From January 2018 to March 2021, a total of 105 adult patients underwent ECMO at the First Affiliated Hospital of Anhui Medical University, People’s Republic of China. The indications for ECMO followed the Extracorporeal Life Support Organization (ELSO) guidelines (10). VA-ECMO is the treatment of choice for various patients with acute biventricular failure, including acute myocardial infarction, fulminant myocarditis, sudden cardiac intervention, awaiting cardiac transplantation, acute right heart failure: acute massive pulmonary embolism and intractable ventricular arrhythmias. VV-ECMO is the treatment of choice for patients with acute respiratory failure due to various causes. The main indications include severe acute respiratory failure due to ARDS patients, lung transplant patients, bronchial asthma, pulmonary embolism, atmospheric airway obstruction, and other causes. This study’s exclusion criteria were as follows: (1) age < 18 years; (2) ECMO duration<2 days or > 28 days; and (3) bridge to heart or lung transplant; (4) periprocedural support for large airway stenosis.

According to the exclusion criteria, 63 patients were recruited for this study (Figure 1). Baseline characteristics, laboratory parameters, ECMO parameters, and in-hospital outcomes were reviewed from medical records. This was a retrospective cohort study in accordance with the Ethical Guideline of the Committee on Human Experimentation of our institution. Due to the nature of this retrospective study, informed consent was waived.

The primary outcome was 30-day survival. The secondary outcomes included survival to ECMO weaning, ECMO duration, kidney injuries accepted continuous renal replacement therapy (CRRT), and ECMO-associated complications (leg ischemia, hospital acquired infections, thrombosis or hemorrhage).

Continuous variables are expressed as the mean ± SD or median (IQR), as appropriate. Categorical data are expressed as numbers (percentages). Continuous variables were compared using Student’s t test or the rank-sum test, as appropriate. Categorical variables were compared by the χ² test. Univariate Cox regression was used to screen for variables that were significantly associated with 30-day survival in the primary cohort. The proportional hazards assumption was checked based on the scaled Schoenfeld residuals using the survival package in the R tool. Potential prognostic factors that were significant in the univariate Cox regression model were entered into the multivariate Cox proportional hazard model, in which the HR, which was used to approximate the risk of an event, was also calculated. To avoid too many variables entering into the final model and influencing the practicality of the model, a strict cut-off value of 0.05 was chosen. The backwards stepwise process based on the Akaike information criterion was used to control the overfitting of the model.

A two-tailed p value <0.05 was considered statistically significant in our study. SPSS software (V.26.0, IBM, New York, USA) and R software (V.3.6.3, R Foundation for Statistical Computing, Vienna, Austria) were used for statistical analysis.

A nomogram based on the results of previous multivariable analyzes was established. The calibration, discrimination and clinical usefulness of the nomogram were calculated to evaluate its performance. The area under the receiver operating characteristic curve (AUC) and Harrell’s concordance index (C-index) were used to assess the predictive capacity of the prediction model (11). The calibration curve was used to analyze the agreement between the nomogram and actual observation. Decision curve analysis was performed to assess the clinical usefulness of the prognostic nomogram by quantifying the standardized net benefits at different threshold probabilities. We plotted Kaplan–Meier curves over the tertile of patients stratified by the scores predicted by the nomograms in the data set to further assess calibration. The model was validated using bootstrapped resampling to quantify any overfitting. Survival curves were used to compare the survival probability between the survival group and the non-survival group defined by the nomogram.

This study used multiple interpolations to fill in the missing data (Supplementary Table S1). A total of 63 patients who underwent ECMO support were included in the study and divided into the survival group and non-survival group. There were 44 (69.84%) males and 19 (30.16%) females who received 28 venoarterial (VA) ECMO and 35 venovenous (VV) ECMO. The demographic and clinical characteristics were shown in Table 1. The age, height, weight, patient source, and comorbidities were not statistically significant between the two groups, and also no statistically significant between the VV-ECMO and VA-ECMO groups (P>0.05) (Supplementary Table S2). Clinical and laboratory data are shown in Table 2. Acute Physiology and Chronic Health Evaluation II (APACHE II) and sequential organ failure assessment (SOFA) score before ECMO initiation, mean daily dose of noradrenaline (NE), mean arterial pressure (MAP) and disseminated intravascular coagulation (DIC) score were statistically significant between the two groups (p < 0.05).

The secondary outcomes, the ECMO duration, length of total hospitalization, length of ICU stay, ECMO weaning, and CRRT were statistically significant between the two groups (p < 0.05) (Table 2).

In univariate analysis, the APACHE II score before ECMO initiation, average daily dose of norepinephrine, and DIC score were statistically significant. In a nonlinear trend (Figure 2), within 30 of the APACHE II score before ECMO initiation, with the score increased, the mortality risk gradually increased; when the APACHE II score was greater than 30, the mortality risk became stable with the score increased (Figure 2A). When the average daily dose of NE was within 1.5 μg/kg/min, the mortality risk gradually with the dose increased; then the average daily dose of NE was greater than 1.5 μg/kg/min, the mortality risk tended to increase with increasing dose (Figure 2B). When the DIC score is within 4, as the score increases, the mortality risk gradually increases; when the DIC score is greater than 4, the mortality risk tends to be stabilized with the score increased (Figure 2C).

Because of the APACHE II score, so the age was not included separately in the multivariate analysis. Multiple epidemiologic studies have observed sex-related differences in admission into the ICU and ICU resources (12, 13). The statistical indicators and sex in the univariate analysis were incorporated into the Cox analysis. The results showed that the Sex, APACHE II and DIC score before ECMO initiation, average daily dose of NE were independent risk factors for prognosis (Table 3).

Multivariate Cox regression analysis was used to screen out the risk factors of prognosis, and was selected to establish a nomogram chart (Figure 3). The specific value of each factor corresponded to the corresponding score, and the total score corresponded to 30-day mortality. The nomogram model was internally validated using the bootstrap repeated sampling method 1,000 times. The consistency index C-index was used to test the model discrimination, the nomogram C-index: 0.886, AIC: 167.584. A value close to 1 indicates that the predictive performance of the model is better. Model calibration: The consistency between the predicted probability and the actual probability is evaluated by drawing a calibration curve. The higher the overlap between the fitting curve and the standard curve, the better the fitting degree (Figure 4). Using decision curve analysis (DCA) of the survival data to evaluate the clinical effectiveness, the modeling curves were all above the two extreme curves, indicating good clinical outcomes (Figure 5).

The nomogram demonstrated good specificity for predicting 30-day survival of ECMO patients. The AUC value of the nomogram was 0.999 (95% CI: 0.96, 1.00), when compared with the NE, Sex, APACHE II and DIC scores, displayed an area under the receiver operating characteristic (AUROC) that was higher in both sets (Figure 6).

To further assess the discriminative ability of the model and the predicted probability of the 30-day survival, Kaplan–Meier curves were stratified by the tertile of the predicted probability calculated from the nomograms (Figure 7). Patients with the lowest predicted 30-day survival (tertile 3) had a worst outcome (0.00%) than patients in tertile 1 and 2 (100.00 and 57.14%) (p < 0.001). Compared with actual survival based on Kaplan–Meier tables, the 10-day survival predicted by the nomogram revealed good estimations of 100.00, 76.19 and 33.33% in tertile 1, 2, 3, respectively (p < 0.001). Compared with actual survival based on Kaplan–Meier tables, the 20-day survival rate predicted by the nomogram revealed good estimations of 100.00, 61.90 and 9.52% in tertile 1, 2, 3, respectively (p < 0.001) (Figure 7).