A total of 79 patients were enrolled in this study after exclusion, of which 20 patients developed chyle leak (CL). Among them, 10 patients were grade A, 10 patients were grade B, and no patients had grade C CL. The median and interquartile range of abdominal drain fluid TG was 2.86 [1.86; 4.99] mmol/L in CL group. Table 1 showed the perioperative variables in the CL and non-CL cohorts, with significant differences observed in operation time, and venous resection (p < 0.05). The number of lymph node harvested was also higher in CL patients (p = 0.069). Table 2 demonstrated the surgical outcome for patients undergoing TP. The hospital stay was prolonged in the CL group compared to the non-CL group, although no statistical difference was observed (p = 0.08). Also, there was no statistically significant difference in postoperative complication morbidity and mortality.
Table 1 Perioperative characteristics of TP patients in the Non-CL and CL cohortsTable 2 Surgical outcome of TP patients in the Non-CL and CL cohortsIdentification of independent risk factor for predicting CLOf the 24 variables, 2 predictive features were selected using the LASSO regression analysis (Fig. 2A and B). The Lambda.1se identified a model with optimal predictive ability with minimal predictors, including operation time and venous resection. These two factors screened from LASSO were then applied to identify independent risk factors by multivariable logistic regression analysis, which venous resection (OR = 4.352, 95% CI 1.404–14.04, P = 0.011) was identified as statistically significant risk factor for CL, as demonstrated in Table 3.
Fig. 2Variables selection by the LASSO regression. A LASSO coefficient profiles of the clinical features. B The optimal penalization coefficient lambda was generated via tenfold cross-validation
Table 3 Multivariate logistic regressionPredictive model constructionAlthough there is no statistical difference for operation time (OR = 1.473, 95% CI 1.015–2.237, P = 0.052), we still believe that it is associated with CL occurrence and included these two risk factors in a nomogram (Fig. 3A), aiming to predict the risk of CL for patients who underwent TP. Figure 3B showed our model. Because of the small sample size, we did not perform random assignment to training and validation sets. The area under the curve (AUC) was 0.752 (95% CI 0.593–0.850) (Fig. 3B). After 500 times bootstrap replicates, the observed AUC was 0.752 (95% CI 0.621–0.859) (Fig. 3C), indicating great discriminative ability. We also compared the AUC value of the nomogram with other single predictors, as depicted in Fig. 3D. Note that the AUC of the individual predictors was consistently smaller than the predictive model, underscoring the robust performance of this model. The calibration curves are shown in Fig. 3E, which showed an ideal consistency between the prediction and actual observation after bootstrap replicates. The Hosmer–Lemeshow goodness-of-fit (GOF) test showed X-squared values of 3.3216 (p = 0.9126), suggesting a good fit in our model. The clinical efficacy of our nomogram was evaluated by DCA and CIC curve with fivefold cross-validation and 500 times bootstrap (Figs. 3F and G). The decision curve analysis demonstrated high net benefits in our cohort. Within the probabilities ranging from 5 to 62%, the model outperforms the “treat-all” or “treat-non” strategies, indicating increased net benefits.
Fig. 3Nomogram construction and validation. A Nomogram for predicting CL after TP. B ROC curve for the prediction model. C ROC curve validation by 500 times bootstrap replicates. D Calibration curves of the risk nomogram. E Rationality curve analysis for the CL risk nomogram. F Decision curve analysis. ‘All’ refers to that all patients have CL and ‘none’ means no patient has CL. G The CIC curve of the nomogram. The red solid line refers to the total patients regarded as high risk for each risk threshold. The blue dashed line refers to those would be actual CL patients
Venous resection also increases risk of Grade B CL after TPSince grade B CL is more important in the clinical practice, we compared the perioperative characteristics and surgical outcome for TP patients in the Non-CL/Grade A CL and Grade B CL cohorts. The results showed that the proportion of venous resection (P = 0.019), the number of harvested lymph nodes (P = 0.032), and intraoperative fluid replacement (P = 0.020) were higher in the grade B CL group (Supplement Table 1). Besides, there was no significant difference in surgical outcomes between the two groups (Supplement Table 2). After backward multivariate logistic regression analysis, venous resection (OR = 4.118, 95% CI 0.971–18.80, P = 0.055) and harvested lymph nodes (OR = 1.049, 95% CI 1.000–1.104, P = 0.051) were selected as risk factors for Grade B CL (Supplement Table 3). However, since only 10 patients developed grade B CL in our research, it was insufficient to further construct a valid predictive model.
Postoperative drainage volume analysis in CL patientsNumerous research have shown the link between CL and oral diet, and the characteristics of drainage fluid change with patient's food intake [7,8,9,10]. In addition to the color of the drainage fluid, the volume is also an important indicator of CL. However, no study has reported the volume of drainage fluid in patients with CL after TP. First, we calculated the mean drainage volume of CL patients within 20 days after surgery (Fig. 4A). We found that the mean drainage volume in the CL group remained above 300 ml for a long time after surgery, but seemed to start decreasing after POD14. Then, since venous resection was the independent risk factor for CL in our model, we discovered that the mean drainage volume in venous resection TP group was generally higher than standard TP group (Fig. 4B). Also, grade B CL group also had significantly higher mean drainage volume than grade A CL (Fig. 4C).
Fig. 4A The changing trend of mean drainage volume in CL patients from POD1 to POD20. B The mean drainage volume for patients after standard TP and venous resection TP from POD1 to POD10. C The mean drainage volume of grade A and B CL from POD1 to POD10
Comments (0)