超声导向中度急性呼吸衰竭患者无创呼吸支持失败的预测模型构建

Construction of a prediction model for non-invasive respiratory strategies failure of patients with moderate acute hypoxemic respiratory failure based on ultrasonic parameters

目的构建超声导向的中度急性呼吸衰竭患者无创呼吸支持失败的预测模型,并评价其价值。方法该研究为前瞻性观察性研究,纳入2020年10月至2023年1月入住苏州大学附属常熟医院重症医学科及急诊、呼吸病房初始治疗为高流量氧疗或无创通气的中度急性呼吸衰竭患者[100 mmHg≤氧合指数(PaO_(2)/FiO_(2))≤200 mmHg],治疗过程中需要气管插管或死亡判定为治疗失败。记录患者入院时及入院后24 h时生命体征、ROX指数、血气分析指标及超声参数。将研究对象按7:3随机(随机数字法)分组分为建模组和验证组,在建模组中对单因素分析差异有统计学意义的因素进行多因素logistic回归分析,构建治疗失败预测模型,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,计算净重新分类指数评价模型的预测能力,决策曲线分析评价净效益,并绘制预测模型诊断的列线图。结果共193例患者纳入研究,建模组137例,验证组56例,两组患者基本资料上差异无统计学意义,112例患者无创呼吸支持失败。单因素分析显示在失败组入院时及24 h PaO_(2)/FiO_(2)、膈肌移动度(DE)均低于成功组(P均<0.01),舒张末期右心室与左心室比值(RV/LV)均高于成功患者(P均<0.05),24 h呼吸频率(RR)及肺部超声评分(LUS)高于成功患者而ROX指数低于成功患者(P均<0.01),此外失败组患者血管活性药物的使用率也显著升高(P=0.001)。多因素logistic回归分析显示治疗失败的独立危险因素是血管活性药物使用(OR=4.709,P=0.012)、24 h较高的RR(OR=1.254,P=0.035)、LUS评分(OR=1.250,P=0.037)、RV/LV(OR=1.057,P=0.008),PaO_(2)/FiO_(2)(OR=0.950,P=0.001)、DE(OR=0.107,P=0.001)为保护因素。ROC分析显示建模组模型B的曲线下面积(AUC)大于模型A,差异有统计学意义(0.872 vs.0.928,P=0.009);验证组中模型A及模型B的曲线下面积分别为0.867,0.932(P=0.07)。校准图显示两个模型校准度均较好(P均>0.05)。净重新分类指数指数分析显示模型B较模型A预测能力提高(P<0.01)。决策曲线显示当预测值小于80%时,模型B的净效益优于模型A。结论超声参数联合临床指标在预测中度急性呼吸衰竭患者无创呼吸支持失败中具有较高的预测价值和准确性。

Objective To construct a predict model based on ultrasomics parameters that can identify moderate acute hypoxemic respiratory failure patients at risk of non-invasive respiratory strategies(NIRS)failure and evaluate its value.Methods This is a prospective observational trial.The patients with moderate acute respiratory failure(100 mmHg≤PaO_(2)/FiO_(2)≤200 mmHg)in intensive care unit(ICU),emergency and respiratory ward of Changshu Hospital Affi liated to Soochow University from Oct 2020 to Jan 2023 were included,NIRS failure is defined as death or intubation while on therapy.At admission time and 24 h after treatment vital signs,biological and ultrasound parameters were determined.The study subjects were randomly(random number)divided into a development group(70%)and a validation group(30%).Univariate and multivariate logistic regression was performed.The treatment failure prediction models were constructed according to ultrasonic parameters combined with clinical parameters.The models were also validated by ROC curves,calibration curves,NRI index and decision curve analysis(DCA).The nomograms were drawn.Results A total of 193 patients were included in the study,137 were allocated to the development group,and 56 to the validation group,there was no statistically significant difference between the two groups.NIRS failed in 112(58%)of 193 patients..Univariate analysis revealed that PaO_(2)/FiO_(2),DE at the time of admission and 24 h in the failure group were found to be statistically lower than the success group,RV/LV was higher(all P<0.05).RR,LUS at 24 h in the failure group were higher and ROX index was lower(all P<0.01).In addition,more patients in the failure group received vasopressors(P=0.001).Use of vasopressors(OR=4.709,P=0.012),RR(OR=1.254,P=0.035),LUS(OR=1.250,P=0.037),RV/LV(OR=1.057,P=0.008),PaO_(2)/FiO_(2)(OR=0.950,P=0.001),DE(OR=0.107,P=0.001)in the development group were independent risk factors for NIRS failure.ROC analysis revealed that model B achieved a larger area under curve(AUC)than model A in the development group,with their AUC values of 0.928 and 0.872(P=0.009),AUC of model A and model B in the validation group were 0.867 and 0.932 respectively(P=0.07).Two prediction models showed a good degree of calibration(all P>0.05).NRI analysis showed signifi cant improvement in the predictive power of model B(P<0.01).DCA showed that the model B had a good net benefi t between the threshold probabilities of 0-80%.Conclusions Ultrasomics parameters combined with Clinical parameters can effectively predict NIRS failure in moderate acute hypoxemic respiratory failure patients.