Study design
We conducted a post hoc reanalysis of a large single-center retrospective cohort. The data collectors were blind to the primary outcome. The analysis included patients who successfully underwent VATS with non-intubated anesthesia from January 1st, 2011, to December 30th, 2018 at the Guangzhou Institute of Respiratory Diseases. The inclusion for non-intubated VATS is the same as described before12-14, which patients were age ≥ 18 years, with body mass index (BMI) ≤ 25, the American Society of Anesthesiologists (ASA) physical status III or less, with no abnormal airway and spinal anatomy, no compromised coagulation, no serious cardiopulmonary dysfunction, and no extensive pleural adhesion. Patients were excluded from this observation if they proceeded to overlapping operations besides lungs, thoracotomy, tracheal surgery, and esophagus surgery. Patients who had invalid or unavailable preoperative basic examination results and incomplete intraoperative and postoperative medical records were also excluded (Fig. 1).
Main operating procedures
All patients received VATS under non-intubated anesthesia.12,15 Two anesthesia methods were initiated in our institution, one was plasma concentration target-controlled infusion (TCI) of propofol and remifentanil, combined with intravenous dexmedetomidine and placed with laryngeal mask airway (LMA), the other method was epidural anesthesia (EA), which maintained anesthesia with ropivacaine with placing the epidural catheter in the fifth or seventh epidural space. It was up to the patients, the experience of surgeons and anesthesiologists to decided which anesthesia method to choose.
The thoracoscopic procedures were consistent with the guidelines of the American Association for Thoracic Surgery.16 Whether to place a chest-drainage tube or not depends on the patient's condition and surgical procedures. All thoracic procedures were divided into four types, such as the nonanatomic wedge resection, including wedge resection, bullectomy, and lung volume reduction surgery; the anatomic resection, including lobectomy and segmentectomy; the mediastinal mass resection; and others procedures, including bilateral sympathectomy, thoracoscopic exploration, lung biopsy, pericardial cyst resection, etc. Surgical time was defined as the interval from skin cutting to wound suturing and a surgical dressing covering. After the operation, the patients were removed LMA or epidural catheter and sent back to the ward, or transfer to the intensive care unit (ICU). If the patient did not place a drainage tube or enter the ICU, the duration of chest drainage and ICU stay were recorded as zero.
Primary endpoint and candidate predictors
The primary endpoint was postoperative complications (PC). PC definition included: pleural effusion, mechanical ventilation, dyspnea, arrhythmia, air leakage, fever, reoperation, cardio-dysfunction, chylothorax, pulmonary embolism, and death. The candidate predictors included preoperative characteristics and intraoperative variables, such as age, gender, BMI, the level of ASA physical status, the previous medical history, revised cardiac risk index (RCRI), stair climbing, values of the forced vital capacity in percent of predicted (FVC% predicted), and the forced expiratory volume in the first second in percent of predicted (FEV1% predicted), left ventricular ejection fraction (LVEF), thoracic procedures, surgical location, preoperative values of leukocyte and neutrophil ratio, anesthesia methods, surgery time, blood loss, intraoperative values of minimum pulse oxygen saturation (SpO2) and arterial partial pressure of carbon dioxide (PaCO2).
Statistical analysis
Statistical analyses of patient distribution were performed with SAS software version 9.2. All analyses were based on the input of complete cases.
All eligible patients were allocated on the ratio of 3:2, and a development data set and a validation data set were established respectively. A risk model and a risk score model were established according to the development data set. In univariate analysis, P < 0.05 candidate predictors were included in the development of the risk model. The best subset of risk factors was selected by the bootstrap method to avoid over-fitting. The scoring method of the risk score model was similar to Sullivan’s17 and was based on the development of the risk model. Continuous variables were classified as clinical significant categories for scoring purposes. Pearson’s contingency coefficient evaluated the degree of correlation between the score levels and the PC risk, and the Cochran-Armitage test was used to examine the trend.
The predictive accuracy of the risk model and the risk score model was assessed by both discriminations measured by the C-statistic and calibration evaluated by the Hosmer- Lemeshow c2 statistic and calibration plot. Furthermore, the risk scoring model was validated by split-sample to evaluate the stability of the model. The area under the receiver operating characteristic (ROC) curve was compared by the nonparametric approach of DeLong.18