Preoperative malignancy risk assessment in pancreatic cystic neoplasms using clinical and laboratory parameters

Abstract

BACKGROUND

Pancreatic cystic neoplasms (PCNs) are increasingly detected due to advancements in radiographic techniques, with a prevalence of approximately 15% in the general population. These lesions range from benign to premalignant and malignant, posing a diagnostic challenge. Accurate differentiation is critical, as premalignant and malignant PCNs often require surgical intervention, while benign cysts may only need monitoring unless symptomatic. Current diagnostic methods, including cross-sectional imaging, endoscopic ultrasonography, and endoscopic ultrasonography-guided fine-needle aspiration/biopsy, are specialized, not universally available, and have variable accuracy. Clinical and laboratory parameters such as carbohydrate antigen 19-9 (CA 19-9), neutrophil-lymphocyte ratio, platelet-lymphocyte ratio, and red cell distribution width (RDW) have been associated with malignancy risk, though only CA 19-9 is guideline-supported.

AIM

To assess the malignancy risk of PCNs using preoperative clinical and routine laboratory parameters.

METHODS

A retrospective cohort study analyzed 70 patients who underwent surgery for PCNs at Ankara Bilkent City Hospital between February 2019 and March 2023. Patients were categorized into group A (benign or low-grade dysplasia, n = 40) and group B (malignancy or high-grade dysplasia, n = 30) based on postoperative pathology. Preoperative demographic and laboratory parameters, including age, RDW, albumin, and CA 19-9, were compared. Univariate and multivariate logistic regression analyses identified independent predictors of malignancy. Receiver operating characteristic curve analysis evaluated predictive performance, with internal validation using bootstrapping.

RESULTS

Group B patients were older (69.86 ± 9.58 years vs 52.74 ± 16.85 years, P < 0.001) and had a higher incidence of diabetes mellitus (57.1% vs 21.4%, P = 0.002). RDW (16.2% vs 13.7%, P < 0.001), platelet-lymphocyte ratio (178 vs 126, P = 0.008), and CA 19-9 (21.7 U/mL vs 9.3 U/mL, P = 0.009) were significantly higher in group B, while albumin was lower (41 g/L vs 45 g/L, P = 0.008). Multivariate analysis identified age [odds ratio = 1.067, 95% confidence interval (CI): 1.014-1.122, P = 0.012] and RDW (odds ratio = 1.784, 95%CI: 1.172-2.715, P = 0.007) as independent predictors. The area under the curve for age, RDW, and their combination was 0.798 (95%CI: 0.695-0.900), 0.801 (95%CI: 0.692-0.911), and 0.858 (95%CI: 0.771-0.944), respectively, with bootstrapped validation confirming stability. Cut-off values of age ≥ 60 years and RDW ≥ 15.5% balanced sensitivity and specificity, increasing malignancy risk 15.3-fold and 22.6-fold, respectively.

CONCLUSION

Age and RDW are independent predictors of malignancy in PCNs, aiding in patient selection for advanced diagnostics and surgery. Larger, multicenter studies are needed to validate these findings.

Key Words: Pancreatic cystic neoplasms; Malignancy risk; Red cell distribution width; Age; Preoperative assessment

Core Tip: This study highlights the utility of simple preoperative parameters - age and red cell distribution width - in predicting malignancy in pancreatic cystic neoplasms. With area under the curve values of 0.798 and 0.801, respectively, and a combined area under the curve of 0.858, these markers can guide patient selection for endoscopic ultrasonography and cytological evaluation, potentially improving surgical decision-making for pancreatic cystic neoplasms.

INTRODUCTION

Pancreatic cystic neoplasms (PCNs) are increasingly detected due to advancements in radiographic techniques, with a prevalence of approximately 15% in the general population[1,2]. These lesions pose a diagnostic challenge due to their heterogeneous nature, ranging from benign to premalignant and malignant[3]. Accurate differentiation is critical, as premalignant and malignant PCNs often require surgical intervention, while benign cysts may only need monitoring unless symptomatic[4]. Current diagnostic approaches rely on cross-sectional imaging (computed tomography, magnetic resonance imaging), endoscopic ultrasonography (EUS), and EUS-guided fine-needle aspiration (FNA)/biopsy[5-7]. However, these methods are specialized, not universally available, and have variable diagnostic accuracy[8,9].

Clinical and laboratory parameters, such as elevated carbohydrate antigen 19-9 (CA 19-9), neutrophil-lymphocyte ratio (NLR), platelet-lymphocyte ratio (PLR), jaundice, and new-onset diabetes mellitus, have been associated with malignancy risk in PCNs[10]. For instance, PLR has been linked to prognosis in various malignancies, including pancreatic cancer, due to its reflection of systemic inflammation[11,12]. However, only CA 19-9 is supported by guideline-level evidence. Recent studies suggest that simple biochemical markers, such as red cell distribution width (RDW), may aid in preoperative risk stratification[13]. This study aims to evaluate the role of clinical and routine laboratory parameters in assessing the malignancy risk of PCNs, focusing on their potential to guide advanced diagnostics and surgical decisions.

MATERIALS AND METHODS

Study design and patient selection

This retrospective cohort study was conducted at Ankara Bilkent City Hospital, Department of General Surgery, from February 2019 to March 2023. The study population included 94 patients who underwent surgery for PCNs during this period. After excluding 24 patients who underwent diagnostic EUS ± FNA and were not recommended for surgery, 70 patients were included. Patients were divided into group A (benign or low-grade dysplasia, n = 40) and group B (malignancy or high-grade dysplasia, n = 30) based on postoperative pathology. Inclusion criteria encompassed cysts in all pancreatic regions (head, body, tail). Exclusion criteria included patients under 18 years (due to differing disease patterns and rarity of PCNs in this age group), those with hematological, autoimmune, or rheumatological diseases, recent blood transfusions (to avoid RDW distortion), or a prior malignancy diagnosis (to exclude confounding factors). This study was approved by the Ethics Committee of Ankara Bilkent City Hospital (Approval No. E-1-23-3616).

Data collection

Preoperative demographic data (age, gender, diabetes mellitus, smoking, comorbidities such as coronary artery disease, hypertension, and others) and laboratory parameters (RDW, NLR, lymphocyte-monocyte ratio, PLR, mean platelet volume, albumin, CA 19-9) were collected from electronic medical records. Imaging characteristics, including cyst size, main pancreatic duct diameter, and presence of mural nodules, were recorded. Pathological findings were obtained from postoperative histopathological reports. Data were retrospectively extracted by trained researchers from the hospital’s electronic database, ensuring consistency through standardized record-keeping protocols.

Statistical analysis

Statistical analyses were performed using IBM SPSS 26.0 (IBM Corp., Armonk, NY, United States). Normality was assessed with the Shapiro-Wilk test. Categorical variables were analyzed using the Pearson χ² test (for expected frequencies > 5) or Fisher’s exact test (for expected frequencies ≤ 5), while continuous variables were compared with the student’s t-test (for normally distributed data) or Mann-Whitney U test (for non-normally distributed data). Univariate logistic regression identified parameters associated with malignancy or high-grade dysplasia at P < 0.1, which were then included in a forward stepwise multivariate logistic regression to determine independent predictors, with variance inflation factor < 5 to assess multicollinearity. Odds ratio (OR) with 95% confidence interval (CI) was calculated. Receiver operating characteristic curve analysis evaluated the predictive performance of independent predictors, with area under the curve (AUC) values and 95%CI reported. Internal validation was performed using 1000 bootstrap resamples to assess model stability. Optimal cut-off values were determined using Youden’s method to balance sensitivity and specificity, and sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated. A P-value < 0.05 was considered statistically significant. The Hosmer-Lemeshow test confirmed model calibration (P = 0.742).

RESULTS

Baseline characteristics

Of the 70 patients, 40 (57.2%) were in group A and 30 (42.8%) in group B. Group B patients were significantly older (69.86 ± 9.58 years vs 52.74 ± 16.85 years, P < 0.001) and had a higher incidence of diabetes mellitus (57.1% vs 21.4%, P = 0.002). Gender distribution showed no significant difference (P = 0.058), with 72.5% females in group A and 45.4% in group B. Group B exhibited higher median RDW (16.2% vs 13.7%, P < 0.001), PLR (178 vs 126, P = 0.008), and CA 19-9 (21.7 U/mL vs 9.3 U/mL, P = 0.009), and lower albumin (41 g/L vs 45 g/L, P = 0.008). Cyst size and lymphadenopathy did not differ between groups, but main pancreatic duct diameter (9.4 mm vs 6.9 mm, P = 0.044) and mural nodule presence (80% vs 5%, P = 0.012) were significantly higher in group B (Table 1).

Table 1 Baseline characteristics of patients with pancreatic cystic neoplasms, n (%)/median (minimum-maximum).

Parameters	Overall (n = 70)	Group A (n = 40)	Group B (n = 30)	P value2
Age, years (mean ± SD)	59.59 ± 16.61	52.74 ± 16.85	69.86 ± 9.58	< 0.001
Gender				0.058
Female	42 (60.0)	29 (72.5)	13 (45.4)
Male	28 (40.0)	11 (27.5)	17 (54.6)
Cigarette smoking	7 (10.0)	2 (4.8)	5 (17.9)	0.107
Presence of chronic pancreatitis	18 (25.7)	9 (21.4)	9 (32.1)	0.315
Major comorbidity
Coronary artery disease	11 (15.7)	5 (11.9)	6 (21.4)	0.328
Hypertension	24 (34.3)	14 (33.3)	10 (35.7)	0.837
Diabetes mellitus	25 (35.7)	9 (21.4)	16 (57.1)	0.002
Other	9 (12.9)	3 (7.1)	6 (21.4)	0.142
Localization of tumor
Head	39 (55.7)	17 (42.5)	22 (73.3)	0.031
Body	21 (30.0)	16 (40)	5 (16.6)	0.070
Tail	9 (12.9)	6 (15)	3 (10.1)
Extensive	1 (1.4)	1 (2.5)	0
Length of tumor (long-axis, mm)	35 (25-55)	30 (20-60)	40 (25-50)	0.838
Mural nodule (preoperative screening)	26 (37.1)	2 (5)	24 (80)	0.012
Mural nodule diameter (mm)	7.8 (5-12)	5.5 (5-6)	8.8 (5-12)	0.035
Pancreatic duct diameter (mm)	8.8 (< 5-15)	6.9 (< 5-11)	9.4 (< 5-15)	0.044
Pathological diagnosis
Serous neoplasms	9 (12.9)	9 (22.5)	0
Mucinous cystic neoplasm	11 (15.7)	9 (22.5)	2 (6.7)
IPMN	16 (22.9)	22 (55)	26 (86.6)
Other1	2 (2.8)	0	2 (6.7)
Laboratory findings, median (IQR)
Red cell distribution width (%)	14.2 (13.5-16)	13.7 (13.3-14.6)	16.2 (14.1-17.3)	< 0.001
Neutrophil-lymphocyte ratio	2.47 (1.95-3.78)	2.33 (1.90-3.53)	2.56 (2.14-4.60)	0.096
Lymphocyte-monocyte ratio	4.01 (2.73-6.50)	4.23 (2.81-6.61)	3.46 (2.50-4.55)	0.112
Platelet-lymphocyte ratio	145 (107-194)	126 (106-157)	178 (118-238)	0.008
Mean platelet volume (10-15L)	8.5 (7.8-9.2)	8.3 (7.9-8.9)	8.7 (7.7-9.3)	0.272
Albumin (g/L)	43 (40-46)	45 (41-46)	41 (34-45)	0.008
CA 19-9 (U/mL)	13.6 (5.2-95)	9.3 (3.2-37.5)	21.7 (12-211.7)	0.009

¹Includes solid pseudopapillary neoplasm (1) and neuroendocrine tumor (1).

²P value for age derived from student’s t-test. P value for gender, presence of chronic pancreatitis, major comorbidities (coronary artery disease, hypertension, diabetes mellitus, other), and localization of tumor derived from Pearson χ² test. P values for cigarette smoking and mural nodule presence derived from Fisher’s exact test. P value for length of tumor, mural nodule diameter, pancreatic duct diameter, and laboratory findings derived from Mann-Whitney U test. P value not calculated for pathological diagnosis due to descriptive nature.

IPMN: Intraductal papillary mucinous neoplasm; IQR: Interquartile range; CA 19-9: Carbohydrate antigen 19-9.

Pathological findings

In group B, 26 (86.6%) patients had malignancies arising from intraductal papillary mucinous neoplasms (IPMNs), 2 (6.7%) had high-grade dysplasia in mucinous cysts, 1 (3.35%) had high-grade dysplasia in a solid pseudopapillary cyst, and 1 (3.35%) had a neuroendocrine tumor.

Regression analysis

Univariate analysis identified age, diabetes mellitus, RDW, PLR, albumin, and CA 19-9 as predictors of malignancy (P < 0.1). Multivariate analysis confirmed age (OR = 1.067, 95%CI: 1.014-1.122, P = 0.012) and RDW (OR = 1.784, 95%CI: 1.172-2.715, P = 0.007) as independent predictors (Table 2). Cut-off values of age ≥ 60 years and RDW ≥ 15.5% were selected to balance sensitivity and specificity, yielding sensitivities of 83.3% and 66.7%, specificities of 70.0% and 82.5%, PPVs of 65.6% and 73.9%, and NPVs of 85.7% and 77.4%, respectively. These thresholds increased malignancy risk 15.3-fold (95%CI: 3.8-61.9) and 22.6-fold (95%CI: 5.2-97.8), respectively (Table 3).

Table 2 Univariate and multivariate regression analyses to predict malignancy or high-grade dysplasia.

Parameters	Univariate analysis OR (95%CI)	P value	Multivariate analysis OR (95%CI)	P value
Age	1.095 (1.042-1.150)	< 0.001	1.067 (1.014-1.122)	0.012
Male gender	2.574 (0.957-6.925)	0.061
Diabetes mellitus	4.889 (1.710-13.977)	0.003
Red cell distribution width	2.240 (1.485-3.380)	< 0.001	1.784 (1.172-2.715)	0.007
Neutrophil-lymphocyte ratio	1.145 (0.927-1.414)	0.208
Platelet-lymphocyte ratio	1.005 (0.999-1.012)	0.092
Albumin	0.914 (0.844-0.989)	0.026
CA 19-9	1.003 (0.999-1.007)	0.152

Variance inflation factor for all variables < 5, indicating no significant multicollinearity. CA 19-9: Carbohydrate antigen 19-9; OR: Odds ratio; CI: Confidence interval.

Table 3 Predictive values of independent predictors for malignancy or high-grade dysplasia at optimal cut-off values, n (%)/%.

Parameters	Cut-off value1	Number2	Sensitivity	Specificity	PPV	NPV	OR (95%CI)
Age (years)	≥ 60	35 (50.0)	83.3	70.0	65.6	85.7	15.3 (3.8-61.9)
Red cell distribution width (%)	≥ 15.5	25 (35.7)	66.7	82.5	73.9	77.4	22.6 (5.2-97.8)

¹Cut-off values selected using Youden’s method to balance sensitivity and specificity.

²Number of patients above the given cut-off values.

PPV: Positive predictive value; NPV: Negative predictive value; OR: Odds ratio; CI: Confidence interval.

Receiver operating characteristic curve analysis

The AUC values for age, RDW, and their combination were 0.798 (95%CI: 0.695-0.900), 0.801 (95%CI: 0.692-0.911), and 0.858 (95%CI: 0.771-0.944), respectively (Figure 1). Bootstrapped AUCs confirmed model stability (mean AUC: 0.855, 95%CI: 0.768-0.942).

Figure 1 Receiver operating characteristic curve analysis. RDW: Red cell distribution width.

DISCUSSION

Accurate preoperative assessment of PCNs remains a significant challenge, despite advances in imaging and endoscopic techniques[8,9]. Established imaging markers such as cyst size, main pancreatic duct dilation, and mural nodules are critical for malignancy risk stratification[5-7]. Clinical parameters like elevated CA 19-9, jaundice, new-onset diabetes mellitus, and PLR have also been linked to malignancy[10]. Our study identifies age and RDW as independent predictors of malignancy in PCNs, with AUC values of 0.798 and 0.801, respectively, and a combined AUC of 0.858, indicating strong discriminatory power for identifying high-risk cases.

The association of RDW with malignancy aligns with previous studies suggesting its role as an inflammatory marker influenced by erythropoietin inhibition and oxidative stress[12]. Hu et al[13] and Yüksel et al[14] reported RDW’s prognostic value in gastric and pancreatic cancers, with elevated levels (cut-off approximately 15%-16%) correlating with poor outcomes. Dang et al[15] further demonstrated RDW’s prognostic significance in patients undergoing pancreatic surgery, linking it to poor immunonutritional status, with a cut-off of 14.8% yielding 68% sensitivity and 72% specificity. Our finding of RDW ≥ 15.5% as a predictor with a 73.9% PPV is consistent with these reports, extending RDW’s utility to preoperative PCN risk assessment. However, unlike studies emphasizing NLR or PLR[10], our analysis found RDW superior, possibly due to its sensitivity to broader inflammatory and oxidative pathways. Celli et al[10] reported NLR’s utility in IPMN malignancy with a sensitivity of 56%-73.1% and specificity of 58%-87% (cut-off approximately 2.5-3.0), but our RDW sensitivity (66.7%) and specificity (82.5%) suggest a more balanced predictive role in PCNs.

Age ≥ 60 years as a predictor is consistent with epidemiological data showing increased pancreatic cancer risk with age, attributed to cumulative exposure to risk factors such as smoking, diabetes, and alcohol, as well as age-related epigenetic changes[11,16]. Our study’s NPV of 85.7% for patients under 60 years supports a conservative approach, aligning with Gonda et al[17], who noted age as a risk modifier in PCN progression (cut-off approximately 55-60 years). This finding contrasts with studies focusing solely on imaging markers[18], highlighting the additive value of clinical parameters. Sun et al[18] emphasized imaging features (e.g., main duct diameter > 10 mm, large mural nodules) without integrating laboratory markers like RDW, achieving high specificity but lacking the non-invasive simplicity of our model.

Compared to existing literature, our study offers a distinct perspective by integrating simple clinical and laboratory markers with imaging. Brugge et al[8] and Yao et al[19] underscored EUS ± FNA’s diagnostic accuracy (25.5%-70%), but noted operator dependency and complication risks, which our age-RDW model could help mitigate by preselecting candidates. Wei et al[20] highlighted the limitations of computed tomography/magnetic resonance imaging alone, supporting our approach of combining age and RDW with imaging to enhance diagnostic precision. Our model’s high NPV and PPV could reduce unnecessary EUS procedures, addressing the accessibility and cost barriers noted by Taya et al[9].

Our findings differ from molecular approaches exploring Kirsten Rat Sarcoma viral oncogene homolog and guanine nucleotide-binding protein alpha-stimulating activity polypeptide mutations in intraductal papillary mucinous neoplasms, which are often multifocal and multiclonal[21]. While molecular analysis can improve cyst classification, its routine clinical application remains limited due to cost and availability, unlike the widely accessible RDW and age parameters we propose. The study, single-center design and small sample size (n = 70) limit generalizability, introducing potential selection bias by including only surgically managed patients. This contrasts with multicenter studies like Ohtsuka et al[5], which provide broader guidelines. Future research should validate our findings across diverse populations, explore RDW’s mechanistic role (e.g., via inflammatory cytokines), and conduct prospective studies with larger cohorts. A nomogram integrating age and RDW could enhance translational impact, though this requires further validation.

CONCLUSION

Age and RDW are independent predictors of malignancy in PCNs, offering a simple, non-invasive approach to risk stratification. These parameters can guide patient selection for advanced diagnostics and surgical intervention, potentially improving outcomes. Larger, multicenter studies are needed to validate these findings and elucidate RDW’s underlying mechanisms in PCN malignancy.