infection or suspected infection in the General Ward of Changshu Hospital Affiliated to Soochow University from September 2018 to September 2019 were included. Patients younger than 18 years, pregnant, abandoned treatment and died within 3 days after admission were excluded. Clinical characteristics, laboratory test results and AGI grading from 48 h before the infection to 24 h after the onset of infection were recorded. The patients were divided into the sepsis and non-sepsis groups according to whether they were diagnosed with sepsis. The patients were allocated randomly to a modeling cohort and a validation cohort with a ratio of 7:3. Univariate and multivariate logistic regression analyses were used to analyze the relevant risk factors for sepsis in the modeling cohort. Three types of diagnostic models were constructed in the modeling cohort: model A (qSOFA model), model B (the combined model of AGI grading and qSOFA score), and model C (the combined model of clinical parameters). The clinical usefulness of the diagnostic models was assessed by receiver operating characteristic curve (ROC), calibration curve and decision curve analysis (DCA) in the validation cohort. The nomograms were developed based on these models.Results:A total of 2 553 patients were enrolled in the study, 1 789 patients in the modeling cohort and 764 patients in the validation cohort. and 326 were diagnosed with sepsis. There was no statistical difference in the basic conditions of patients in the two groups. Univariate analysis showed that age, gender, the source of infection, temperature, heart rate, polypnea, changes in consciousness, severe edema, hyperglycemia, white blood cell, C-reactive protein and procalcitonin, hypotension, hypoxemia, acute oliguria, coagulation disorders, hyperlacticemia, capillary filling damage or piebaldskin, AGI grading and qSOFA score were significantly correlated with sepsis (all P<0.01). Multivariate logistic regression analysis showed that age ( OR=1.027, P<0.01), source of infection ( OR=2.809, P=0.03), hypotension ( OR=35.449, P<0.01), hypoxemia ( OR=57.018, P<0.01), and AGI grading ( OR=19.313, P<0.01) were significantly associated with sepsis. ROC analysis showed that the area under the curve (AUC) of model A, B and C were 0.784, 0.944 and 0.971 in the modeling cohort, and 0.832, 0.975 and 0.980 in the validation cohort, respectively. The sensitivities were 63.9%, 89.5% and 97.5% in the modeling cohort, and 72.7%, 90.9% and 96.6% in the validation cohort; and the specificities were 90.8%, 90.3% and 88.1% in the modeling cohort, and 92.2%, 94.5% and 92.8% in the validation cohort, respectively. AUC of model B and C were significantly higher than that of model A ( P<0.01). Model A in the validation cohort was poorly calibrated, with low accuracy and high risk of missed sepsis diagnosis ( P=0.044). The net benefits of model B and C were better than that of model A. Conclusions:AGI grading combined with qSOFA score has a high predictive value and accuracy in the diagnosis of sepsis.