Risk Identification and Prediction of Myocardial Injury After Laparoscopic Pheochromocytoma/Paraganglioma Resection:A Ambispective Cohort Study.

MI-PPGL is a single-center observational ambispective cohort study.On the basis of retrospective study, the research team plans to build a structured database to investigate the incidence of myocardial injury in laparoscopic PPGL-resection, and further analyze myocardial injury related risk factors. In particular, timing data such as vital signs(blood pressure,heart rate)will be included to construct an efficient and robust myocardial injury prediction model. At the same time, a prospective cohort study is carried out to verify the model, so as to test the prediction ability of myocardial injury and reduce the incidence of myocardial injury.

The investigators expect to enroll 700 patients, including at least 550 patients retrospectively and 150 patients prospectively.In this study, the main endpoint events of the prediction model are binary outcome. Conservatively estimated according to the "10EPV" principle, that is, each predictive factor included in the model needs at least 10 positive outcome endpoint for estimation (10 events per variable). The investigators expected 5 to 8 predictors to be included in the model, and at least 80 positive events to be included. The incidence of perioperative myocardial injury is 12~20%, so the estimated sample size was at least 666 patients. Considering the absence of data or subject withdrawal from the study. so the investigators expected to include 700 patients, including at least 550 retrospectively and 150 prospectively.

STATA (version 15.0; Stata Corp., TX, USA) and R 3.6.1 software (R Foundation for Statistical Computing, Vienna, Austria) will be used for statistical analysis. Binary logistic regression was used to screen risk factors and stratify risk levels. P<0.05 was considered statistically significant. For predictive modeling, clinical databases were 9:1 or 8: 2. Randomly split into training samples and verification samples. In the training samples, optimal subset method and LASSO regression will be used for feature selection.Receiver operating characteristic curve (ROC curve) was used to represent the model differentiation, and Nomogram was used to represent the predictive factors of multiple logistic regression. In the verification samples, Hosmer-Lemeshow goodness of fit test was used to test the calibration degree of the model, and P>0.05 was the acceptable level of estimated fitting of the model. Decision curve analysis (DCA curve) was used to verify the clinical applicability.