Evaluation of a predictive model for pancreatic fistula based on amylase value in drains after pancreatic resection

The American Journal of Surgery (2014) -, -–-

Evaluation of a predictive model for pancreatic fistula based on amylase value in drains after pancreatic resection Stefano Partelli, M.D.1, Domenico Tamburrino, M.D., Stefano Crippa, Ph.D.1, Enrico Facci, M.D., Claudio Zardini, M.D., Massimo Falconi, M.D.*,1 Department of Surgery, Ospedale Sacro Cuore Don Calabria, Negrar, Verona, Italy KEYWORDS: Pancreatic fistula; Drain; Amylase; Surgery

Abstract BACKGROUND: Amylase value in drains (AVD) is a predictor of pancreatic fistula (PF). We evaluated the accuracy of an AVD-based model. METHODS: Two hundred thirty-one patients underwent pancreatoduodenectomy with pancreaticojejunostomy (PDPJ) or pancreatoduodenectomy with duct-to-mucosa (PDDTM) and distal pancreatectomy (DP). Patients with AVD greater than 5,000 U/L on postoperative day (POD) 1 underwent AVD measurement on POD5. RESULTS: Sensitivity and specificity of POD1 AVD greater than 5,000 in predicting PF were 71% and 90%, respectively. The sensitivity and specificity of POD5 AVD greater than 200 were 90% and 83%, respectively. AVD greater than 1,000 (for PDPJ) and 2,000 U/L (PDDTM and DP) represented the most accurate cutoffs on POD1. AVD greater than 200 (PDPJ), 300 (PDDTM), and 50 U/L (DP) represented the cutoffs with the highest sensitivity in predicting PF on POD5. CONCLUSION: AVD-based model for predicting PF after pancreatic resection is an accurate tool, although AVD cutoffs should be evaluated for each type of operation. Ó 2014 Elsevier Inc. All rights reserved.

Pancreatic fistula (PF) is the most common and challenging complication after pancreatectomy.1–3 The presence of a PF is associated with a higher mortality risk, a longer length of hospital stay, increased costs, delayed administration of adjuvant treatments, and poorer quality of life.4–6 Different predictors of PF have been proposed.7–12 Although * Corresponding author. Tel.: 139-0715965781; fax: 139 071 596 4669. E-mail address: [email protected] Manuscript received December 2, 2013; revised manuscript March 28, 2014 1 Present address: Pancreatic Surgery Unit, Ancona University Hospital-Universita` Politecnica delle Marche, Ancona, Italy. 0002-9610/$ - see front matter Ó 2014 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.amjsurg.2014.03.011

a correct prognostication of a PF cannot prevent the occurrence of this complication, the postoperative prediction of PF can influence the management of abdominal drains preventing an early or late removal. In 2005, Molinari et al13 conducted a prospective study that demonstrated the accuracy of a predictive model based on amylase value in drain (AVD) measured on postoperative day (POD) 1 and POD5. On the basis of their results, the same group published a randomized clinical trial that showed the benefit in terms of complications in the group of patients with an early drain removal after pancreatic resection.14 The AVD predictive model was the only criterion for the management of abdominal drains. Nevertheless, this model was estimated on the

2

The American Journal of Surgery, Vol -, No -, - 2014

basis of a large cohort that included both distal pancreatectomy (DP) and pancreatoduodenectomy (PD). Moreover, the external validity of the AVD predictive model has not been demonstrated so far. The aim of this study was to validate the accuracy of the AVD-based model in predicting PF in a cohort of consecutive patients who underwent pancreatic resection.

PD. The pancreatic texture was defined as ‘‘soft’’ or ‘‘firm’’ by the evaluation of the operating surgeon. The MPD diameter was measured by the pathologist on final histologic examination. In all patients, AVD was measured in both drains on POD1. AVD was also measured on POD5 for those patients who maintained at least one drain. Drains were usually removed by the operating surgeon on the basis of POD1 AVD and/or POD5 AVD. Nevertheless, as drain management protocol was not standardized, some surgeons applied the protocol proposed by Molinari et al,13 whereas some others evaluated also the quality of fluids as well as intraoperative findings (ie, pancreatic texture, MPD diameter, bleeding, risk of biliary fistula). In all the cases, drains were not removed in the presence of fresh blood, biliary and/or enteric liquid. Drains were always left in place until POD5 if POD1 AVD greater than 5,000 U/L. PF was defined according to the International Study Group of Pancreatic Fistula as any measurable volume of fluid on or after POD3 with amylase content greater than 3 times the serum amylase activity.15 PF was then classified as grade A, B, or C as defined by the International Study Group of Pancreatic Fistula guidelines.15

Methods Study population Between January 2011 and May 2012, 231 consecutive patients underwent PD or DP. For all patients, demographics, pathologic examinations, operative details, and postoperative outcomes were retrospectively collected and analyzed. Histology revealed a ductal adenocarcinoma in 116 patients (50%), a neuroendocrine neoplasm in 30 patients (13%), and a cystic tumor in 35 (15%) patients. In 27 patients (12%), pancreatic resection was performed for other reasons (chronic pancreatitis, metastatic lesions, or other uncommon tumors).

Surgical procedure For PD, a pylorus-preserving procedure was always performed. Reconstruction of the pancreatic remnant included both PD with pancreaticojejunostomy (PDPJ) and PD with duct-to-mucosa (PDDTM) anastomosis. Each surgeon was free to choose the type of pancreatic anastomosis, although PJ has been the preferred reconstruction in the first months of the study. PJ was performed with a single-layer interrupted suture using nonabsorbable stiches between the pancreatic capsule and jejunal seromuscular layer. DTM was performed using 8 interrupted 4.0 or 5.0 polydioxanone (PDS-II; Johnson and Johnson Co., New Brunswick, NJ, USA) between the pancreatic duct and jejunal mucosa. Two ‘‘easy flow’’ drains were routinely placed (12 mm; Chimed R Livorno, Livorno, Italy) adjacent to the anastomosis. The right-sided drain was placed posterior to the hepaticojejunostomy and anterior to the PJ. The left drain passed posterior to the PJ and anterior to the hepaticojejunostomy. DP always included en bloc splenectomy and the pancreatic stump was always sutured with interrupted nonabsorbable stitches. The main pancreatic duct (MPD), when identified, was routinely sutured with a single nonabsorbable stitch. One ‘‘easy flow’’ drain was routinely placed (12 mm; Chimed R Livorno) near the pancreatic stump. Another drain was routinely placed above the superior pancreatic margin.

Perioperative management Postoperative management of patients did not include a specific protocol. Prophylactic octreotide was administered to prevent PF only in those patients who underwent

Statistical analysis Distribution of continuous variables is reported as median and interquartile range (IQR) (25th and 75th percentiles). Categorical variables are presented as numbers and percentages. The comparison between subgroups was carried out using Student t test or Mann–Whitney U test for continuous variables. Qualitative data were compared using the chi-square test or Fisher’s exact test when necessary. Study of potential prognostic factors for PF was carried out using logistic analysis. Logistic regression was performed for multivariate models with P values and 95% confidence intervals estimated by the Wald method. The predictive power of POD1 AVD and POD5 AVD was assessed by calculating the area under the receiver–operator characteristic (ROC) curve. All tests were 2-sided. Statistical analyses were performed using SPSS 16.0 (SPSS, Inc, Chicago, IL). P values were considered significant when less than or equal than .05.

Results Clinical and operative characteristics The main demographics, clinical and operative characteristics are listed in Table 1. The overall rate of PF was 36% (n 5 83). The frequencies of PF for DP, PDPJ and PDDTM were 59%, 35% and 16%, respectively (P 5 .007). The univariate and multivariate analyses of PF predictors are summarized in Table 2. On multivariate analysis, independent predictors of PF were the type of operation (DP [odds ratio, OR 5.395], P 5 .025 and PDPJ [OR

S. Partelli et al. Table 1 and DP

Drain amylase predict pancreatic fistula

3

Clinical and operative characteristics of 231 patients who underwent pancreatic resection with comparison of PDPJ, PDDTM,

Variable

PDPJ, n (%)

PDDTM, n (%)

DP, n (%)

P value

Sex (male) Age (years)* Operative time (minutes)* Blood transfusion (yes) Postoperative complications (yes) Pancreatic fistula (yes) Relaparotomy (yes) Length of hospital stay (days)*

51 64 330 18 39 27 0 13

38 58 340 11 32 13 4 11

32 73 190 4 47 43 4 11

.011† .243 ,.0001† .007† .009† ,.0001† .147 .005†

(66) (58–70) (300–380) (23) (51) (35) (0) (9–26)

(47) (50–70) (300–400) (14) (39.5) (16) (5) (9–15)

(44) (51–71) (162–235) (5.5) (64) (59) (5) (9–15)

DP 5 distal pancreatectomy; PDDTM 5 pancreatoduodenectomy with duct-to-mucosa anastomosis; PDPJ 5 pancreatoduodenectomy with pancreatojejunostomy. *Values are expressed as median (interquartile range). † P values were considered significant when less than or equal than .05.

7.497], P 5.014), the presence of a ‘‘soft’’ pancreatic texture (OR 5.258, P 5 .003), and a POD5 AVD greater than 200 U/L (OR 16.852, P 5 .001).

Evaluation of AVD-based model The median POD1 AVD and POD5 AVD for the entire cohort were 2,568 (IQR 165–23,414 U/L) and 454 U/L (IQR 60–2,064 U/L), respectively. Overall, 73 patients Table 2

(32%) had POD1 AVD greater than 5,000 U/L. POD5 AVD was measured in 142 patients (61%). Of those 142 patients, 83 (58%) had POD% AVD . 200 U/L. The sensitivity and specificity of POD1 AVD greater than 5,000 in predicting PF were 71% and 90%, respectively. The sensitivity and specificity of POD5 AVD greater than 200 in predicting PF were 90% and 83%, respectively. Fifty-six patients (24%) had an AVD-based predictive model of PF (POD1 AVD . 5,000 U/L or POD1 AVD % 5000 with POD5 AVD . 200

Univariate and multivariate analysis of predictors of pancreatic fistula after pancreatic resection

Variable Sex Male Female Age (years) %60 .60 Blood transfusion No Yes Type of operation PDDTM PDPJ DP Pancreas texture Pancreas firm Pancreas soft Main pancreatic duct diameter (mm) .4 %4 POD1 AVD (U/L) %5,000 .5,000 POD5 AVD (U/L) %200 .200

OR

95% CI

P value

OR

95% CI

P value

1 .965

d .563–1.652

.896

1 1.009

d .589–1.728

d .974

1 2.125

d 1.010–4.471

.047*

1 3.315

d .786–13.970

d .103

1 7.497 2.825

d 3.525–15.945 1.327–6.014

d ,.0001* .007*

1 4.330 5.395

d 1.201–15.614 1.412–20.610

d .025* .014*

1 18.941

d 8.465–42.381

d ,.0001*

1 5.258

d 1.760–15.709

d .003*

1 9.480

d 3.278–27.412

d ,.0001*

1 1.766

d .278–11.216

d .546

1 18.087

d 9.018–36.274

d ,.0001*

1 2.583

d .866–7.705

d .089

1 21.359

d 10.390–43.911

d ,.0001*

1 16.852

d 6.844–41.495

d .001*

AVD 5 amylase value in the drain; CI 5 confidence interval; DP 5 distal pancreatectomy; OR 5 odds ratio; PDDTM 5 pancreatoduodenectomy with duct-to-mucosa anastomosis; PDPJ 5 pancreatoduodenectomy with pancreatojejunostomy; POD 5 postoperative day. *P values were considered significant when less than or equal than .05.

4

The American Journal of Surgery, Vol -, No -, - 2014

Evaluation of an AVD cutoff for pancreatic fistula prediction

Figure 1 Histogram depicting the median AVD on POD1 grouped according to different types of operation.

U/L). Of those 56 patients, 52 patients (93%) had a PF and they represent only the 62% of patients with PF (n 5 83). Among 31 patients with a negative AVD-based model, an abdominal collection occurred in 29 (94%) patients. The median POD1 AVD and POD5 AVD as well as the AVD sensitivity and specificity differed from operation to operation. In particular, the median POD1 AVD after DP was significantly higher compared with median POD1 AVD after PDPJ and PDDTM (7,363 vs 1,554 vs 521 U/L, P , .0001) (Fig. 1). Similarly, the median POD5 AVD reflected the same significant differences between the 3 types of operation (838 vs 360 vs 193 U/L, P , .0001).

Figure 2 ROC curve based on amylase value in drains on POD1 (area under the curve: .876; P , .0001).

The area under the ROC curve for POD1 AVD was .876 (P , .00001) for the entire cohort (Fig. 2). The area under the ROC curve for POD1 AVD was .860 (P , .00001) after PDPJ, .950 (P , .00001) after PDDTM, and .726 (P 5 .001) after DP. The area under the ROC curve for POD5 AVD was .875 (P , .00001) after PDPJ, .894 (P , .00001) after PDDTM, and .928 (P , .0001) after DP. AVD greater than 1,000 and 2,000 U/L represented the cutoffs associated with the highest sensitivity in predicting PF on POD1 for PDPJ, PDDTM, and DP, respectively. AVD greater than 200, 300, and 50 U/L represented the cutoffs associated with the highest sensitivity in predicting PF on POD5 (Table 3). According to the highest sensitivity value, we then considered 2 different cutoffs of POD1 AVD that were 1,000 U/L for PDPJ and 2,000 U/L for PDDTM and DP. Positive and predictive values of these cutoffs are illustrated in Table 4.

Comments This study demonstrates that AVD-based model for predicting PF after pancreatic resection is a valid and accurate tool, although AVD cutoffs should be evaluated for each type of operation and they possibly differ from institution to institution. PF after pancreatic surgery represents the most challenging complication and the main reason behind mortality rate after these operations.1–6 Several studies analyzed predictive factors associated with the risk of developing PF. These factors include age older than 65 years, preoperative jaundice, soft pancreatic texture, small pancreatic duct, exocrine pancreatic function, longer operative time, high-volume transfusion, and intraoperative bleeding.7–12 Some of these variables were also considered in the current analysis as we found that pancreatic texture and the type of pancreatic anastomosis were independent predictors of PF. The possibility of predicting PF preoperatively involves a better selection of patients by balancing the benefits from surgery with the probabilities of having this complication postoperatively. The prediction of PF in the first postoperative days allows a more accurate and safer management of abdominal drains. An early drain removal could be associated with the development of abdominal collection when a PF is not promptly detected. On the other hand, the presence of long-term prophylactic abdominal drains is associated with a significant risk of ascending infections.16 It has been observed that positive cultures of drainage fluid on POD7 increased to 31% when drains are removed on POD8.16 Therefore, drain removal at POD4 or POD5 reduces the rates of intra-abdominal infections. Moreover, a randomized clinical trial demonstrated a significant increase in the rate of abdominal collections among patients who were drained after PD compared with those who were

S. Partelli et al. Table 3 PDPJ

PDDTM

DP

Drain amylase predict pancreatic fistula

5

Cutoff variation of sensitivity and specificity of POD1 AVD and POD5 AVD POD1 AVD (U/L)

Sensitivity (%)

Specificity (%)

POD5 AVD (U/L)

Sensitivity (%)

Specificity (%)

1,000 1,000–2,000 .2,000 %2,000 2,000–3,000 .3,000 %2,000 2,000–4,000 .4,000

96–100 78–96 %78 100 85–100 %85 97–100 84–97 %84

0–74 74–86 .86 0–85 85–87 .87 0–37 37–54 .54

200 200–400 .400 %300 300–450 .450 %50 50–200 .200

93–100 88–93 %88 100 83–100 %83 93–100 85–93 %85

0–88 88–94 .94 0–80 80 .80 0–57 57–88 .88

AVD 5 amylase value in the drain; DP 5 distal pancreatectomy; PDDTM 5 pancreatoduodenectomy with duct-to-mucosa anastomosis; PDPJ 5 pancreatoduodenectomy with pancreatojejunostomy; POD 5 postoperative day; POD1 AVD 5 postoperative day 1 amylase value in drains; POD5 AVD 5 postoperative day 5 amylase value in drains.

not drained.17 These results support the hypothesis that drains may be harmful as the longer the drains are left, the higher is the risk of infection. In this light, several efforts have been made in an attempt to validate an accurate model for predicting PF after surgery especially for a better drain management.13,18–21 A prospective diagnostic study from Molinari et al13 analyzed 137 consecutive patients who underwent pancreatic resection (101 PD and 36 DP). In all the patients the levels of drain fluid amylase were determined starting from POD1 to POD5. The study demonstrated that only the level of amylase in drains on POD1 had a predictive value for the appearance of PF, and in particular when the level of amylase was greater than 5,000 U/L with a sensitivity of 93% for PD and 100% for DP. Moreover, for patients who presented amylase level greater than 5,000 U/L, only those with amylase levels greater than 200 U/L on POD5 developed complications. On the basis of these results, the same group designed a prospective randomized clinical trial demonstrating that, in patients with AVDs less than or equal to 5,000 U/L, early removal (POD3) was associated with a lower rate of PF and abdominal complications after standard pancreatic resections.14 In our experience, amylase level in drains was assessed on POD1 and POD5 when prophylactic drains were maintained. Essentially, we incorporated these measurements as part of our clinical pathway on the basis of the study by Molinari et al.13 Nevertheless, the management of drains was not always influenced by the amylase values as 69 patients with POD1 AVD less than or equal to 5,000 U/L maintained at least one of the drains until POD5. In this study, the cutoff value for POD1 Table 4 PDPJ PDDTM DP

proposed by Molinari et al13 resulted in a low sensitivity for the entire cohort. Consistent with these results, Kawai et al18 found that for POD1 AVD greater than 4,000 U/L, the sensitivity was only 62%. Moreover, several authors demonstrated a higher sensitivity for lower levels of POD1 AVD.19–21 In particular, Sutcliffe et al21 found a significantly lower cutoff of POD1 AVD (350 U/L) with a sensitivity of 100%. The cutoff proposed by Molinari et al13 was then considered not accurate to confidently exclude a PF. We also observed that different types of operations and/or reconstruction were associated with different median POD1 AVD. The consequence of this result is that the predictive value of POD1 AVD had to differ between different operations. We then found a cutoff level of amylase in drains with 100% sensitivity for different types of operations. As early drain removal could be associated with the development of abdominal collections, sensitivity was the most important parameter considered in the cutoff choice. On the other side, specificity can be increased only measuring also POD5 AVD. Also, Molinari et al13 reported different median values of AVD after DP (3,230 U/L) and PD with pancreatogastrostomy (2,700 U/L) compared with PDPJ (666 U/L). It is possible that the differences on POD1 AVD could also be explained by the wide heterogeneity of drain fluid composition. The analytical performance of methods for measuring amylase (including at least the dilution test and the recovery test) should be validated in a matrix comparable to pancreatic juice. As a consequence, it is difficult to compare amylase levels measurement in pancreatic fluids between different institutions. There are some limitations to our study. The

Positive and negative predictive value according to different types of operation POD1 AVD (U/L)

Sensitivity (%)

Specificity (%)

PPV (%)

NPV (%)

%1,000 %2,000 %2,000

96–100 100 97–100

0–74 0–85 0–37

66 43 26

96 100 97

AVD 5 amylase value in the drain; DP 5 distal pancreatectomy; NPV 5 negative predictive value; PDDTM 5 pancreatoduodenectomy with duct-tomucosa anastomosis; PDPJ 5 pancreatoduodenectomy with pancreatojejunostomy; POD 5 postoperative day; POD1 AVD 5 postoperative day 1 amylase value in drains; PPV 5 positive predictive value.

6

The American Journal of Surgery, Vol -, No -, - 2014

main limit of the cutoffs that we propose is the low specificity. This means that a significant portion of patients will have an uneventful postoperative course despite a high drain amylase level on POD1. In these patients, a further measurement of AVD on POD5 could be helpful for deciding about drain removal. The absence of a standardized protocol in drain management was possibly associated with a late drain removal in some patients. As a consequence, it is likely that some fistulas were the consequence of drain pressure gradient across the anastomosis. On the other hand, the strengths of this study include the standardization of surgical techniques as well as the large number of patients included in the analysis. In conclusion, an AVDbased model is helpful and accurate in predicting PF. The AVD cutoff differs significantly between different types of pancreatic resection and also between different types of anastomosis reconstruction. It is likely that the accurate AVD cutoff for predicting PF should be determined in each institution rather than be extrapolated by the current available reported experiences.

7. Schmidt CM, Choi J, Powell ES, et al. Pancreatic fistula following pancreaticoduodenectomy: clinical predictors and patient outcomes. HPB Surg 2009;2009:404520. 8. Muscari F, Suc B, Kirzin S, et al. Risk factors for mortality and intraabdominal complications after pancreatoduodenectomy: multivariate analysis in 300 patients. Surgery 2006;139:591–8. 9. Sierzega M, Niekowal B, Kulig J, et al. Nutritional status affects the rate of pancreatic fistula after distal pancreatectomy: a multivariate analysis of 132 patients. J Am Coll Surg 2007;205:52–9. 10. Fuks D, Piessen G, Huet E, et al. Life-threatening postoperative pancreatic fistula (grade C) after pancreaticoduodenectomy: incidence, prognosis, and risk factors. Am J Surg 2009;197:702–9. 11. Winter JM, Cameron JL, Yeo CJ, et al. Biochemical markers predict morbidity and mortality after pancreaticoduodenectomy. J Am Coll Surg 2007;204:1029–36; discussion, 1037–8. 12. Schnelldorfer T, Ware AL, Sarr MG, et al. Long-term survival after pancreatoduodenectomy for pancreatic adenocarcinoma: is cure possible? Ann Surg 2008;247:456–62. 13. Molinari E, Bassi C, Salvia R, et al. Amylase value in drains after pancreatic resection as predictive factor of postoperative pancreatic fistula: results of a prospective study in 137 patients. Ann Surg 2007;246:281–7. 14. Bassi C, Molinari E, Malleo G, et al. Early versus late drain removal after standard pancreatic resections: results of a prospective randomized trial. Ann Surg 2010;252:207–14. 15. Bassi C, Dervenis C, Butturini G, et al. Postoperative pancreatic fistula: an international study group (ISGPF) definition. Surgery 2005; 138:8–13. 16. Kawai M, Tani M, Terasawa H, et al. Early removal of prophylactic drains reduces the risk of intra-abdominal infections in patients with pancreatic head resection: prospective study for 104 consecutive patients. Ann Surg 2006;244:1–7. 17. Conlon KC, Labow D, Leung D, et al. Prospective randomized clinical trial of the value of intraperitoneal drainage after pancreatic resection. Ann Surg 2001;234:487–93; discussion, 493–4. 18. Kawai M, Tani M, Hirono S, et al. How do we predict the clinically relevant pancreatic fistula after pancreaticoduodenectomy?dan analysis in 244 consecutive patients. World J Surg 2009;33:2670–8. 19. Kong J, Gananadha S, Hugh TJ, et al. Pancreatoduodenectomy: role of drain fluid analysis in the management of pancreatic fistula. ANZ J Surg 2008;78:240–4. 20. Okano K, Kakinoki K, Suto H, et al. Persisting ratio of total amylase output in drain fluid can predict postoperative clinical pancreatic fistula. J Hepatobiliary Pancreat Sci; 2011 [Epub ahead of print]. 21. Sutcliffe RP, Battula N, Haque A, et al. Utility of drain fluid amylase measurement on the first postoperative day after pancreaticoduodenectomy. World J Surg 2012;36:879–83.

References 1. Yeo CJ, Cameron JL, Sohn TA, et al. Six hundred fifty consecutive pancreaticoduodenectomies in the 1990s: pathology, complications, and outcomes. Ann Surg 1997;226:248–57; discussion, 257–60. 2. Gouma DJ, van Geenen RC, van Gulik TM, et al. Rates of complications and death after pancreaticoduodenectomy: risk factors and the impact of hospital volume. Ann Surg 2000;232:786–95. 3. Cameron JL, Riall TS, Coleman J, et al. One thousand consecutive pancreaticoduodenectomies. Ann Surg 2006;244:10–5. 4. Pratt WB, Callery MP, Vollmer Jr CM. Risk prediction for development of pancreatic fistula using the ISGPF classification scheme. World J Surg 2008;32:419–28. 5. Tsiotos GG, Farnell MB, Sarr MG. Are the results of pancreatectomy for pancreatic cancer improving? World J Surg 1999;23:913–9. 6. Bartoli FG, Arnone GB, Ravera G, et al. Pancreatic fistula and relative mortality in malignant disease after pancreaticoduodenectomy. Review and statistical meta-analysis regarding 15 years of literature. Anticancer Res 1991;11:1831–48.