Alternative Fistula Risk Score and First Postoperative Day Drain Fluid Amylase as Predictors of Pancreatic Fistula After Pancreatoduodenectomy


 Objectives: The alternative fistula risk score (aFRS) and the first postoperative day drain fluid amylase (DFA) are predictors of the occurrence of clinically relevant postoperative pancreatic fistula (CR-POPF). No consensus has been reached on which of the scores is a better predictor; moreover, their combined predictive power remains unclear. To our knowledge, this association had not yet been studied.Methods: This study assessed the predictive effect of aFRS and/or DFA on CR-POPF in a retrospective cohort of 58 patients following PD. The Shapiro-Wilk and the Mann-Whitney tests were applied for assessing the distribution of the samples and for comparing the medians, respectively. The receiver operating characteristics (ROC) curve and the confusion matrix were used to analyze the predictive models.Results: The aFRS values were not statistically different between patients in the CR-POPF and non-CR-POPF groups (Mann-Whitney U test: 59.5, p=0.12). The DFA values were statistically different between the CR-POPF and non-CR-POPF groups (Mann-Whitney U test: 27, p=0.004). The aFRS and DFA were independently less predictive for CR-POPF, compared to combined aFRS + DFA.Conclusions: The combined model involving aFRS>20% + DFA≥5,000 U/L was the most effective predictor of CR-POPF occurrence following PD.


Introduction
The high morbidity and mortality rates of pancreatoduodenectomy (PD), performed for the treatment of periampullary diseases, are mainly associated with pancreaticojejunal anastomosis. Of all anastomoses performed during PD, pancreaticojejunal anastomosis is the most fragile and is susceptible to complications such as clinically relevant postoperative pancreatic fistula (CR-POPF). CR-POPF occurs in 3-45% of PD procedures and increases hospital stay, readmissions, and reoperations. The occurrence of CR-POPF can also delay the start of adjuvant therapy and reduce the overall survival of patients following PD for malignant diseases. Moreover, it consistently increases health care costs [1][2][3][4] . The critical factor for predicting the occurrence of CR-POPF is risk stratification by assessing patient-and procedure-related criteria and creating a fistula risk profile 3 .
The first internationally accepted definition of postoperative pancreatic fistula (POPF) was established in 2005 by the International Study Group of Pancreatic Fistula (ISGPF). 1 It was defined as abdominal drain output of any measurable volume of fluid on and after postoperative day 3 with amylase levels in the drained fluid 3 times higher than the institutional normal serum amylase upper limit. This classification stratified patients into grade A (benign clinical course), grade B (patients at moderate risk), and grade C (critical patients requiring invasive intervention). 1 A 2016 review of the definition of POPF 2 considered grades B and C as CR-POPF and renamed grade A as biochemical leak (BL), a pancreatic secretion extravasation without clinical implications. The grade B POPF is the fistula itself, and requires changes in postoperative management, including oral fasting, enteral or parenteral diet support, and antibiotic therapy in cases of infection. Moreover, it may also require an invasive intervention with percutaneous or endoscopic drainage of intra-abdominal collections. Grade C POPF occurs when grade B POPF is complicated with organ failure, clinical instability, need for reoperation, or death. 2 These terms are summarized in Table 1.
The alternative fistula risk score (aFRS) was described in 2017 4 and externally validated in multicenter studies. [4][5][6][7][8][9] This score considers three predictive factors of CR-POPF: pancreatic parenchyma texture, diameter of the main pancreatic duct (Wirsung's duct), and the patient's body mass index (BMI). The aFRS classifies the risk for developing CR-POPF into three groups: low-(aFRS 0% to 5%), intermediate-(aFRS >5% to 20%), and high-risk (aFRS >20%) groups. 4 These terms are summarized in Table 1. CR-POPF occurrence can also be predicted by measuring the first postoperative day drain fluid amylase (DFA). 10,11 Visual changes in the appearance of the drain fluid only become evident from postoperative day 5, 12 whereas amylase levels in the drain fluid increase already on postoperative day 1 due to its early and imperceptible extravasation through the 4 anastomosis. Several cutoff values of DFA, [11][12][13][14][15][16][17][18][19][20][21][22][23][24] ranging from 90 U/L 21 to 5,000 U/L, 12 have been described for determining the high risk of developing CR-POPF. The most cited cutoff is ≥5,000 U/L; 12 thus, motivating surgeons to remove the abdominal drain early when DFA values are below the aforementioned cutoff. 25 Several studies 4,10-12 have pointed out the importance of measuring the aFRS and DFA. With varying diagnostic values, 4,11-24 the determination of these scores facilitates a more individualized management, considering if the patient is a low-or high-risk of developing CR-POPF. 3,4,[25][26][27][28][29] However, there is still no consensus on which of these scores is better, and whether combining both scores would result in increased accuracy in the selection of patients with a high risk of developing CR-POPF. To our knowledge, this association had not yet been studied. probability; exp: exponential function.

Study design, patients, and data collection
We carried out a retrospective cohort of 58 patients who underwent PD for malignant or benign periampullary disorders. The postoperative progression was based on the length of hospital stay, duration of abdominal drain use, use of antibiotics and parenteral nutrition, and occurrence of pancreatic, biliary, or lymphatic fistulas. The clinical or surgical complications were assessed individually based on the need for interventions, such as percutaneous drainage of intra-abdominal collections or reoperation.
The stratification according to POPF grades followed the ISGPF guidelines. 2 The CR-POPF risk assessment used two predictive scores that had been previously described in the literature: aFRS 4 and DFA. 10,11 The aFRS was calculated using the online calculator available at www.pancreascalculator.com. 4 The abdominal drain fluid was collected on postoperative day 1 to measure the amylase levels. 10,11 The original paper that describes aFRS 4 classified patients with aFRS>20% as having a high risk for CR-POPF; we adopted this value in the present study as well. A DFA of ≥5,000 U/L is considered to indicate a high risk for CR-POPF 12 ; we also adopted this value in the present study.
The study included adult patients of both genders who underwent elective PD due to malignant or benign periampullary disorders. We excluded patients with medical records having insufficient information for calculating the aFRS or without a DFA value. The medical records of 58 patients were initially reviewed and 18 were excluded from the study for insufficient information (Fig. 1).

Statistical analysis
This study assessed the correlation between aFRS and/or DFA and the occurrence of CR-POPF by performing a descriptive analysis of the data comparing the scores individually (aFRS or DFA) or combined (aFRS + DFA). The means, medians, and percentage distribution of each variable were calculated, and the corresponding boxplots were traced. The Shapiro-Wilk normality test was applied to assess the distribution of the samples and the two-tailed non-parametric Mann-Whitney test to compare the medians. The receiver operating characteristic (ROC) curve and the confusion matrix were used to analyze the predictive models.

Biodemographic data
The patients were divided into two groups: those who evolved with (CR-POPF group) and without (non-CR-POPF group) CR-POPF. The biodemographic, surgical, and pathological data are summarized in Table 2.
All surgeries were PDs (Whipple procedure) and had a curative intent in cases of malignant neoplasia.
Laparotomy was the access route used in all patients. The reconstruction strategy consisted of a single loop with a duct-to-mucosa pancreaticojejunal anastomosis. Abdominal drainage was performed with a silicone tubulolaminar drain placed near the pancreaticojejunal anastomosis. Multivisceral and vascular resection was performed in 1 (2.5%) case. The mean surgical time was 338.5 ± 123.42 min (150 to 630 min). The mean estimated intraoperative blood loss was 570 ± 427.71 mL (50 to 1500 mL). The Wirsung's duct diameters were ≤3 mm in 26 (65%) patients and >3 mm in 14 (35%). The pancreatic parenchyma texture was considered firm in 8 Postoperatively, biliary fistula was observed in 3 (7.5%) cases; 2 were surgically and 1 was conservatively treated, and fistulas were resolved in all cases. Lymphatic fistula was observed in 2 (5%) cases, which both progressed favorably after clinical treatment.
There were 6 (15%) cases of pancreatic fistula, namely 4 (10%) with grade B and 2 (5%) with grade C. The mean length of hospital stay was 16 and 12.9 days in the CR-POPF and non-CR-POPF groups, respectively (Mann-Whitney test; p>0.05). The mean duration of abdominal drain use was 15.6 and 11.9 days in the CR-POPF and non-CR-POPF groups, respectively (Mann-Whitney test; p>0.05). be applied to at least 5 cases, and it is noted that for some parameters, the fistula group has only 1 or 2 patients, obtaining an unreliable p value that can lead to confusion in interpretation of the data.

Correlation between aFRS and CR-POPF
The mean aFRS was 22.38% and 12.23% in the CR-POPF and non-CR-POPF groups, respectively. We  The analysis of the variability of aFRS in relation to the presence of CR-POPF demonstrated that the median aFRS was higher in the CR-POPF group compared to that in the non-CR-POPF group (Fig. 3).
The distribution of the aFRS values was non-parametric (p<0.05; Shapiro-Wilk normality test). There were no significant differences between the CR-POPF and non-CR-POPF groups (Mann-Whitney U test: 59.5, p=0.12). No-CR-POPF CR-POPF data in the boxplots in this analysis (DFA and CR-POPF - Fig. 4), which corroborates the strong discriminatory power of DFA.
The distribution of DFA values was non-parametric (p<0.05; Shapiro-Wilk normality test). There were differences between the CR-POPF and non-CR-POPF groups (Mann-Whitney U test: 27, p=0.004). This indicates that DFA has a greater discriminatory power for the occurrence of CR-POPF compared to aFRS.  The values of the metrics of the predictive models are summarized in Table 3.

Discussion
CR-POPF is one of the most serious complications and is responsible for the high morbidity and mortality rates of PD. The morbidity rates range from 30 to 50% in large reference centers; however, the mortality rates have been reduced to less than 5% due to recent advances in perioperative management. 1,2,12 The clinical impact of CR-POPF leads to significantly higher health costs by increasing hospital stay and the need for invasive interventions. 3,4,12 This study assessed the association between aFRS and DFA and the occurrence of CR-POPF and compared the models for predicting the occurrence of CR-POPF.
The incidence of CR-POPF in the present study was 15%, which is in agreement with that reported in the literature. [1][2][3][4] The CR-POPF group showed a trend towards smaller Wirsung's duct diameters and greater BMIs, which are parameters that are used for calculating aFRS. The Wirsung's duct diameter and the pancreatic parenchyma texture are criteria considered subjective since these measurements are intraoperatively determined by the surgeon´s autonomy. However, studies 30,31 suggest that the determination of these criteria by experienced surgeons is accurate and intraoperative measurements can correlate with corresponding findings on preoperative images (three-phase computed tomography or nuclear magnetic resonance). 32 The calculation of the aFRS is inherent to the patient and its use in the literature has shown to be of great value for screening patients, especially those at low risk of CR-POPF occurrence. [4][5][6]8,9 The lower values (aFRS ≤5%) in the present study, although not very accurate, have a high Se and NPV. As predicted by the low risk score (aFRS ≤5%), none of the patients in the present study evolved with CR-POPF. Lao et al. 5 retrospectively evaluated the external validation of aFRS in 370 patients who underwent PD and found a low accuracy (0.46) and high Se (0.92) and NPV (0.94). The present study corroborated these results and, thus, suggests the use of this indicator for selecting patients at low risk for CR-POPF.
The correlation of the aFRS alone with the CR-POPF showed a trend towards fistula occurrence since the aFRS value increased, which agrees with results found in the literature. [4][5][6]8,9 DFA measurement is a simple procedure; however, it requires the presence of the abdominal drain.
Bassi et al. 12 correlated the DFA with the occurrence of CR-POPF and encouraged the strategy of early removal of the abdominal drain depending on DFA values. 25 The cutoff provided in the literature 11-24 ranged from 90 U/L 21 to 5,000 U/L. 12 Giglio et al. 13 performed a meta-analysis with 13 studies and 4,416 patients to assess the accuracy of 11 different DFA cutoff values as predictors of CR-POPF, and found the highest Sp for the value of 5,000 U/L.
The correlation between DFA and CR-POPF showed a trend towards fistula occurrence as DFA values increased, which is in agreement with findings from other studies. [11][12][13][14][15][16][17][18][19][20][21][22][23][24] In the present case series, the DFA showed a high discriminatory power, since the means of the two groups had very different magnitudes. The DFA boxplot visually showed its strong predictive power by a significant independent separation of data.
Moreover, the accuracy of the DFA for predicting the occurrence of CR-POPF was higher than that of the aFRS when the scores were analyzed individually. For these reasons, the DFA alone proved to be a strong predictive parameter for CR-POPF occurrence, and with a greater discriminatory power compared to the aFRS.
The most accurate model in the present study for predicting CR-POPF occurrence was aFRS >20% + DFA ≥5,000 U/L. There was an increase in accuracy when the scores were combined. Notably, this combination used the two values of aFRS 4 and DFA 12 obtained in the literature for determining the high risk for CR-POPF.
The combination of these two scores is of great interest since their methods and advantages for predicting CR-POPF seem broad and overlapping. To our knowledge, this association between aFRS and DFA had not yet been studied.
Therefore, using the aFRS and DFA, it was possible to predict CR-POPF occurrence intraoperatively and on postoperative day 1, respectively, with greater accuracy for the combined aFRS + DFA. Although our sample is small, this information may help in further studies to identify patients at high risk for CR-POPF (aFRS >20% and DFA ≥5,000 U/L) who may be candidates for more rigorous postoperative monitoring, and patients at low risk for CR-POPF (aFRS ≤20% and DFA <5,000 U/L) who may be candidates for fast-track protocols such as early return to normal diet, removal of the abdominal drain, and discharge. Nevertheless, our results need to be confirmed with prospective clinical studies. [23][24][25][26][27][28][29][35][36][37][38][39][40][41][42][43] Kawai et al. 29 argued that early discharge (around postoperative day 5) can be safe and economically beneficial for patients with good progression following PD, while others encourage the omission of the prophylactic drain in patients with low risk of CR-POPF, as well as its early removal when the DFA is <5,000 U/L. [23][24][25][27][28][29][35][36][37][38][39][41][42][43] However, most institutions still use the abdominal drain routinely and remove it on or after postoperative day 7. 29 Postoperative management is planned based on existing complications, which affect the length of hospital stay and the patient's morbidity and mortality. 12 Regarding the length of hospital stay and duration of abdominal drain use, the present case series showed a trend for the CR-POPF group to stay longer in the hospital compared to the non-CR-POPF group, and an association between the decision to remove the drain and the decision to discharge.
This study was limited by the small sample size. Moreover, it was a retrospective and observational cohort study. The association between aFRS and DFA may help differ the patients with low risk from those with high risk of developing CR-POPF, but further studies with a larger number of patients are required to confirm the possible prediction of CR-POPF occurrence with the combined use of the scores.

Conclusion
The data obtained in this study demonstrated that the combined use of aFRS and DFA increased the accuracy for predicting CR-POPF in patients who underwent PD. Prospective studies with a larger number of cases are required to confirm these results.
postoperative pancreatic fistula, Grade B POPF: fistula requiring a change in postoperative management. Grade C POPF: grade B with organ failure, clinical instability, need for reoperation, or death.

Fig. 2. Analysis of the distribution of aFRS categories in relation to the occurrence of CR-POPF.
The patients were divided into three columns, according to aFRS categories: those with low-(aFRS 0% to 5%), those with intermediate-(aFRS >5% to 20%), and those with high-risk (aFRS >20%) for developing CR-POPF.
Each column was divided into patients who developed CR-POPF (dark blue) and patients who not developed CR-POPF (light blue). aFRS: alternative fistula risk score; CR-POPF: clinically relevant postoperative pancreatic fistula.