Prognostic Factors of Complications After Thoracoscopic Lobectomy for Non-Small Cell Lung Cancer-A Retrospective Study

Background: Video-assisted thoracoscopic surgery, safe and minimally invasive, is the rst strategy recommended for non-small cell lung cancer. The purpose of this study was to determine the risk factors for postoperative cardiopulmonary complications (including cardiac and pulmonary complications) in patients with NSCLC who underwent video-assisted thoracoscopic surgery (VATS). Methods: We retrospectively collected information of 3142 lung cancer patients undergoing VATS tumor resection at Jiangsu Provincial People's Hospital from January 2017 to June 2018, and established a clinical prediction model using the factors selected by univariate analysis. Results: A total of 305 in 3142 patients developed postoperative cardiopulmonary complications. In univariate analysis, age, PNI, CCI, long-term smoking history before surgery, conversion to thoracotomy, albumin before surgery, pre-albumin, Δalbumin and pleura adhesion were all associated with cardiac and pulmonary complications. Multivariate analysis showed that age, PNI, CCI, long-term smoking history before surgery, conversion to thoracotomy, pre-albumin, Δalbumin were important independent risk factors for complications. Finally, age, PNI, CCI, long-term smoking history before surgery, conversion to thoracotomy, Δalbumin variables were included in the model (AUC=0.743). Conclusion: Nutritional status (PNI, pre-albumin), CCI, age, long-term smoking history before surgery, conversion to thoracotomy were all independent risk factors for postoperative complications and poor prognosis. Experienced surgeons should instruct the operation of high-risk patients to avoid long-term hospitalization and possible poor prognosis after surgery.

Introduction has been used in the prediction of postoperative outcomes [12]. In a retrospective study conducted by West China Hospital in China, perioperative serum ALB level was used as an independent predictor to predict the occurrence of PCCs. However, Loftus held the opinion that serum ALB could only re ect nutritional status when the visceral protein is stable [12]. Therefore, some studies have proposed to use PNI to predict postoperative lung cancer risk [13], and found that the incidence of pneumothorax, pneumonia and extrapulmonary infection in the PNI group was signi cantly higher than that in the normal PNI group.
In addition to nutritional status, age and CCI may also be predictors of PCCs. Some studies found that elderly patients were more likely to have PCCs [14,15], but a prospective American research suggested that age could not predict the occurrence of major postoperative complications [16], and one patient's physique-related indexes could better predict the occurrence of PCCs. The Charlson Comorbidity Index (CCI) is also a good indicator for evaluating postoperative complications [15,17], and CCI has many overlaps with the criteria for assessing frailty, so both can re ect frailty. In addition, it is still controversial whether intraoperative conversion to thoracotomy and LSHS before surgery can be used as risk factors for PCCs. Two retrospective studies suggested that the different survivals of patients who received successful VATS and converted to thoracotomy. There was no signi cant difference in recurrence [18,19], but the incidences of atrial brillation, pneumothorax, and acute renal failure were higher in patients who converted to thoracotomy in Bongiolatti S' study [20].
Studies showed that patients with LSHS in forefoot surgery were more likely to develop PCCs [21,22].
However, in the studies of spinal surgery, head and neck surgery, there was no correlation between LSHS and complications [23,24]. There are also different voices on whether the clinical stage of lung cancer can predict the incidence of PCCs [25,26] In this study, we retrospectively collected the incidences of PCCs in 3142 patients who underwent VATS for NSCLC in Jiangsu Provincial People's Hospital, and evaluated various factors related to PCCs, including age, gender, PNI, CCI, LSHS and ΔALB. We established a prediction model based on the data of all patients in order to evaluate the incidence of PCCs before surgery.

Study patients
Totally, 3149 cases recruited from Jiangsu People Hospital were divided into groups A and B. After data processing, we deleted samples with > 25% dada missed and meaningless data. The nal sample size was 3142, consisting of 305 patients who had PCCs and 2837 patients who had never developed any complications. Inclusion criteria were: (1) target diseases were operable primary NSCLCs; (2) only standardized single-lobectomy with systematic mediastinal lymph node dissection (SMLND) was operated by a VATS procedure; (3) patients must nish the entire clinical pathway according to our institutional policies during the hospitalization; (4) patients with loss of accurate records on estimated variables were not considered. Baseline data collected for this study included: age, gender, BMI, LSHS, CCI (Charlson Comorbidity Index),  operation time, FEV1, TNM staging, conversion to thoracotomy, WBC, ALB before operation, albumin change(ΔALB), LDL, postoperative stage, tumor location and pleura adhesion. Among these variables, the pleura adhesion was classi ed into four types (no, light, medium, and high); tumor location into ve types (the top right, top left, bottom right, bottom left and middle right group); smoking history into two types (non-smoking and more than 10 years). CCI is the Chalson comorbidity index [17]. Constantly, blood samples were measured at 1-7 days before surgery and one day after surgery.

Statistical analysis
After data processing, we described continuous normally distributed data with mean±standard deviation (SD) and non-normal data with medians and corresponding interquartile range (IQR) (25% and 75% percentiles). Categorical variables were expressed as absolute frequencies and proportions (%). Clinicopathologic characteristics were compared between the two groups. For categorical variable, chisquare method was used to analyze the difference between two groups. For continuous variable, student t-test and variance analysis were used for variables with normality and variance homogeneity, while Wilcoxon rank sum test for variables failed to satisfying the requirements above. Variables with signi cant meaning (P<0.05) were selected for extra analysis, and the relative results of difference analysis were visualized. Then we established multivariable logistics regression model using the variables with signi cant difference de ned in univariate analysis. Results of regression analysis were expressed as odds ratio (ORs) and 95% con dence interval (95%). The receiver operating characteristics (ROC) curve was drawn for evaluating the model, and nomogram analysis was performed to describe the variable scores for risk prediction. All statistics analysis was performed in R software 4.0.2 version.

Results
A total of 3142 patients undergoing VATS were included in this retrospective study. Table 1 summarizes the baseline characteristics. The average age was 56 (48,64) years, 57.7% (1813/4142) of the patients were women, and the average preoperative ALB level was 43.6 (40.5, 47.8) g/L. Hypoalbuminemia was de ned in 235 patients with preoperative albumin lower than 35g/L. The average albumin decreased to 37.7 (34.8,40.3) g/L on the rst day after surgery. The numbers of patients with CCI grades 0, 1, >=2 were 1825 (58%), 898 (28.6%), 421 (13.4%), respectively. During the operation, 82 patients converted to thoracotomy for various reasons. The average preoperative WBC level was 8.51 g/L [6.2, 11.9], and the average LDL level was 2.62 [2.1, 3.1] g/L. Preoperative TNM stages 1, 2 and 3 were observed in 2962, 124 and 63 patients, respectively. 54.7% (, p < 0.001) of patients had different degrees of pleural adhesions. [Comparison between PCCs group and non-PCCs group]: Table 1 shows the perioperative demographic difference between patients with or without PCCs. Figure Table1 respectively show the distribution and differences of age, PNI, SIRI, ΔALB, CCI, LSHS between the two groups with or without PCCs. Age (p < 0.001), WBC (p < 0.001) and ALB (P = 0.024) were higher in the PCCS group, while PNI (P < 0.001) was lower in the PCCs group, and the median differences in these variables between the two groups were statistically signi cant. Meanwhile, CCI (p < 0.001), conversion to thoracotomy(p < 0.001) showed difference between the two groups.
The proportions of patients in TNM stages 2 and 3 (p = 0.027) and other pathological parameters showed no signi cant difference in patients without PCC.
[Univariate and multivariate analyses]: As shown in Table 2 (Fig. 2). The AUC of this model was 0.743, suggesting good tting. Each variable in nomogram was scored. The cumulative sum of these scores was the "total score" that predicted the probability of complication (Fig. 3).

Discussion
The logistic regression model is an e cient tool to evaluate the incidence of PCC in patients with NSCLC under VATS. In this study, we observed that the incidence of PCCS after VATS was generally higher than 10%, and the incidence of PCCs was low. The model has a strong ability to predict the prognosis of NSCLC, with a positive predictive value of 0.743. In external veri cation, the sensitivity and speci city were still relatively high, indicating that our risk prediction model had a good performance in predicting PCCs in patients after NSCLC.
The preoperative changes in ALB, ALB and PNI affect the incidence of PCC after endoscopic surgery in lung cancer patients. Many studies have found that preoperative serum albumin can be used as an important indicator for predicting postoperative results and managing comorbidities [8, 11, 12, 28-31].
Malnutrition is a risk factor that can be optimized before surgery, and an accurate predictor of adverse reactions after surgery. A prospective cohort study exploring postoperative complications of major abdominal surgery found that ΔALB ≥ 10g / L produced a sensitivity of 77.1% and a speci city of 67.2% (AUC: 78.3%) to predict complications. Patients with ΔAlb ≥ 10 g/L on POD 1 had a three-fold increase in the overall risk of postoperative complications [11]. Another prediction model for short-term postoperative complications of gastric cancer showed that with a cut-off ΔALB of 19.0%, patients with ΔALB ≥ 19.0% had a higher risk of postoperative complications (62.3 vs. 13.7%, P < 0.001), and longer postoperative hospital stay (22.1 13.5 vs. 17.5 4.2, P < 0.001) than those with ΔALB < 19.0% [8]. Another prediction model for postoperative complications of lung cancer also reached the same conclusion [31]. Not only abnormal serum albumin before and after surgery, preoperative hypoalbuminemia is also a strong risk factor for PCCs. When the serum ALB was lower than 3.0g/L, the incidences of almost all complex complications showed an upward trend [29]. However, some researchers have proposed that ALB cannot be used as an independent biomarker of malnutrition. These visceral proteins can only be used as a marker of nutritional status among patients in a stable state. Therefore, studies have begun to use the prognostic nutritional index (PNI) as an immunonutrient index to predict PCCs of various types of malignant tumors. Other studies found that low preoperative PNI was an important independent predictor of Clavie-Dindo CII PCCs [13]. In studies related to lung transplantation, multiple regression analysis showed that even after adjusting for other confounding factors, the preoperative PNI score was signi cantly correlated with postoperative survival, which indicated that PNI can be used as a parameter for evaluating the prognosis of lung transplantation [32]. Age and CCI are sensitive indicators of the incidence of PCCs after endoscopic surgery in lung cancer patients [3].
This study found that advanced age was also associated with postoperative complications, which is consistent with many other studies. A study of Kim revealed that > 60 years old was a high-risk factor for postoperative infection, bringing with a risk 1.45 times of that in the low-age group. In prospective studies of lung transplantation, age was also an independent prognostic factor of PCCs [32]. However, more studies suggested that frailty could better predict postoperative outcomes than age. Among the 19 articles reviewed by Beggs, eight showed that frailty was signi cantly associated with mortality and PCCs [33]. As more and more patients over 75 years of age receive VATS, the vulnerability may have considerable value in evaluating this tool for perioperative evaluation. A short-term prognostic model for the outcome of hepatobiliary, pancreatic, spleen and bile duct surgery showed that the risk index of physical weakness (RAI) was a quick predictor of comorbidities and performance status [34]. Another prospective study clearly concluded that the frailty measurement was better than the strati cation by age, and the Rockwood vulnerability index could indicate the outcome of elderly patients receiving EGS. On the contrary, age and American anesthesiologist score (ASA) cannot predict postoperative and major complications (ASA in Table 2) [16].
CCI is also an indicator for evaluating preoperative frailty, and predict complications after surgery. In a study of Whitmore, although neither CCI nor ASA scores could completely predict the occurrence of PCCs, CCI was bene cial to consult and evaluate patients with spinal surgery before surgery [35] Studies have also compared the diagnostic abilities of (Chalson Comorbidity Score, Elixhauser Comorbidity Score, Comorbidity Score, Centers for Medicare and Medicaid Services-Graded Disease Category [CMS-HCC]) with that of prescription-based chronic disease scores (CDS), nding that almost all scores were better than prescription-based scores, and combining the two scores does not improve the predictive value [36]. More CCI-related studies have suggested using age-adjusted CCI for predicting postoperative tumor complications [37][38][39]. Tomonari A also found that higher CACI was an independent predictor of poor survival in a retrospective study of pancreatic cancer. In this study, age and CCI showed a correlation with complications in both univariate and multivariate analyses. In order to avoid interference between the two factors during modeling, CCI replaced CACI.
Intraoperative conversion to thoracotomy and LSHS were related to the incidence of PCC in lung cancer patients after endoscopic surgery. A conclusion is drawn from the analysis of the causes of accidental thoracotomy during thoracoscopic lobectomy for lung cancer. Respiratory complications were more common in the conversion group (p = 0.012), but accidental thoracotomy did not increase the overall surgical morbidity and mortality [40]. A multi-center retrospective analysis in Italy showed that unplanned thoracotomy was signi cantly associated with postoperative morbidity [15]. Among Bertolaccini L's research, intraoperative vascular injury, pleural adhesions, age > 60 years and resection locations were all factors that affect the incidence of PCCs and mortality. Experience was an independent factor to predict conversion [15]. However, studies of the same type put forward different points of view. A single-center study concluded that there was no signi cant difference in survival and recurrence between groups with successful VATS and conversion to thoracotomy, but the operation time and hospital stay were longer in the group with conversion to thoracotomy [14]. Another retrospective study also showed that unplanned thoracotomy did not affect the prognosis. [41] In this study, we found that patients with a long history of smoking before surgery had a signi cantly higher incidence of PCCs than patients who had never smoked. A retrospective study in Japan found that smokers had a higher PCCs incidence than non-smokers. The incidence of respiratory-related complications gradually increased with the number of smoking years [42]. Another retrospective study of 666 patients in Jampan found that long-term smoking history before surgery and TNM stage were predictors of complications after pneumonectomy [25]. However, Matsuoka K believed that the length of time to quit smoking before surgery had no relationship with the frequency of PCCs, even if smoking restarted within two weeks before surgery. Related studies also found that LSHS before surgery was more correlated with postoperative complications of foot and ankle surgery, adult spinal deformity surgery, and head and neck microvascular reconstruction surgery than those who had never smoked [23,24,43].
There are still some limitations in this study. First, this study is limited by the inherent weak points of single center retrospective analysis, and the potential selection bias may weaken the predictive ability of our prediction model, because PCCs arises from a series of complex factors. In addition, the sample size of a single center may limit the reliability of this analysis. Second, although the predictive power of frailty index is higher than that of age, it is impossible to collect relevant indexes to calculate frailty index prospectively because this study is a retrospective analysis. Third, the accuracy of incidence calculation is likely to depend on the clinical experience of surgeons. However, quantitative analysis of human experience may be di cult.  ). The grant funders had no role in the study design, data collection, analysis and interpretation, or preparation of the manuscript.

Availability of data and materials
The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Ethics approval and consent to participate
The work was approved by the ethical committee of Nanjing Medical University ("F","CH","Nanjing Med U", "FWA00001501", "NANJING", 11/21/2004), and an IRB (Institutional Review Board) approval was given prior to this study. I have read and have abided by the statement of ethical standards for manuscripts submitted to BMC Pulmonary medicine. All methods were carried out in accordance with relevant guidelines and regulations. The need for informed consent was waived by the ethical committee of Nanjing Medical University ("F","CH","Nanjing Med U", "FWA00001501", "NANJING", 11/21/2004) due to its retrospective nature.

Consent for publication
Not applicable.

Figure 2
Receiver operating characteristic (ROC) curve to determine the optimal cut-off of multivariate analysis results, showed an AUC of 0.743.