We retrospectively collected the data for IPMN cases registered between January 1994 and September 2008. Eleven tertiary medical centers with specialized biliary-pancreatic surgeons, gastrointestinal radiologists, and gastrointestinal pathologists participated in this study.

We designed standardized case-report forms containing fields for IPMN clinical, radiological, and pathological characteristics and survival data. A standard protocol for interpreting radiological and pathological data was made by experienced radiologists and pathologists at Seoul National University Hospital.

For standardized radiological information, images from cross-sectional studies (contrast-enhanced computed tomography [CT] and magnetic resonance imaging [MRI]) and pancreatography (magnetic resonance cholangiopancreatography [MRCP] or endoscopic retrograde cholangiopancreatography [ERCP]) were used to classify the type of IPMN. Standardized criteria included IPMN morphology (main, branch, or mixed), anatomic location of the main lesion, distribution, multifocality, presence of mural nodule, lesion size, septation, wall thickening, calcification, and parenchymal atrophy. Analyses of these factors were based mainly on CT, and information from other imaging methods was used to support the information gained from CT.

Branch duct type IPMN was defined as a cystic dilatation of a branch pancreatic duct or a pancreatic cyst communicating with the pancreatic duct without main duct dilatation. The term 'without main duct dilatation' indicated a maximal main pancreatic duct diameter of 5 mm or less. This was thought to be the most restricted and well-represented criteria for branch duct type IPMN (3, 5, 10-12). Main duct type IPMN was radiologically defined as a main pancreatic duct dilated > 5 mm. Mixed type IPMN was defined as radiologic evidence of both main and branch duct type IPMN (3, 5, 10-12). Wall thickening was considered to be present when the wall was ≥ 2 mm in maximum thickness and covered at least one-third of the cystic lesion or the dilated pancreatic duct. Pancreatic parenchymal atrophy was defined as a situation in which the ratio of the diameter of the main pancreatic duct to the width of the pancreatic parenchyma at the same location was greater than 0.5 (12). A mural nodule was defined as a soft tissue nodule attached to the lesion wall, or as a septum in the pancreatic cystic lesion or the dilated main pancreatic duct. Mural nodule status only was applied in this study. Mural nodule size was not investigated, because CT is not the gold standard for measuring mural nodule size (12).

During the above-mentioned period, a total of 388 patients preoperatively diagnosed with IPMN and cystic dilatation of a branch duct underwent surgical resection. Pathological confirmation was performed in all 11 medical centers. Patients with main ducts dilated to larger than 5 mm (n = 151) were excluded from this study. Thus, a total of 237 patients were eligible for this study. We analyzed the medical records, results of preoperative laboratory and radiologic examination and postoperative pathologic examination, and follow-up data.

The study subjects consisted of 137 men and 100 women of mean age 63.1 yr (range, 38-83 yr). There were 148 (62.4%) symptomatic patients. Frequently manifested symptoms and signs included abdominal pain, weight loss, jaundice, diabetes mellitus, and pancreatitis. Elevated serum carcinoembryonic antigen (CEA, > 5 ng/mL) was identified in 9 (3.8%) patients, and elevated serum carbohydrate antigen 19-9 (CA 19-9, > 37 U/mL) was detected in 31 (13.1%) patients. The mean cystic tumor size was 28.1 ± 18.0 mm. Thirty-one (13.1%) patients had mural nodules on CT (Table 1). Combined extrapancreatic malignancies were detected in 65 (27.4%) patients. Stomach cancer was the most common malignancy (26 patients, 11.0%), followed by colorectal cancer (14 patients, 5.9%), other hepatobiliary-pancreatic cancer (12 patients, 5.1%), and urologic cancer (5 patients, 2.1%). Among the various operative procedures, pancreaticoduodenectomy (including pylorus-preserving procedures) was most common (108 [45.6%] patients). For limited resection, duodenum-preserving resection of the head of the pancreas, pancreatic head resection with segmental duodenectomy, median pancreatectomy, and excision were performed in 4 (1.7%), 3 (1.3%), 12 (5.1%), and 14 (5.9%) patients, respectively. Pathologic examination according to the AJCC 6th edition manual showed malignant IPMN (carcinoma in situ or invasive IPMC) in 39 (16.5%) patients and invasive IPMN in 26 (11.0%) patients. Extrapancreatic tumor extension (pT3) was identified in 8 (30.8%) patients with invasive IPMN, and metastatic lymph nodes were identified in 3 (11.5%) patients with invasive IPMN (n = 26) (Table 2).

Differences between categorical variables were analyzed using the chi-squared test and Fisher's exact test, while the Student t-test and Mann-Whitney U test were used for comparisons among continuous variables. The statistical correlation between outcomes and categorical/continuous factors was determined using logistic regression analysis. Statistical significance was defined as P < 0.05 (two-sided P values).

Parameters identified by univariate analysis with P < 0.05 were entered into multivariate analysis to identify independent malignant and invasive predicting factors.

Receiver-operator characteristic (ROC) curves were used to determine optimal score cutoff levels for the prediction of malignant and invasive IPMN. All analyses were carried out using SPSS version 15.0 for Windows (SPSS, Chicago, IL, USA).

This study was approved by institutional review board of the Seoul National University Hospital (IRB No. H-1007-125-324). The board waived the requirement for informed consent.

The subjects were divided into two groups: nonmalignant and malignant IPMN. The nonmalignant IPMN group (n = 198) included patients with adenomas and borderline tumors, and the malignant IPMN group (n = 39) included patients with carcinoma in situ and invasive IPMC. On univariate analysis, elevated CA 19-9 (P = 0.002), larger cyst size (P < 0.001), the presence of mural nodules (P < 0.001), wall thickening (P = 0.036), calcification (P = 0.041), and parenchymal atrophy (P = 0.037) were statistically significant (Table 3). When multivariate analysis was performed for these factors, larger cyst size (P = 0.021) and the presence of mural nodules (P = 0.001) remained significant (Table 4).

Adenomas, borderline tumors, and carcinoma in situ were classified as noninvasive IPMN. Diabetes mellitus (P = 0.036), elevated serum CEA (P = 0.027) or CA 19-9 (P < 0.001), larger cyst size (P = 0.001), presence of mural nodules (P < 0.001), wall thickening (P = 0.005), and septation (P = 0.024) were significant variables on univariate analysis (Table 5). On multivariate analysis, elevated CEA (P = 0.043), larger cyst size (P = 0.034), and mural nodules (P < 0.001) were statistically significant (Table 6).

The scoring rule was developed using a logistic regression model with the following form:

Logit (Y = 1) = ln (P(Y = 1)/[1-P(Y = 1)]) = b₀ + b₁X₁ + b₂X₂ + ... + b_nX_n,

where Y = 1 indicates malignant or invasive IPMN, and ln is the natural logarithm. The formula shows that the predictors (X₁-X_n) are linear, and on an additive scale related to the logit (Y = 1), with relative weights b₁-b₂; that is, the regression coefficients. This means that, for a particular subject with a certain X₁-X_n profile, the regression coefficients multiplied by the patient's X values may simply be added to obtain the logit and, by logistic transformation, the probability of Y = 1. Commonly, however, the regression coefficients are multiplied by a number (for instance, 10) and rounded to the nearest value to easily obtain applicable scores per predictor (13).

First, this formula was applied to malignancy-predicting factors. The regression coefficient of cyst size was 0.045, and that of mural nodule presence was 2.237. These coefficients were multiplied by 10 and rounded off. Then the formula was constructed as follows:

Malignancy-predicting score = 22.4 (presence of mural nodule [0 or 1]) + 0.5 (size of cyst [mm]).

The malignancy-predicting scores were calculated on whole subjects based on this scoring formula, and optimal cutoff values were determined using a receiver-operator characteristic (ROC) curve (Fig. 1). The optimal cutoff value was determined to be 14, and subjects were divided into two groups: 14 points or more, and less than 14 points. Significant differences (P < 0.001) were identified between the two groups, with sensitivity of 71.8%, specificity of 60.0%, and accuracy of 61.6%.

These results indicated that branch duct type IPMN with mural nodule should be regarded as malignant IPMN itself. If preoperative CT does not identify mural nodule, cyst size greater than 28 mm should be regarded as indicating malignant IPMN.

Invasiveness-predicting scores were obtained in the same way. The regression coefficient of elevated CEA was 3.216, that of cyst size was 0.061, and that of mural nodule presence was 3.660. The scoring formula was as follows:

Invasiveness- predicting score = 36.6 (presence of mural nodule [0 or 1]) + 32.2 (elevated serum concentration of CEA [0 or 1]) + 0.6 (size of cyst [mm]).

After calculating the invasive-predicting scores, the best cutoff value was determined to be 21 points by ROC curve. Based on a 21-point cutoff, the two groups were shown to be significantly different (P < 0.001) with sensitivity of 76.9%, specificity of 72.5%, and accuracy of 73.1% (Fig. 2). These results indicated that patients with mural nodule or elevated serum CEA should be considered to have invasive IPMN. If the patient had no evidence of mural nodules or elevated serum CEA, cyst size greater than 35 mm was regarded as indicating the presence of invasive IPMN.