Patients and Baseline Characteristics
Data of consecutive patients with complete records of preoperative haemoglobin were analysed retrospectively. All patients were received open PD at two university hospitals in China between May 2015 and May 2019. Figure 1 showed the flowchart of this study. Local ethics committee approved the usage and publication of these data. Written informed consent was not considered necessary by the ethics committee because of the blinded data and retrospective design. (Approval letter No. 2018BJYYEC-196-02).
Baseline characteristics included age, gender, body mass index (BMI), American Society of Anaesthesiologists (ASA) classification, preoperative obstructive jaundice. Age-adjusted Charlson comorbidity index (aCCI) was used to assess the comorbidities [8]. In China, anaemia was defined according to both the level of haemoglobin and gender. Haemoglobin less than 12g/L in male and 11g/L in female were defined as anaemia [9]. All patients recruited received no preoperative blood transfusion and supplement therapy.
Several nutritional variables such as albumin, nutritional risk and malnutrition were included. The nutritional risk was defined by the tool named nutritional risk screening 2002 (NRS2002) [10] and malnutrition was defined by the global leadership initiative malnutrition (GLIM) diagnosis criteria [11].
Intraoperative and Postoperative Data
The operation method of open pancreatoduodenectomy was unified in our two institutes due to long term cooperation. Intraoperative data included duration of the procedure, volumes of blood loss, intraoperative red blood cell, and fluid infusion. The malignant and benign pathologies were also recorded, especially pancreatic ductal adenocarcinoma (PDAC).
Complications were recorded totally according to the Claviene-Dindo (CD) classification system (Minor: I-II; Major: III-V) [12]. We defined that all postoperative outcomes recorded were happened until discharge. Postoperative pancreatic fistula (POPF) was defined and graded according to the 2016 International Study Group of Pancreatic Surgery (ISGPS) classification and clinically relevant POPF (CR-POPF) contained both grade B and C [13]. Nonfistulous complications like postpancreatectomy haemorrhage (PPH), delayed gastric emptying (DGE), biliary fistula, abdominal infection, cardiac and cerebrovascular events were also included and ISGPS definitions and classifications of PPH and DGE were followed [14,15].
In-hospital reoperation rate, postoperative length of stay (LOS), 30-day readmission rate, perioperative mortality and total hospital costs were recorded. Total hospital costs only contained the direct cost on the hospitalization bill including fees for operation, drugs and medical equipment, nursing care and other medical service such as consultation.
Propensity Score Matching (PSM)
Propensity score matching was applied to achieve a balance between two groups. We selected variables those were significantly different between two groups in the original data analysis by groups comparison and logistic analysis, including age, albumin, Charlson comorbidity index and preoperative obstructive jaundice to generate the propensity score and binary logistic regression with selected variables was used to generate continuous propensity scores from 0 to 1. Patients were matched by a matching ratio 1:1 based on the propensity score with a standard caliper width of 0.02 [16].
Statistical Analysis
The data were collected and checked by two staffs to ensure accuracy at the two institutions. IBM SPSS Statistics (Ver. 26.0, IBM Corp., Armonk, NY, USA) was used to do the statistical analysis by professional statisticians. Categorical data were analysed using the chi-square test or Fisher exact test. Continuous data was tested by Student’s unpaired t test. Charlson comorbidity index was shown by median and interquartile range (IQR), and analysed using Mann-Whitney U test. Multivariable logistic regression analysis was used to evaluate the relationship between risk factors and anaemia and postoperative severe complications respectively, which was expressed as an odds ratio (OR) with 95% confidence intervals. We determined the risk factors by referring to several published articles and what we had in our database, including age, sex, comorbidities, nutrition related variables, pathology and some intraoperative items [7,17]. We did the logistic analysis of the risk factors of anaemia in total cohort in order to reduce the error caused by missing cases and we did the analysis of the risk factors of complications in the paired cohort in order to prevent the influence of bias. P values of less than 0.05 were considered statistically significant.