Background
Congenital heart disease (CHD) complicated with airway anomaly, a rare combination of lesions, approximately accounts for 3–4% of patients [
1]. Tracheobronchial malacia, tracheal stenosis, and tracheoesophageal fistulas are the major presentations of the anomalies, and surgical intervention may be needed occasionally [
2,
3]. Airway stenosis (AS) is the major subtype and 50–70% of the patients are reported to have a concomitant cardiac anomaly [
4‐
6].
Depending on the various presenting symptoms, different treatment strategies may be adopted. For severe conditions, surgical intervention may be needed. However, for AS patients with moderate respiratory symptoms, conservative management may be preferred. Cheng et al. indicated that compared to normal children, tracheal growth and tracheal diameter enlargement seem to be faster in congenital tracheal stenosis (CTS) children, especially after infancy [
7]. In a large retrospective analysis of the Society of Thoracic Surgeons Congenital Heart Surgery Database (STS-CHSD), 6861 (3.4%) patients were complicated with airway anomalies and only 428 (0.2%) patients underwent tracheal operations during the same hospitalization [
8].
Most of the studies focused on CHD patients who underwent tracheal surgery, so the data on the conservative group are scarce [
9,
10]. As we mentioned, most CHD patients with tracheal anomalies may not require tracheal intervention, thus, studies exploring these populations are warranted to better understand the outcomes and prognosis. Consequently, in this study, we aim to retrospectively investigate the CHD patients who underwent conservative management of AS in our center, to provide novel insight to AS management.
Methods
Patient population and management
Patients with the diagnoses of CHD complicated with AS in Fuwai Hospital were reviewed from July 2009 to December 2022. For patients diagnosed with AS and CHD, computered tomography (CT) scan or fiberoptic bronchoscopy would be prescribed. After adequate multidisciplinary team (MDT) consultation and preoperative evaluation by cardiac surgeons, cardiologists, echocardiologists, and radiologists, patients with stenotic airway segment detected by fiberoptic bronchoscopy or CT scan and significant respiratory symptoms would undergo tracheal surgery, such as slide tracheoplasty. However, in patients with AS reported by bronchoscopy or CT but mild or asymptomatic respiratory symptoms, conservative treatment on AS would be preferred. Consequently, we enrolled patients who met the following criteria: diagnosis of CHD, AS determined by fiberoptic bronchoscopy or CT scan, no significant respiratory symptom, and no tracheal surgery during hospitalization.
Definitions
The Society of Thoracic Surgeons- European Association for Cardiothoracic Surgery (STAT) mortality category was selected to risk-stratify the CHD procedures. Complex CHD was defined as CHDs excluding simple shunts (atrial septal defect, ventricular septal defect and patent ductus arteriosus), pulmonary artery slings and vascular rings [
11]. The primary outcome was defined as prolonged mechanical ventilation (PMV). According to previous reports and our clinical experience, PMV was defined as postoperative mechanical ventilation for more than 14 days [
12‐
14].
Statistical analysis
Descriptive analysis of baseline data was presented by median (interquartile range, IQR) for continuous variables, or frequency (percentage) for categorical variables. Wilcoxon rank sum test were used to compare the continuous data and the chi-square test or Fisher exact test was used to analyze categorical data. Variables with missing values were detected and processed by the Mice package of R. The imputation methods were selected based on the type of variables that continuous data were imputed using Predictive Mean Matching, and categorical data were imputed using Logistic Regression Imputation. The imputed dataset was compared with the original dataset to verify the bias generated in the process. The imputed dataset was used for the following statistical analysis.
Logistic regression was used to assess the association between perioperative indicators and prognosis. Variables with potential predictive power were first included in the univariate logistic regression analysis, and those with P < 0.1 were further included in the multivariate logistic regression analysis.
We tested the accuracy of the model with the Hosmer-Lemeshow test and presented it with calibration plots. The receiver operating characteristic (ROC) curve was plotted to determine the model’s ability to accurately identify PMV in children with CHD undergoing conservative treatment of AS. We also identified the net benefit of the model by decision curve analysis (DCA). Finally, a nomogram was adopted to visualize the model.
Due to the lack of an external validation cohort, we internally validated the model using machine learning method. Cross-validation was selected to examine the accuracy and analyze the overfitting of the predictive model. After randomly divided all the data in the original cohort into k groups, (k-1) of them were taken as the training cohort for constructing the model, and the remaining 1 group was treated as the validation cohort (k-fold cross-validation). The model constructed from the training cohort was used to predict the outcome of the validation cohort. To reduce the influence of random grouping, the above process was repeated 200 times. The other method we used for internal validation was bootstrap validation. The training cohort was constructed by sampling (with replacement) from the original cohort. Although the probability of each patient being selected was the same, the training cohort differed significantly from the original cohort due to the use of sampling with replacement. Similarly, the process was repeated 200 times. We used these two methods to obtain the corrected area under curve (AUC) values and 95% confidence intervals for model correction, respectively.
Based on the results of logistic regression, the possible risk score for each patient was calculated, and the patients were divided into high-risk and low-risk groups according to the optimal cutoff value of the ROC curve. We further compared the differences in other postoperative indicators between the two groups, including postoperative hospital stay, ICU stay, reintubation, ECMO implication, and in-hospital death.
All the statistical analyses were performed using R-studio (version 4.2.2), P < 0.05 was considered to be statistical significance.
Discussion
This is the first study focusing on CHD patients with AS who underwent conservative airway treatment. Plenty of studies have investigated the surgical outcomes of CHD patients who underwent airway surgery, while the data on AS without intervention are scarce. Riggs and colleagues retrospectively reviewed the associated airway anomaly in pediatric patients who underwent heart surgery from the STS-CHSD, and this study provided insights in patients with AS but without surgical intervention that the operative mortality of 5.9%, major morbidity of 21.2%, and postoperative tracheotomy of 5.6%, respectively [
8]. As a result, we sought to take a deep insight into this population and provided more information on their perioperative characteristics, by retrospectively reviewing the patients who met the criteria from 2009 to 2022 in the National Center for Cardiovascular Center of China.
The mechanical ventilation duration is an important perioperative indicator for children undergoing cardiac surgery, especially children with airway anomalies, yet there was no consensus definition of PMV in children. According to a systemic review, this definition varied from 48 h to 6 months [
15]. Polito et al. retrospectively investigated the mechanical ventilation duration of patients after complex congenital cardiac surgery, and the cohort comprised 362 patients, of whom 41 (11%) required mechanical ventilation for ≥ 7 days (median ventilation duration for 362 patients: 1.5 days, range: 0–7 days) [
13]. However, the mechanical ventilation duration of CHD patients was longer when complicated with airway anomaly. McMahon et al. reported that pediatric patients with CHD and bronchopulmonary dysplasia had a median postoperative mechanical ventilation duration of 15 days (range: 1-141) [
16]. For patients with CHD and CTS, the median ventilation duration was 9 days (IQR, 5-20.75) [
11]. Given that the coexisting airway stenosis may extend the mechanical ventilation duration compared to patients with CHD only, we established the standard of PMV at a higher level (≥ 14 days). We also referenced results from PICU, where patients with various conditions, not limited to cardiovascular disease. A cohort enrolled children in PICU showed the incidence of PMV (≥ 14 days) is 33.2%, identifying an elevated rate of extubation failure, increased hospitalization costs, and higher mortality after 1-month discharge in patients who received PMV [
17].
The PMV is associated with ICU stay duration, re-intubation, perioperative complications, and increased mortality [
18,
19]. There were several research focusing on the predictive factors for mechanical ventilation after pediatric cardiac surgery. Gaies et al. developed a model to predict the mechanical ventilation duration after pediatric cardiac surgery, revealing that age, prematurity, extracardiac/genetic anomalies, underweight, preoperative mechanical ventilation, higher STAT category (STAT 4 and 5), and cardiopulmonary bypass duration were the independent predictors [
20]. According to a multicenter study performed by Gupta et al., the odds of mechanical ventilation after cardiac surgery were associated with patient characteristics, surgical risk category, and cardiac center volume [
21]. Similarly, Maisat et al. identified risk factors for postoperative mechanical ventilation for pediatric pulmonary vein stenosis, including male sex, low body weight, preoperative oxygen supplement, high PVS severity score, intraoperative red blood cell transfusion low preintervention PaO2/ FIO2 ratio, and high preintervention right ventricular systolic pressure [
22]. According to our study, four predictive factors were identified when focusing on children with CHD and AS: weight at CHD surgery, CPB duration, complex CHD, and comorbid tracheobronchomalacia. CHD, especially great vessel anomalies can cause airway compression, contributing to tracheobronchomalacia [
23]. According to Chen et al., the combination of CHD and tracheobronchomalacia was associated with PMV, ICU-stay, hospital-stay, and mortality [
24]. For a subset of patients with airway compression or tracheobronchomalacia, invasive airway intervention may be waived after the relief of compression or stenosis by cardiac surgery.
Patients included in our cohort had a relatively young age (median 0.70, IQR 0.39–1.54), which could be attributed to the growth of children’s airways. As their bodies grow, the length, diameter, and cross-sectional area of the airway increase, and this growth process continues into adulthood [
25]. In comparison to the trachea in adults, the size of the trachea during infancy is approximately 50%, 36%, and 15% of the length, diameter, and cross-sectional area, respectively [
26]. The growth of the trachea has been observed in patients with CTS as well, with the diameter approaching normal values by the age of nine [
7].
Our study developed a nomogram-based prediction model and the corresponding risk scores could be calculated. The predictive model identified patients who had a high risk for PMV, and those patients may tend to have worse postoperative outcomes, including postoperative ICU-stay and hospital-stay, reintubation, ECMO use, postoperative tracheotomy, and in-hospital death (Table
2). Therefore, early identification of this group is essential that more attention should be paid on intensive care of these patients in the perioperative period, and more comprehensive as well as routine surveillance of these patients should be set up after discharge to further improve the prognosis.
Efforts were made to shorten the mechanical ventilation duration. According to a randomized clinical trial by Blackwood et al., compared to usual care in the pediatric ICU, a sedation and ventilator liberation protocol including assessment of sedation levels, spontaneous breathing trials, and non-invasive ventilator resulted in a statistically significant reduction in time to the first successful extubation [
27]. Tracheotomy during ventilator adoption is also considered to be beneficial in reducing mechanical ventilation. Each pediatric intensive care unit has different options for the timing of tracheostomy [
28]. A recent review suggested that performing a tracheotomy early may improve important medical outcomes [
29]. Other management included specific body positions for receiving mechanical ventilation and intraoperative protective ventilation [
30,
31]. Strategies for reducing the duration of mechanical ventilation in patients with congenital heart disease combined with AS are still under discussion. However, one thing is certain, early identification of those at risk for PMV is necessary.
Limitations
This study has several limitations to be addressed. First, as the present study was a single-center retrospective study, validation of an external cohort was lacking. The model’s predictive power should be tested in an external cohort by calculating each patient’s risk score and comparing it to their actual outcomes. Although we performed adequate internal validation to assess whether the model was overfitted, we were still unable to confirm whether the model was applicable to all patients. A subsequent prospective study may serve as an external validation cohort to evaluate the external applicability of this model. Second, patients included in the study suffered not only from AS but also various cardiac malformations, the influence of cardiac surgery could not be ignored, even if we tried to reduce this imbalance, the final results might still be influenced. Finally, our study included all the patients who met the inclusion criteria between July 2009 and January 2022, it is possible that the sample size was inadequate, albeit to the rarity of CHD complicated with AS [
32]. The optimal sample size for the predictive model was calculated according to the 4-step sample size calculation method proposed by Riley et al. [
32]. Since no similar prediction model was available for our reference, the parameters we set when using the 4-step method were based on the model we built, the calculation results could only be used for model evaluation and subsequent model refinement. With an outcome incidence of 0.265, an AUC value of 0.847, and 4 variables planned to be included as candidates, a sample size of 300 was calculated using the
pmsampsize package of R, and 80 of these outcome events should be observed.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.