1 Introduction
Extracorporeal membrane oxygenation (ECMO) has played vital roles in the treatment of severe COVID-19 pneumonia. A recent worldwide meta-analysis including 1896 critically ill patients with COVID-19 who received ECMO support showed that in-hospital mortality was reduced to 37% compared with 50%–80% in patients without ECMO support. However, the mortality rate differed greatly between different centers [
1–
4]. Heterogeneity of experience among ECMO centers may have played an important role [
5]. The intensive care unit (ICU) is the main site in which ECMO-supported patients receive treatment. However, treatment in the ICU is associated with frequent adverse medical events because of the high intensity of therapeutic activities per day, the high quantity and complexity of medical data, and sudden changes in patients’ clinical conditions. The quality of ICU has a crucial effect on patients’ prognosis [
6,
7].
In 2015, the China National Critical Care Quality Control Center (China-NCCQC) released 15 specific national clinical quality control indicators for critical care medicine. These 15 quality control indicators were recently found to be good predictors of the prognosis of critically ill patients, and a quality scoring system was defined to further optimize the system [
8]. In addition, certain capacity parameters of the ICU, such as patient-to-bed ratio and nurse-to-bed ratio, were proven to be substantially associated with the incidence of ventilator-associated pneumonia [
9]. ECMO was treated as a high-tech salvage strategy for critically ill patients, and its performance and management were a comprehensive reflection of the ICU level. The authors hypothesized that quality control plays an important role in the prognosis of ECMO-supported patients. Therefore, China-NCCQC carried out the ECMO quality improvement action (EQIA) study to investigate the associations among ICU capacity parameters, quality control indicators, and in-hospital mortality of V-V ECMO-supported patients.
2 Methods
2.1 Study population
The NCIS database of the National Health Commission of the People’s Republic of China was designed to collect detailed information of patients from 1700 tertiary hospitals. With reference to the authors’ previous research [
10], information from ECMO-supported patients included in the database from 1 January 2017, to 31 December 2019, were screened. The following indications for V-V ECMO initiation were in accordance with the Extracorporeal Life Support Organization (ELSO) guidelines [
11,
12]: severe acute respiratory distress syndrome and refractory hypoxemia (PaO
2/FiO
2 of < 60 mmHg for > 6 h or < 50 mmHg for > 3 h) or severe hypercapnic respiratory failure (pH of < 7.20 with PaCO
2 of > 80 mmHg for > 6 h) after maximizing traditional therapies, particularly the use of prone positioning. Patients with relative contraindications, such as central nervous system hemorrhage, systemic bleeding, and contraindications to anticoagulation, were rigorously excluded by intensivists. Among the 1700 tertiary hospitals in the database, 318 hospitals conducted V-V ECMO from 2017 to 2019. All these patients were enrolled in the study, and all information available in the database was recorded. The data collection was approved by the National Health Commission of China, and the study was approved by the Institutional Review Board of Peking Union Medical College Hospital (approval number: SK1828). A waiver for individual consent was granted by the research ethics board because of the retrospective and deidentified nature of the data.
2.2 Exposure variables
In this study, ICU quality was assessed by quality control and capacity indicators [
6,
7]. The 15 quality control indicators was classified into three categories: structural indicators, procedural indicators, and outcome indicators. The structural indicators were the percentage of ICU patients among all inpatients, the percentage of ICU bed occupancy of the total inpatient bed occupancy, and the percentage of patients with an Acute Physiology and Chronic Health Evaluation II (APACHE II) score of ≥ 15 among all ICU patients. The procedural indicators were the 3- and 6-h Surviving Sepsis Campaign (SSC) bundle compliance rates, the rate of microbe detection before administration of antibiotics, the percentage of ICU patients receiving prophylaxis for deep vein thrombosis, the percentage of unplanned endotracheal extubation, the percentage of extubated patients reintubated within 48 h, the percentage of patients with unplanned ICU transfers, the 48-h ICU readmission rate, the incidence of ventilator-associated pneumonia, the incidence of catheter-related bloodstream infection, and the incidence of catheter-associated urinary tract infection (CAUTI). The outcome indicator was ICU mortality. The six capacity indicators assessed in this study were the ICU patient-to-bed ratio, severe patient-to-bed ratio, intensivist-to-bed ratio, nurse-to-bed ratio, intensivist-to-patient ratio, and intensivist-to-severe patient ratio. All parameters and their definitions are listed in Tables S1 and S2.
2.3 Study methods
The main outcome was in-hospital mortality, and the primary diagnosis at discharge was determined to be the reason for V-V ECMO initiation. The association between the center volume and prognosis was also assessed. On the basis of previous studies and the current ECMO situation in China, high-volume centers were defined as centers that performed ≥ 50 V-V ECMO runs in 3 years [
3,
10,
13,
14]. The cutoff value of each indicator was identified in accordance with the quality control data in China from 2015 to 2020 and clinical practice experience [
8]. Then, the patients were grouped in accordance with the quality indicators of their hospitals, and the in-hospital mortality rates of V-V ECMO-supported patients between hospitals with different qualities were compared. The association between ICU quality and in-hospital mortality was comprehensively explored.
In total, 31 provinces, municipalities, and autonomous regions in the Chinese mainland were included in the study (data from Hong Kong, Taiwan, and Macao were excluded). The age divisions and geographic and economic region divisions are listed in Table S3. Costs are expressed in USD after conversion using the annual average exchange rate between RMB and USD in 2019 (1 USD to 6.8985 RMB).
2.4 Statistical analysis
First, the baseline characteristics of all patients were stratified by time period and compared using Chi-square test and variance analysis for categorical and continuous variables, respectively. Univariate and multivariate logistic regression models with death during hospitalization as the dependent variable were applied to identify factors related to in-hospital mortality. The risk factors were screened in accordance with literature search, expert consensus, and expert experience. Factors of specific focus were the baseline characteristics (demographic characteristics, economic regions in which the hospitals were located, and season at admission) and patients’ chronic underlying diseases and ECMO-related complications. The results are expressed as the odds ratio (OR) with 95% confidence interval (CI).
Second, generalized linear mixed models were performed with the quality-related parameters included in the models as the random variable to assess the association between ICU quality and in-hospital mortality. Variables that were statistically different in the first step, including patients’ baseline characteristics, chronic underlying diseases, and ECMO-related complications, were adjusted in the models to eliminate potential confounding. The cutoff values for categorizing the quality-related parameters were assigned in accordance with the clinical implications and data distribution. The associations with in-hospital mortality were evaluated in separate models because of the multicollinearity among these parameters. A subgroup analysis of patients in high-volume centers was performed to reduce the bias caused by factors in low-volume centers. All P values presented were two-sided, with P < 0.05 considered statistically significant. All analyses were conducted using SAS statistical software version 9.4 (SAS Institute Inc., Cary, NC, USA).
In addition, SPSS Modeler software version 18.0 (IBM Corp., Armonk, NY, USA) was used to conduct apriori association rule learning analysis and plot charts. This analysis conducted using the 17 most frequently used association rules, and the minimum requirements were determined as a support degree of ≥ 10% and confidence of ≥ 47%. Furthermore, the association rules were reported in accordance with descending support and confidence and lift values corresponding to the support of the association rules.
3 Results
3.1 Baseline characteristics of V-V ECMO-supported patients and identification of confounders for quality control association analysis
In total, 2563 V-V ECMO-supported patients in 318 hospitals from 2017 to 2019 were identified in the NCIS database. The number of V-V ECMO cases increased annually, totaling 386, 719, and 1458 cases in 2017, 2018, and 2019, respectively, whereas the in-hospital mortality remained constant (32.4%, 29.4%, and 27.4%, respectively; P = 0.137; Fig.1). The baseline characteristics of these patients are listed in Table S4 and Figs. S1–S3. Univariate and multivariate regression analyses were performed to screen out factors significantly associated with in-hospital mortality. The results showed that lung cancer, intracranial hemorrhage, hypoxic–ischemic encephalopathy, multiple organ dysfunction syndrome, kidney injury, coagulation disorder, bacteremia, and shock were independently correlated with increased in-hospital mortality. By contrast, acute respiratory distress syndrome, age of 21–30 years, and location in a region with a mid-level gross domestic product were protective factors (Fig.2 and Fig.3; Tables S5 and S6). All these factors were treated as confounders in the subsequent analyses for quality control indicators.
Apriori association rule learning analysis was conducted to determine the risk factors for in-hospital mortality of V-V ECMO-supported patients by using the 17 most frequently used association rules. In accordance with the rule that a positive correlation was present between patient factors and in-hospital mortality and divided by the threshold of ≥ 200 cases, the following seven factors were confirmed to be associated with in-hospital mortality of V-V ECMO-supported patients: received continuous renal replacement therapy (n = 640), received blood transfusion for hemorrhage (n = 601), received noninvasive ventilation (n = 599), underwent cardiopulmonary resuscitation (n = 597), complicated with bacteremia (n = 286), complicated with kidney injury (n = 274), and complicated with coagulation disorder (n = 268). The combination of coagulation disorder, kidney injury, and continuous renal replacement therapy was the most lethal risk factor for in-hospital mortality of V-V ECMO-supported patients (support, 11.08%; confidence, 48.59%; Fig.4 and Table S7).
3.2 Association between ICU quality and in-hospital mortality of V-V ECMO-supported patients
All the quality control indicators and capacity parameters from hospitals where ECMO procedures were performed were collected, and mortality between hospitals with different qualities was compared. The results showed lower in-hospital mortality rates in V-V ECMO-supported patients who were treated in hospitals with a higher proportion of ICU patients among total inpatients (≥ 1.5%), lower 3- and 6-h SSC bundle compliance rates (< 60% and < 90%, respectively), a lower reintubation rate within 48 h after extubation (< 1.5%), a lower CAUTI incidence rate (< 1.5 per 1000 urinary catheter line days), and a lower total ICU mortality rate (< 7%; P = 0.045, 0.002, 0.0002, 0.007, 0.008, and 0.004, respectively). With respect to the ICU capacity parameters, the in-hospital mortality rate significantly decreased from 30.8% to 24.5% when the 3-year total number of V-V ECMO-supported patients exceeded 50 (P = 0.0009, Tab.1).
Next, univariate and multivariate analyses were performed to examine the association between ICU quality and in-hospital mortality of V-V ECMO-supported patients. After adjustment for the patients’ baseline characteristics, chronic underlying diseases, and other potential confounders that were screened out in the first step, the multivariate analysis showed that the reintubation rate within 48 h, CAUTI incidence rate, and total ICU mortality rate were independent risk factors for higher in-hospital mortality (OR [95% CI]: 1.25 [1.05–1.48], P = 0.012; 1.18 [1.01–1.39], P = 0.043; and 1.23 [1.04–1.45], P = 0.015, respectively), and treatment in a high-volume V-V ECMO center was a protective factor for the patients’ prognosis (cutoff of ≥ 50 cases within the 3-year period: OR, 0.69; 95% CI, 0.57–0.83; P = 0.0001; Tab.2).
3.3 Subgroup analysis of V-V ECMO-supported patients in high-volume centers
In accordance with the division criteria, 864 patients (33.7% of the total patients) were treated in 11 high-volume V-V ECMO centers. The in-hospital mortality rate was significantly lower in high-volume centers than in low-volume centers (24.5% vs. 30.8%, respectively; P = 0.001; Table S8). The proportion of ICU patients among the total inpatients and the proportion of ICU bed occupancy among the total inpatient bed occupancy were significantly higher in high-volume centers than in low-volume centers (3.9% vs. 1.8%, P = 0.023 and 3.2% vs. 1.6%, P = 0.006, respectively). Moreover, the proportion of patients with an APACHE II score of ≥ 15 among all ICU patients was slightly lower (35.7% vs. 52.9%, P = 0.065), the reintubation rate within 48 h after extubation was significantly lower (0.8% vs. 1.6%, P = 0.045), and the ICU patient-to-bed and nurse-to-bed ratios were significantly higher (49.7 vs. 38.6, P = 0.043 and 2.63 vs. 2.05, P = 0.037, respectively; Tab.3).
A subgroup analysis was performed on patients in high-volume centers only to eliminate the potential bias induced by center volume on the association between ICU quality and the in-hospital mortality of V-V ECMO-supported patients. Besides ECMO-related complications, the reintubation rate within 48 h and the ICU mortality rate were still independent risk factors for in-hospital mortality of V-V ECMO-supported patients (OR [95% CI]: 0.725 [0.625–0.84], P < 0.0001 and 1.219 [1.14–1.305], P < 0.0001, respectively). In high-volume centers, the ICU patient-to-bed ratio, intensivist-to-severe ICU patient ratio, and nurse-to-severe ICU patient ratio were significantly associated with the prognosis (OR [95% CI]: 0.985 [0.979–0.991], P < 0.001; 0.967 [0.944–0.991], P = 0.007; and 1.016 [1.006–1.026], P = 0.002, respectively; Tab.4).
4 Discussion
This study comprehensively summarized the performance of V-V ECMO in China from 2017 to 2019. Furthermore, ICU quality was multifacetedly evaluated using structural indicators (capacity parameters), process indicators (quality control indicators), and outcome indicators (ICU mortality), and the association between ICU quality and the prognosis of V-V ECMO-supported patients was thoroughly analyzed. In addition to ECMO-related complications, the volume of ECMO centers and certain ICU quality control indicators were independently associated with in-hospital mortality.
The correlations between the indicators of quality control and the prognosis of V-V ECMO-supported patients were thoroughly explored. The unplanned reintubation rate and the ICU mortality rate were independently associated with in-hospital mortality in V-V ECMO-supported patients. Previous studies showed that after a successful spontaneous breathing test and extubation, 10%–25% of patients required reintubation, and reintubation increased the risk of ventilator-associated pneumonia and was associated with high mortality [
15–
17]. Ionescu et al. [
18] recently found that older age, use of paralytics, and high positive end-expiratory pressure predicted reintubation. In the treatment of V-V ECMO-supported patients, ICU-acquired weakness caused by long-term application of sedatives and muscle relaxants and an inappropriate positive end-expiratory pressure setting may lead to an increase in the reintubation rate, which reflects the efficacy of respiratory therapy to a certain extent. The reintubation rate, together with the CAUTI rate and total ICU mortality rate, mirror the quality of care in the ICU, as confirmed by the study that showed patients with a high incidence rate had higher in-hospital mortality [
19–
21].
The volume of ECMO centers has been identified as an independent risk factor for patient outcomes [
3,
13], and the association between the center volume and prognosis of V-V ECMO-supported patients was first illustrated in a developing country. In the present study, V-V ECMO cases of > 50 runs within 3 years was confirmed to be a protective factor for patients’ prognosis. Patients in high-volume centers were much older and had a higher rate of respiratory comorbidities than those in low-volume centers; however, they had a lower in-hospital mortality rate. Greater experience in the delivery of ECMO is of great importance for improving outcomes; however, given the complexity of ECMO, a better prognosis could be achieved by enhanced resources and personnel, and adherence to evidence-based care (all of which are ICU quality control indicators). A subgroup analysis of patients in high-volume centers was performed to rule out the potential influence of center volume, and the association between ICU quality and in-hospital mortality was further confirmed. Besides the reintubation rate and total ICU mortality rate, the intensivist-to-severe ICU patient ratio and nurse-to-severe ICU patient ratio were significantly associated with the prognosis of V-V ECMO-supported patients in high-volume centers.
A previous study showed that a high nurse-to-patient ratio was independently associated with a lower risk of in-hospital mortality [
22]. The ELSO guidelines [
12] recommend maintenance of a 1:1 patient-to-nurse ratio when patients are on ECMO support. The nurse-to-bed ratio in low-volume centers in the present study was 2.05, consistent with data in a nationwide survey showing a calculated patient-to-nurse ratio of approximately 2:1 [
7]. However, this finding is far from adequate. Actually, the gap between medical human resources and ICU expansion was even greater, especially after the COVID-19 outbreak [
4]. The number of trained personnel per center was an independent risk factor for the prognosis of ICU patients, especially critically ill patients, supported by the present study’s finding of an association between the intensivist-to-severe ICU patient ratio and nurse-to-severe ICU patient ratio with in-hospital mortality of V-V ECMO-supported patients. Rational allocation of medical resources needs to be strengthened to further improve the prognosis of V-V ECMO-supported patients.
5 Limitations
This study had three main limitations. First, this was a retrospective observational study, and the observed differences may be subject to unobserved confounding factors. However, all available influencing factors, including indicators of ICU quality control, capacity parameters, and ECMO-related quality indicators, were analyzed to minimize the risk of bias. Second, some important information, such as the patient conditions requiring the initiation of ECMO support and the ECMO treatment details, could not be obtained because of limitations of the database. However, these problems were not the focus of this study. Finally, differences in treatment and heterogeneities among ICU management protocols could have biased the results because of the long observation window. However, the results from high-volume centers further strengthened the effect of ICU quality control on the prognosis.
6 Conclusions
The number of V-V ECMO cases increased and in-hospital mortality remained constant from 2017 to 2019. In addition to ECMO-related complications, the volume of ECMO centers and certain ICU quality control indicators (reintubation rate within 48 h after extubation and total ICU mortality) were independently associated with in-hospital mortality. The results of the subgroup analysis of 864 patients in 11 high-volume centers further strengthened these findings.
PDF
(2383KB)
