Perioperative fluids and complications after pancreatoduodenectomy within an enhanced recovery pathway

Optimized fluid management is a key component of enhanced recovery (ERAS) pathways. Implementation is challenging for pancreatoduodenectomy (PD) and clear guidance is missing in the respective protocol. The aim of this retrospective study was to evaluate the influence of perioperative intravenous (IV) fluid administration on postoperative complications. 164 consecutive patients undergoing PD within ERAS between October 2012 and June 2017 were included. Perioperative IV fluid and morbidity (Clavien classification and comprehensive complication index (CCI)) were assessed. A threshold of more than 4400 ml IV fluid during the first 24 h could be identified to predict occurrence of complications (area under ROC curve 0.71), with a positive and negative predictive value of 93 and 23% respectively. More than 4400 ml intravenous fluids during the first 24 h was an independent predictor of overall postoperative complications (adjusted odds ratio 4.40, 95% CI 1.47–13.19; p value = 0.008). Patients receiving ≥ 4400 ml were associated with increased overall complications (94 vs 77%; p value < 0.001), especially pulmonary complications (31 vs 16%; p value = 0.037), as well as a higher median CCI (33.7 vs 26.2; p value 0.041). This threshold of 4400 ml intravenous fluid might be a useful indicator for the management following pancreatoduodenectomy.

As postoperative complications were higher in patients with increased perioperative fluids, the potential correlation between perioperative fluids and postoperative complications was assessed. As presented in Fig. 1, the amount of IV fluids POD 0 weakly correlated with CCI (r = 0.168, p value = 0.020).
Fluid threshold. In order to identify a critical threshold, a ROC curve analysis of IV fluids POD 0 (ml/24 h) and overall complications was performed (Fig. 2). The Area Under the Curve (AUC) was 0.71 (95% CI 0.59-0.83), which is above the 0.70 cut-off indicating moderate accuracy 8 . The optimal threshold of IV fluids POD 0 as predictor of complications was set at 4400 ml. This threshold provided a sensitivity of 69% and a specificity of 63% with a positive predictive value of 93% and a negative predictive value of 23%. The ROC curve analysis for IV fluids POD 0 and major complications was not contributive (AUC = 0.55). Further ROC curve analyses of IV Table 1. Patients preoperative and surgical characteristics. ASA American Society of Anesthesiology; BMI Body Mass Index, SD standard deviation, n number, IQR interquartile range; WHO World Health Organization; IV intravenous; POD postoperative Day.

Overall (n = 164)
Age, years, mean (SD) 65 (12) Gender (m:f) 86:78 BMI, kg/m 2 , mean (SD) 25.6 (4.7) ASA, n (%)    Predictors for overall and major complications. In order to evaluate the potential predictive role of perioperative fluid on postoperative complications, a binary logistic regression analysis with preoperative and perioperative characteristics including the identified 4400 ml threshold of IV fluids on POD 0 was performed and presented in Tables 2 and 3. On multivariable analysis, the 4400 ml IV fluids POD 0 was an independent predictor of overall postoperative complications (adjusted OR 4.40, 95% CI 1.47-13.19; p value = 0.008), but not of major complications.
Postoperative outcomes according to critical fluid threshold. According to this 4400 ml of IV fluids POD 0, postoperative outcome was compared as shown in Table 4. There were 61 patients who received less  www.nature.com/scientificreports/ than 4400 ml and 103 patients who received 4400 ml or more. The overall complication rate and the CCI were significantly higher in the ≥ 4400 ml group (94 vs 77%; p value = 0.001, and 33.7 vs 26.2; p value = 0.041, respectively). The 90 days after surgery mortality was not different in both groups: 3% in the < 4400 ml group (one patient died of hemorrhagic shock and one patient of septic shock) and 2% in the ≥ 4400 ml group (one patient died of bleeding shock and one patient of disease progression with carcinosis after 61 postoperative days); p value = 0.629. The rate of postoperative pulmonary complications was significantly increased in patients receiving higher perioperative volume (31 vs 16%; p value = 0.037). The increased rate of respiratory complications in patients receiving more than 4400 ml was especially observed for minor (Clavien I-II) complications (21 vs 5%; p value = 0.006). There was no significant difference among detailed pulmonary complications, such as atelectasis (8 vs 5%; p value = 0.748), pneumonia (18 vs 10%; p value = 0.179), pleural effusion (12 vs 10%; p value = 0.800), respiratory failure (10 vs 5%; p value = 0.332), and pneumothorax (1 vs 0%; p value = 1.000). Regarding hemodynamic related complications, there was no significant difference in terms of cardio-vascular dysfunction (13 vs 10%; p value 0.500), and of renal failure (3 vs 5%; p value = 0.322). Between the two groups. In the < 4400 ml group, one patient had a renal failure classified AKIN (Acute Kidney Injury Network) 1 and one patient AKIN. 3.
In the group with more than 4400 ml, two patients were classified AKIN 1 and three patients AKIN 3. No significant difference in pancreatic surgery specific complications such as delayed gastric emptying or pancreatic fistula was observed between both groups. The use of postoperative nasogastric tube was 3% (2/61) in the < 4400 ml group compared to 17% (18/103) (p value = 0.007). The rate of nasogastric tube reinsertion was comparable (43% vs 42%; p value = 0.913). The rate of postoperative parenteral nutrition started in the postoperative period was not different between both groups (55 vs 49%; p value = 1.000). The number of relapartomy needed for postoperative hemorrhage was 8 (8%) in patients with ≥ 4400 ml compared to 5 (8%) in patients with < 4400 ml (p value = 1.000). The proportion of patients requiring intensive care unit was 21% in the < 4400 ml group and 19% in the ≥ 4400 ml group (p value = 0.841).

Discussion
This study described a higher administration of perioperative IV fluids among patients presenting with postoperative morbidity. The correlation of perioperative IV fluids and postoperative cumulative complications expressed by CCI was weak and meant that a small proportion of complications could be correlated to the amount of IV fluids. However, a critical threshold of 4400 ml IV fluid administration during the first 24 h was identified as an independent predictor of postoperative complication. More than 4400 ml IV fluids POD 0 was especially associated with increased respiratory complications, but neither pancreatic fistula nor delayed gastric emptying. www.nature.com/scientificreports/ The issue on optimal fluid management in abdominal surgery is still open to debate. Previous studies assessing fluid therapy and outcome in PD perfectly reflects this situation. A fair amount of published studies including PD, predominantly retrospective cohort, described an increased rate of postoperative complications with increased fluids [9][10][11][12][13] . However, most randomized controlled studies [14][15][16] failed to show any significant difference in postoperative outcome when comparing liberal to restrictive fluid therapy.
Comparison between studies on fluid therapy in PD is difficult because of the various definition of fluids (units, measure, and duration) and heterogeneous data. Moreover, most studies did not encompass the use of a systematic enhanced recovery protocol. Nevertheless, our center has a more restrictive fluid administration as in some other centers with a median intraoperative fluid administration of 3500 ml in comparison to 6000 ml for Kulemann et al. 10 and 5000 ml for Weinberg et al. 9 . This is reflected in the intraoperative infusion rate too, as we have a median intraoperative infusion rate of 9 ml/kg/h (IQR 7-11), in comparison of 14 ml/kg/h for Eng et al., and corresponding to a restrictive fluid administration (< 10 ml/kg/h) 11,13 , although no standard definition exists. The rate of renal failure in the present study (3 and 5% in the < 4400 ml and ≥ 4400 ml groups, respectively) was in accordance to the 7% rate reported by Weinberg et al. 9 . Other parameters were comparable with the literature: similar demographics in term of age (mean age 65, going from 53.5 to 69.0 in a systematic review 17 ), gender (male 52% versus 50.3% 10 to 59% 9 ), mean BMI (25.6 kg/m 2 versus 25.8 kg/m 212 to 26 kg/m 29 ). As in most studies, a large majority of patients had malignant tumor as indication for PD (86% versus 73.4% 10 to 88% 12 ). The ASA score may be lower in the present study with 33.5% of patients with ASA-score of 3 or 4, which ranged from 34.5 10 to 89.5% 12 in other studies. In term of operative parameter, the mean operative time tend to be lower in our data (335 min versus 420 min 9 to 445 min 11 ) but with similar vascular resection rate (22% versus 24% for Behman et al. 12 ) and blood loss (median blood loss 440 ml versus 350 ml 9 to 909 ml 11 ).
As avoiding fluid overload might improve postoperative outcome after PD, the way to achieve this euvolemia is still matter of debate. Goal directed fluid therapy requires cardiac output monitoring that may be invasive and lacks easily identifiable target. For these reasons it is a challenging measure to implement in daily clinical practice. Further guidance are awaited from the OPTIMISE II trial 18 , a multicenter international trial of cardiac output-guided fluid therapy with low-dose inotrope infusion compared with usual care in patients undergoing major elective gastrointestinal surgery. Meantime, a threshold of perioperative IV fluid is a simple and reproducible way to estimate the fluid balance. The ROC analysis identified a threshold of 4400 ml perioperative fluid for complications. Previously, specific thresholds for perioperative IV fluid were identified, for example for open colorectal surgery (> 3500 ml) 19 or loop ileostomy closure (> 1700 ml) 20 . Although the above mentioned threshold of 3500 ml for perioperative fluids was determined for open colorectal surgery, it was previously used by extrapolation for pancreatodudenectomy 7 . Data for pancreatic resection are sparse and Bruns et al. identified a ratio (infusion rate/glomerula function rate) of 0.15 as predictive of pancreatic fistula 21 . Available data for liver surgery identified a threshold of an increased weight of 3.5 kg on the second postoperative day which was associated with increased major complications 22 , but no specific data on fluids were provided. In the present study, the weight difference did not vary between patients with or without complications, possibly indicating that fluid balance evaluation is a complex process.
Noteworthy, the 4400 ml threshold was related to increased complications and length of stay. Especially postoperative minor pulmonary complications were increased in patients receiving 4400 ml or more. Similarly, Eng et al. reported an increased rate of pulmonary complications (29 vs 5%, p < 0.01) related to intraoperative infusion when comparing patients receiving more or less than 13.95 ml/kg/h 11 . The meta-analysis of Garland et al 23 found no difference between restrictive and liberal fluid therapy on overall morbidity (Odds Ratio 1.17, 95% CI 0.92-1.50) and pulmonary complications (Odds Ratio 0.69, 95% CI 0.43-1.10). While causes of postoperative respiratory complications are multifactorial, a recent systematic review 24 identified enhanced recovery (risk ratio 0.35, 95% CI 0.21-0.58) and goal directed therapy (risk ratio 0.87, 95% CI 0.77 to 0.98) among other interventions as effective measure to reduce their occurrence. Postoperative pulmonary complications are probably also related to other complications, but the study design as well as the lack of date of occurrence of each complications prevented to conduct a separate detailed analysis of predictive factors, including other complications, of pulmonary complications. As enhanced recovery also encloses goal-directed fluid therapy, the direct effect of fluid management is difficult to isolate. The threshold of 4400 ml would be more on the restrictive side as opposed to the liberal side when considering the various cut-offs reported in the systematic review on fluid regimens in pancreatoduodenectomy by Garland et al 23 . A fear of a too restrictive fluid regimen is potential tissue hypoperfusion leading to an increased rate of renal failure or altered wound healing. For example, the RELIEF (Restrictive versus Liberal Fluid Therapy for Major Abdominal Surgery) trial observed an increased rate of acute kidney injury associated with restrictive fluid therapy (8.6% in the restrictive fluid group and 5.0% in the liberal fluid group) among high risk patient undergoing major abdominal surgery 25 . Of notice, in the RELIEF trial the median intravenous fluid during and up-to 24 h surgery was 3700 ml in the restrictive group and 6100 ml in the liberal group. In the present study no increased renal failure was observed in patients receiving less than 4400 ml.
Regarding pancreatic-specific complications such as delayed gastric emptying and pancreatic fistula, the present study did not observed increased risk associated with increased perioperative fluid infusion. This is in line with Garland et al. 23 who reported in a meta-analysis no significant difference in occurrence of delayed gastric emptying or pancreatic fistula when comparing restrictive to liberal fluid therapy in PD patients. However, some retrospective studies not included in the abovementioned meta-analysis [26][27][28] were suggesting an increase in pancreatic fistulas associated with excessive fluid. Moreover, a study from Bannone et al. reported an increased incidence of postoperative acute pancreatitis and pancreatic fistula in high-risk patients after pancreatoduodenectomy with an intraoperative fluid administration of ≤ 3 ml/kg/h 29 . In the present study, the median intraoperative fluid administration was 9 ml/kg/h and no increased rate of pancreatic fistula was observed between both groups.
Limitations of the present study are inherent to the retrospective analysis and to the limited study sample obtained in a single center. Moreover, enhanced recovery relies on compliance to protocol elements, including Scientific Reports | (2020) 10:17898 | https://doi.org/10.1038/s41598-020-74907-y www.nature.com/scientificreports/ fluid management, which has altogether a strong influence on the postoperative outcome 7 . However, in order to avoid redundancy with previously published results 7 , this study focused on fluid management and impact of ERAS compliance was not assessed here. In addition, the amount of IV fluid administered in the first 24 h of surgery is not an entirely modifiable factor, as it is closely related to patient's comorbidity, disease's prognosis and surgery's extent. In order to validate and generalize this threshold of 4400 ml iv fluids on POD 0 and associated findings, further study including PD performed within an enhanced recovery setting need to be conducted. Once available, this will provide useful guidance on how to refine fluid management of patients undergoing PD to avoid both fluid overload and kidney insufficiency.
In conclusion, the present study emphasizes the potential deleterious consequence of fluid overload especially on pulmonary complications. A threshold for perioperative intravenous fluid at 4400 ml might be a useful indicator in the management after pancreatoduodenectomy within an enhanced recovery pathway.

Methods
Study design. This retrospective cohort study included all consecutive patients undergoing elective PD within an ERAS program in a tertiary referral centre (Lausanne University Hospital (CHUV), Switzerland) between October 2012 and June 2017. This cohort of patients was previously included and merged with data of three other institutions in a multicentre study 7 . Patients younger than 18 years old, and those who were opposed to the use of their data were not included. The study protocol was approved by the Institutional Review Board (CER-VD # 2016-01815) and patients gave informed consent. This study was reported in accordance with the STROBE guidelines 30 .
Enhanced recovery protocol, fluid management and data collection. The enhanced recovery protocol for PD was initiated in October 2012 according to the ERAS guidelines 5 and was previously detailed 7 . During the surgery, an infusion of Ringer Lactate at a rate of 3-5 ml/kg/h was initiated and bolus of 500 ml were given in case of decreased cardiac output (assessed by pulse pressure variation, oesophageal Doppler or transoesophageal echocardiography) or increased lactate with a minimal diuresis set at 0.5 ml/kg/h. Vasopressors were liberally used during and after the operation to counterbalance the vasodilator effect of epidural. In the postoperative period, an infusion of 1000 ml/24 h of Ringer Lactate was instilled, with 250 ml bolus of Physiogel in case of diuresis less than 2 ml/kg/4 h. Free oral drinks were encouraged from the day of surgery. Preoperative and demographic characteristics, intraoperative data and postoperative outcomes were prospectively collected using the ERAS Interactive Audit System (EIAS). The postoperative follow-up was 30 days.
Fluids assessment and outcomes. Perioperative fluids were defined as intravenous (IV) infusion, including crystalloids, colloids and blood products given during the first 24 h of surgery (postoperative day 0: POD 0). The total IV volume administration on POD 0 (IV fluids POD 0) was used as parameter to define critical threshold.
Postoperative complications were graded using the Clavien classification 31 . Grade I and II were defined as minor and grade IIIa to IVb as major complications. Grade V was reported as mortality. The Comprehensive Complication Index 32 (CCI) was calculated to obtain a continuous morbidity scale. Complications, including postoperative pulmonary complications were reported according to the EPCO (European Perioperative Clinical Outcome) definitions 33 . Accordingly, postoperative pulmonary complications included respiratory infection (antibiotics for suspected respiratory infection and with one or more criteria: sputum, lung opacities, fever, with blood cell > 12 × 10 9 /l), respiratory failure (postoperative PaO 2 < 8 kPa, PaO 2 : FIO 2 ratio < 40 kPa, or arterial saturation < 90% requiring oxygen therapy), pleural effusion (demonstrated by chest radiograph), atelectasis (lung opacification with shift toward affected area and compensatory over-inflation in the adjacent non-atelectatic lung), pneumothorax (air in the pleural space), bronchospasm (newly detected expiratory wheezing treated with bronchodilators) and aspiration pneumonitis (acute lung injury after the inhalation of regurgitated gastric content. Primary outcome was overall postoperative morbidity. Secondary outcomes were major complications, CCI, and length of stay. Primary length of stay was the number of days spent in the hospital from the primary operation until discharge. Total length of stay was the addition of primary length of stay with the number of days following hospital readmission. Specific complications related to pancreas surgery were reported according to the respective ISGPS guidelines 34-37 . Statistical analysis. Descriptive statistics for categorical variables were reported as numbers and percentages, while continuous variables reported as median and interquartile ranges (IQR) or mean and standard deviations (SD). Categorical variables were compared with the Chi-Square or Fisher's exact test. The Student's t test and Mann-Whitney U test were used for continuous normally and non normally distributed data, respectively. Correlation between CCI and perioperative fluids was assessed by the Spearman's test. The diagnostic ability was illustrated by a receiver operating characteristic (ROC) curve. An area under the curve (AUC) of > 0.7 was considered statistically significant. Threshold was identified according to the Youden index. Potential predictors of overall and major complications were assessed by binary logistic regression analysis and multiple logistic regression analysis was further performed including all previous independent variables with a p value < 0.1. All p values were two-sided and p < 0.05 was considered statistically significant. Data analysis was performed with SPSS version 25 (IBM Corp, Amonk, NY).