Clinicopathological differences and risk factors of early- versus late-onset pancreatic cancer: A systematic review and meta-analysis Authors: Mohammad Alzoubi, Rahaf Al Jibarat, Ahmad Omar Saleh, Reem Eisa, Haneen Mosa Al-sha’ar, Tuqa M Alfreijat, Juman Habboub, Rania Rashed, Rahaf Hanoun, Raghad Alkhateeb, Zaed Hamady Published: 2026 ABSTRACT SummaryPancreatic cancer is one of the most aggressive malignancies worldwide, and the incidence is increasing in patients under the age of 50. We systematically reviewed the clinicopathological features of early-onset pancreatic cancer (EOPC) and late-onset pancreatic cancer (LOPC) to help develop tailored therapeutic strategies and earlier diagnosis for younger patients. A total of 21 studies involving 1,813,719 patients were included, comprising 139,554 patients with EOPC and 1,674,165 with LOPC. EOPC was more commonly associated with tumors in the pancreatic tail, poor or undifferentiated histology, lymph node involvement, and abdominal pain at presentation, whereas LOPC more frequently involved tumors in the pancreatic head. No significant differences were observed between EOPC and LOPC in tumor resectability, surgical resection rates, or the prevalence of jaundice, cachexia, and diabetes mellitus. Overall, EOPC demonstrated a distinct clinicopathological phenotype characterized by more aggressive pathological features and advanced nodal disease, emphasizing the need for age-specific pancreatic cancer management.Keywords: pancreatic neoplasms, systematic review, early-onset, young Introduction Pancreatic cancer (PC) is one of the most aggressive and lethal malignancies globally; it is the fourth most common cause of cancer-related death in high-income countries,1 with a five-year survival rate of fewer than 9%. PC is projected to be the second cause of cancer death in 2030.2,3 PC is associated with family history, diabetes, smoking, certain genetic mutations such as CDKN2A and BRCA, pancreatitis, and obesity,4 and a common genetic variant has been identified as associated with PC.5 PC affects people in older age groups, and the median age at diagnosis is 71 years6 Of all PC cases, 8%–10% present in those less than 50 years of age.7 The majority of early-onset PC (EOPC) cases were diagnosed at an advanced stage.8 The prognosis of EOPC compared to late-onset pancreatic cancer (LOPC) remains under debate, with various studies showing contradictory findings.9,10,11,12,13,14 The cause of this early presentation of PC remains unknown. Early-stage PC is difficult to diagnose as it frequently exhibits vague symptoms, which results in a diagnosis at an advanced stage when curative treatment is no longer achievable.15 Studies investigating differences in clinicopathological features between patients with early-onset pancreatic ductal adenocarcinoma (PDAC) and those with average-onset PDAC (AOPC) have shown conflicting results. Some studies have shown that they share the same clinicopathological features.16 Others have shown that younger patients with PC have unique clinicopathological characteristics compared with older patients.17,18 Therefore, we conducted this meta-analysis to compare EOPC and LOPC according to their clinicopathological features. Results Search results and study selection The initial electronic search identified 6124 studies. After duplicates removal via EndNote (n = 218) and Rayyan (n = 223), we were left with a total of 5683 novel studies. Title and abstract screening excluded 5643 studies, leaving 40 studies for full-text review. Ultimately, this systematic review and meta-analysis includes 21 studies.8,11,12,14,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32 The process is outlined in the PRISMA flowchart diagram (Figure 1). PRISMA flow chart of the screening process Characteristics of included studies Our meta-analysis included 21 studies published between 2015 and 2024, comprising a total of 1813719 patients who met our inclusion criteria. Of the 1813719 patients, 139554 had EOPC, and 1674165 had LOPC. A summary of baseline characteristics of the included patients is presented in Table 1. Further study details can be found in Table S1. Table 1Baseline characteristics of the included studiesStudy IDEOPC cutoffStudy designCancer typeNo. of patients EOPC/LOPCGenderAgeEOPCLOPCMaleFemaleMaleFemaleEOPCLOPCNtala C 2018, RLH, UK19≤50 YearRetrospective analysisPDAC35/ 334251017915545.7 (33–50)166.2 (51–85)1Ohmoto A 2015, US16≤40 YearRetrospective analysisPDAC17/ 89111651138035.5264.852Javier E 2020, US11≤50 YearRetrospective analysisPDAC12,137/ 194,9256,9265,21197,52197,40444.5 ± 4.6370.2 ± 10.33Piciucchi M 2015, SAH, Italy14≤50 YearRetrospective analysisPDAC25/ 26817812814045 (43.5–46.6)470.2 (69.1–71.2)4Ramai D 2020, US18≤40 YearRetrospective analysisPAC1,181/134,91972445769,26465,65535 ± 3.9369 ± 11.23Rashad N 2024, Egypt20≤50 YearRetrospective analysisPAC84/328513320412443.6 ± 6.6364.7 ± 8.23Saadat L 2021, US21≤50 YearRetrospective analysisPancreatic Cancer15,710/232,9248,8746,836115,840117,084NANAvan Dongen J 2021, Netherlands22<60 YearRetrospective, population-based cohort studyPDAC1,551/8,7478487034,3574,39055 (51–57)574 (68–80)5Takeda T 2022, Japan12≤50 YearRetrospective analysisPDAC and Histologic Variants127/1,519775082569446 (22–50)668 (51–92)6Whitley A 2023, Czech Republic23≤50 YearRetrospective analysisPDAC1,324/17,5648534719,2928,27243.99 ± 4.89368.12 ± 9.393Rémond M 2024, France8≤50 YearRetrospective analysisPDAC113/765625139537046 (41–48)566 (59–72)5Ansari, D 2019, US24≤50 YearRetrospective analysisPDAC4,523/68,3832,6251,89835,02433,35946 (19–49)669 (50–103)6Beeghly-Fadiel, A 2016, VUMC,US25≤50 YearRetrospective analysisPDAC118/1,289714769859164.3 ± 10.63Bunduc, S 2018, Bucharest26≤45 YearRetrospective analysisPDAC24/1241685866NANACastet, F 2023, Spain27≤50 YearRetrospective analysisPDAC139/19773661029545.2 (41.3–48.5)574.4 (72–78)5Eguchi, H 2016, Japan17≤40 YearRetrospective analysisPDAC526/35,61934218421,42714,19234.1 ± 5.0365.6 ± 9.03Huang, B 2022, US28<55 YearRetrospective analysisPDAC80,908/632,71447,53333,375317,062315,652NANAJayakrishnan, A 2023, US29≤50 YearRetrospective analysisPAC20,822/339,94211,7939,029174,398165,54445 (42–48)570 (62–78)5Kang, J 2017, South Korea30≤45 YearRetrospective analysisPDAC34/6652410396269NANAMa, M 2022, China31≤50 YearRetrospective analysisPDAC44/477341027420346.2 ± 4.3364.4 ± 7.73Mendis, S 2024, Australia32≤50 YearProspective collected multi-site dataset analysisPDAC112/1,571545881275845 (20–50)772 (51–97)7Data presented as1 mean (range),2 median,3 mean ± SD,4 mean (95% CI),5 median (IQR),6 median (range),7 median (min-max). PDAC, pancreatic ductal adenocarcinoma; PAC, pancreatic adenocarcinoma; EOPC, early-onset pancreatic cancer; LOPC, late-onset pancreatic cancer. Baseline characteristics of the included studies Data presented as1 mean (range),2 median,3 mean ± SD,4 mean (95% CI),5 median (IQR),6 median (range),7 median (min-max). PDAC, pancreatic ductal adenocarcinoma; PAC, pancreatic adenocarcinoma; EOPC, early-onset pancreatic cancer; LOPC, late-onset pancreatic cancer. Quality assessment Most of the included studies have a low to some risk of bias according to the NIH Quality Assessment Tool for Observational Cohort and Cross-Sectional Studies. Of the 21 studies evaluated, nine were rated good, 11 fair, and only one poor. As shown in Table S2, a consistent limitation across all studies was the insufficient or poorly described sample size, and none employed multiple exposure measurements. As well, outcome assessor blinding was largely absent, with nearly all studies reporting “no” or “not reported” in this domain. Efficacy outcomes Location of tumor Head and uncinate process of the pancreas Compared to EOPC, LOPC was more frequently located in the head of the pancreas (odds ratio [OR], 0.96, 95% confidence interval [CI], [0.94; 0.97], p < 0.0001), prediction interval [PI], [0.94; 0.97]) with moderate heterogeneity (I2 = 24.8%). The distribution of tumors in the uncinate process did not differ significantly between LOPC and EOPC (OR, 0.88, 95% CI, [0.61; 1.26], p = 0.478) PI, [0.09; 8.95]), and no heterogeneity was observed (I2 = 0%). For pancreatic head tumors, no significant difference between subgroups was observed according to the EOPC age cutoff (p = 0.1580) (Figures 2A and 2B). The influence plot identified Huang (2022 US) as a highly influential study (Figure S1A). Excluding this study slightly altered the pooled effect size to OR, 0.94 (95% CI, [0.91; 0.96], p < 0.0001), with heterogeneity (I2) = 20%. The direction, magnitude, and statistical significance of the association remained consistent, indicating that the overall conclusions are robust and not driven by any single study. (Figure S3). Forest plot illustrating the comparison of tumor location between EOPC and LOPC (A) Pancreatic head, (B) uncinate process, and (C) pancreatic tail. ORs with 95% confidence intervals (CIs) are presented for each anatomical site, reflecting the relative probability of tumor occurrence in EOPC versus LOPC. An OR greater than 1 indicates a higher likelihood of EOPC, whereas an OR less than 1 suggests a lower likelihood compared to LOPC. Abbreviations: OR, odds ratio; CI, confidence interval. The tail part of the pancreas EOPC was significantly more frequently associated with tail tumors compared to LOPC (OR, 1.16, 95% CI, [1.13; 1.18], p < 0.0001) PI, [1.13; 1.18]), with moderate heterogeneity observed (I2 = 30.8%). No significant difference between subgroups was observed according to the EOPC age cutoff (p = 0.0888). (Figure 2C). The influence plot identified Huang (2022, US) as a highly influential study (Figure S1-B). Excluding this study resulted in a nonsignificant association (OR, 1.06, 95% CI, [0.97; 1.16], p = 0.208) and further reduced heterogeneity (I2 = 17.8%) (Figure S3). These results indicate that the pooled effect estimates for tail tumors are sensitive to the inclusion of Huang (2022, US) and should be interpreted with caution. Body part of the pancreas LOPC showed a non-significant trend toward a higher frequency of tumors located in the pancreatic body compared to EOPC (OR, 0.86, 95% CI, [0.73; 1.01], p =0.065; PI, [0.54; 1.37]) with observed significant heterogeneity (I2 = 75.8%). A significant difference between subgroups was identified based on the EOPC age cutoff (p < 0.0001). Using an EOPC age cutoff of ≤40 years (OR, 0.64, 95% CI, 0.53–0.76; p < 0.0001) and ≤45 years (OR, 0.25, 95% CI, 0.09–0.70), LOPC was significantly more frequently located in the pancreatic body. In contrast, with an EOPC age cutoff of <60 years (OR, 1.19, 95% CI, 1.02–1.38), LOPC was significantly less frequently located in the pancreatic body. No significant differences between EOPC and LOPC were observed when the EOPC age cutoff was ≤ 50 or <55 years (Figure 3A) The influence plot indicated that Eguchi (2016, Japan) was a highly influential study (Figure S1C). After excluding this study, the pooled effect size slightly shifted (OR, 0.98, 95% CI, [0.96; 1.01], p = 0.174), and heterogeneity dropped to I2 = 62.6% (Figure S3). Both the effect size and its significance remained consistent, supporting the robustness of the results independent of any single study. Forest plot of tumor distribution by pancreatic region in EOPC versus LOPC Forest plot comparing tumor distribution between EOPC and LOPC across additional pancreatic regions: (A) pancreatic body, (B) combined body and tail, and (C) entire pancreas or unspecified location. ORs with 95%CIs are reported for each category, representing the relative frequency of tumor sites in EOPC compared with LOPC. OR values above 1 indicate increased odds in EOPC, while values below 1 indicate decreased odds relative to LOPC. Abbreviations: OR, odds ratio; CI, confidence interval. Tail and body part of the pancreas No significant difference was observed in the distribution of tumors between the tail and body parts of the pancreas when comparing EOPC and LOPC (OR, 1.03, 95% CI, [0.99; 1.07], p = 0.207), with no heterogeneity observed (I2 = 0%). No significant difference between subgroups was observed according to the EOPC age cutoff (p = 0.9698) (Figure 3B). The influence plot indicated that Javier (2020, US) was a highly influential study (Figure S1D). Excluding this study altered the pooled effect size to (OR, 0.97, 95% CI, [0.74; 1.27], p = 0.809) without changing the heterogeneity (I2 = 0%) (Figure S3). Excluding this study did not substantially affect the effect estimates or heterogeneity, supporting the robustness of these findings. Whole pancreas or not otherwise specified EOPC demonstrated a non-significant trend toward more frequent involvement of whole pancreas or tumors not otherwise specified compared to LOPC (OR, 1.08, 95% CI, [0.99; 1.16], p = 0.068), with significant heterogeneity (I2 = 81.5%). A significant difference between subgroups was identified based on the EOPC age cutoff (p < 0.0001). Using an EOPC age cutoff of ≤ 40 years (OR, 1.27, 95% CI, [1.13; 1.44], p < 0.0001) and ≤ 50 years (OR, 1.08, 95% CI, [1.04; 1.14], p = 0.0005), EOPC was significantly more frequently affected by the whole pancreas or not specified. No significant differences between EOPC and LOPC were observed when the EOPC age cutoff was <60 or <55 years (Figure 3C) The influence plot identified Ramai (2020, US), Javier (2020, US), van (2021, Netherlands), and Whitley (2023, Czech Republic) as highly influential studies (Figure S2). Excluding these studies changed the pooled effect size to (OR, 0.97, 95% CI, [0.96; 0.99], p =0.002) and reduced heterogeneity to I2 = 73% (Figure S3). Thus, the overall effect estimates and heterogeneity were sensitive to the inclusion of these studies. Tumor resectability and surgical resection rate There was no significant difference between EOPC and LOPC in tumor resectability (OR, 0.99, 95% CI, [0.45; 2.17], p = 0.9725; PI, [0.00; 3556.25]), with substantial heterogeneity (I2 = 90.4) (Figure 4A). However, EOPC showed a non-significant trend toward a higher surgical resection rate compared with LOPC (1.25, 95% CI, [0.92; 1.71], p = 0.1565; PI, [0.51; 3.11]), with significant heterogeneity (I2 = 90.6%). Subgroup analysis demonstrated no significant difference according to the EOPC age cutoff (p = 0.3147) (Figure 4B). For the surgical resection rate, the influence plot identified Ntala (2018, RLH, UK) and Ma (2022, China) as highly influential studies (Figure S4). Excluding these studies changed the pooled effect size to (OR, 1.54, 95% CI, [1.36; 1.74], p < 0.0001) and the heterogeneity to (I2 = 90.7%) (Figure S5). Consequently, the pooled effect estimate shifted to demonstrate a significantly higher surgical resection rate in EOPC compared with LOPC. Heterogeneity remained largely unchanged after excluding these studies, indicating that the overall effect estimate was sensitive to its inclusion. Forest plot of tumor resectability and surgical resetion rates between EOPC and LOPC Forest plot comparing EOPC and LOPC in terms of (A) tumor resectability and (B) surgical resection rates. ORs with 95% CIsare shown for resectable, borderline resectable, and unresectable disease categories. ORs greater than 1 denote higher odds in EOPC, whereas ORs less than 1 indicate lower odds compared to LOPC. Abbreviations: OR, odds ratio; CI, confidence interval. Presenting symptoms Abdominal pain EOPC was significantly presented with abdominal pain compared to LOPC (OR, 1.72, 95% CI, [1.48; 1.99], p < 0.0001; PI, [1.40; 2.12]) with significant heterogeneity (I2 = 72.9). No significant difference between subgroups was observed according to the EOPC age cutoff (p = 0.1029). (Figure S6A) Influence diagnostics using the influence plot indicated that no individual study exerted a disproportionate influence on the pooled effect size (Figure S7A). Jaundice EOPC demonstrated a non-significant trend toward a lower frequency of presentation with jaundice compared with LOPC (OR, 0.84, 95% CI, [0.69; 1.01], p =0.063; PI, [0.64; 1.09]), with significant heterogeneity (I2 = 63.3%). No significant difference between subgroups was observed according to the EOPC age cutoff (p = 0.1223). (Figure S6B) Influence diagnostics using the influence plot indicated that no individual study exerted disproportionate influence on the pooled effect size (Figure S7B). Cachexia (weight loss) There were no significant differences between EOPC and LOPC in patients presenting with cachexia (OR, 0.85, 95% CI, [0.53, 1.37], p=0.5096; PI, [0.22; 3.32]), with substantial heterogeneity (I2= 75.6%). Subgroup analysis demonstrated a statistically significant difference according to the EOPC age cutoff (p = 0.0213). Using an EOPC age cutoff of ≤ 40 years (OR, 0.36, 95% CI, [0.19; 0.70]), EOPC was significantly less likely to present the patients with cachexia compared with LOPC. No significant differences between EOPC and LOPC were observed when the EOPC age cutoff was ≤ 45 or ≤ 50 years (Figure S6C) To further investigate this finding, an influence plot was generated, indicating that Ma (2022, China) was a highly influential study (Figure S7C). Consequently, when this study was excluded, the pooled effect size changed to (OR, 0.61, 95% CI, [0.45; 0.82]), with significantly increased cachexia associated with LOPC, and heterogeneity decreased to 57.3% (Figure S8). Thus, this suggests that Ma (2022, China) contributed substantially to between-study heterogeneity and affected the overall estimate. Consequently, when this study was excluded, the pooled effect size changed to (OR, 0.61, 95% CI, [0.45; 0.82]), with significantly increased cachexia associated with LOPC, and heterogeneity decreased to 57.3% (Figure S8). Thus, this suggests that Ma (2022, China) contributed substantially to between-study heterogeneity and affected the overall estimate. DM symptoms There were no significant differences between EOPC and LOPC in patients presenting with DM symptoms when the cutoff was set as ≤ 50 years (OR, 0.94, 95% CI, [0.23; 3.82], p =0.933; PI, [0.00; 222.26]) with substantial heterogeneity (I2 = 85.7%) (Figure S6-D). The influence plot indicated that Piciucchi (2015, SAH, Italy) and Ma (2022, China) were highly influential studies (Figure S7-D). When these two studies were excluded, the pooled effect size changed to (OR, 0.47, 95% CI, [0.11; 1.93]) (Figure S8). Excluding these studies did not substantially affect the effect estimates, supporting the robustness of these findings. Differentiation grade of the tumor Undifferentiated (grade IV) EOPC was significantly associated with an undifferentiated tumor compared to LOPC (OR, 1.25, 95% CI, [1.22; 1.28], p < 0.0001; PI, [1.22; 1.29]), with moderate heterogeneity (I2 = 20.2%). No significant difference between subgroups was observed according to the EOPC age cutoff (p = 0.2624). (Figure S9A) The influence plot identified Jayakrishnan (2023, US) as a highly influential study (Figure S10A). Excluding this study did not materially change the pooled effect (OR, 1.28, 95% CI, [1.22; 1.34]) and heterogeneity (I2 = 23%). Thus, the overall effect estimates and heterogeneity were robust to the exclusion of this study, supporting the stability of the findings (Figure S11). Poor differentiation (grade III) EOPC was significantly associated with poorly differentiated tumors compared to LOPC (OR, 1.13, 95% CI, [1.02; 1.25], p =0.021; PI, [0.85; 1.49]), with substantial heterogeneity (I2 = 89.2%). A significant difference between subgroups was identified based on the EOPC age cutoff (p < 0.0001). When the cutoff was set as < 55 years (OR, 1.30, 95% CI, [1.27; 1.32]) and < 60 years (OR, 1.57, 95% CI, [1.32; 1.88]), EOPC was significantly associated with poorly differentiated tumors compared to LOPC. No statistically significant differences between EOPC and LOPC were detected at the ≤ 45 and ≤ 50-year cutoffs. (Figure S9B) The influence plot indicated that no individual study exerted disproportionate influence on the pooled effect size (Figure S10-B). Moderate differentiation (grade II) EOPC was non-significantly more frequently associated with moderately differentiated tumors compared to LOPC (OR, 1.10, 95% CI, [0.97; 1.25], p = 0.125; PI, [0.75; 1.62]), with substantial heterogeneity (I2 = 93.9%). A significant difference between subgroups was identified based on the EOPC age cutoff (p < 0.0001). Using an EOPC age cutoff of < 55 years (OR, 1.26, 95% CI, [1.24; 1.29]) and < 60 (OR, 1.63, 95% CI, [1.39; 1.92]), EOPC was significantly more frequently associated with moderately differentiated tumors compared with LOPC. No significant differences between EOPC and LOPC were observed when the EOPC age cutoff was ≤ 40, ≤ 45, or ≤ 50 years (Figure S9C) The influence plot identified van Dongen (2021, Netherlands) as a highly influential study (Figure S10C). Excluding this study did not materially change the pooled effect (OR, 1.04, 95% CI, [0.94; 1.14]) and heterogeneity (I2 = 93.9%) (Figure S11). Thus, the overall effect estimates and heterogeneity were robust to the exclusion of this study, supporting the stability of the findings. Well differentiation (grade I) There was no significant difference between EOPC and LOPC in well differentiation frequency (OR, 1.01, 95% CI, [0.85; 1.19], p =0.913; PI, [0.62; 1.66]), with substantial heterogeneity (I2 = 84.9%). A significant difference between subgroups was identified based on the EOPC age cutoff (p = 0.0008). Using an EOPC age cutoff of < 55 years (OR, 1.13, 95% CI, [1.09; 1.17]) and < 60 (OR, 1.77, 95% CI, [1.31; 2.40]), EOPC was significantly more frequently associated with well-differentiated tumors compared with LOPC. No significant differences between EOPC and LOPC were observed when the EOPC age cutoff was ≤ 40, ≤ 45, or ≤ 50 years (Figure S9D). The influence plot identified Ansari (2019, US) and Van Dongen (2021, Netherlands) as highly influential studies (Figure S10D). Excluding these studies did not change the pooled effect to (OR, 1.05, 95% CI, [0.97; 1.14], p = 0.198), and heterogeneity was reduced to (I2 =47.7%) (Figure S11). The exclusion of these studies did not substantially affect the effect estimates or heterogeneity, supporting the robustness of these findings. Lymph nodes status The rate of lymph node positivity was 18.8% (5,554/29,554) in the EOPC group compared with 15% (89,755/598,387) in the LOPC group. Pooled analysis showed that EOPC was significantly more frequently associated with positive lymph nodes compared to LOPC (OR, 1.31, 95% CI, [1.27; 1.35], p < 0.0001; PI, [1.25; 1.38]), with moderate heterogeneity (I2 = 58.4%). No significant difference between subgroups was observed according to the EOPC age cutoff (p = 0.0589). (Figure S12A). Lymph node positivity occurred in 13.1% (3,859/29,554) of patients with EOPC and 14.5% (86,940/598,387) with LOPC. However, there was no significant difference between EOPC and LOPC in negative lymph node frequency (OR = 0.85, 95% CI, 0.50–1.44, p = 0.542; PI, [0.13; 5.71]) or unknown status (OR = 1.24, 95% CI, 0.84–1.83, p = 0.270; PI, [0.31; 4.96]). Substantial heterogeneity was observed (I2 = 98.7% and 98.7%, respectively). A significant difference between subgroups was identified based on the EOPC age cutoff for both negative lymph node (p < 0.0001) (Figure S12B) and unknown status (p = 0.0013) (Figure S12C). When the age threshold was defined as ≤50 years (OR, 1.42, 95% CI, 1.20–1.69; p < 0.0001), EOPC was significantly more likely to be associated with negative lymph node status. Conversely, using a cutoff of ≤40 years (OR, 0.53, 95% CI, 0.47–0.60; p = 0.0003), EOPC was significantly less likely to be associated with negative lymph node status. Risk factors Table 2 summarizes the pooled risk factors for early-onset EOPC versus late-onset LOPC. Patients with EOPC were more likely to be male, current smokers, and belong to non-Hispanic Black, Hispanic, non-Hispanic Asian/Pacific islander, or other racial/ethnic groups compared with those with LOPC. (Figure S13 and S14). Similarly, a history of previous radiotherapy was more frequently reported among patients with EOPC. (Figure S15) In contrast, never smokers, ex-smokers, and non-Hispanic White individuals were significantly more common in the LOPC group. (Figures 12A and 12B) Furthermore, a personal history of previous cancer and the diagnosis of multiple malignancies were more frequently observed among patients with LOPC. (Figure 1, Figure 2, Figure 3, Figure 4and S16) However, no significant differences were identified between EOPC and LOPC with respect to body mass index (BMI) (Figure S17), history of alcohol excess (Figure S13), genetic syndromes, chronic pancreatitis, family history of PC, family history of any cancer, or first-degree family history of cancer (Figure S16). Table 2Pooled effect estimates of risk factors in early-onset pancreatic cancer versus late-onset pancreatic cancerRisk factorNo. of studiesEOPC (events/total)LOPC (events/total)Pooled effect size (OR/MD)95% CIp valueI2 (%)95% Prediction intervalBMI5Total: 314Total: 3,2570.37[–0.03; 0.76]0.06690%[–0.19; 0.93]Gender, male2181,033/139,554848,837/1,674,1611.33[1.29; 1.38]<0.000182.2%[1.21; 1.47]History of alcohol excess896/ 1,0323,029/ 40,1051.47[0.79; 2.70]0.220884.4%[0.20; 10.56]Current smoker11272/ 1,2464,210/42,0891.62[1.26; 2.10]0.000270.6%[0.78;3.38]Never smoke8359/1,1539,434/ 41,2240.61[0.52; 0.71]< 0.000166%[0.50; 0.73]Ex-smoker359/ 7001,790/36,1500.720.52; 0.9910.04500%[0.35; 1.46]Non-Hispanic Black721,456/130,911175,740/1,537,0471.51[1.48; 1.56]<0.000155.9%[1.45; 1.59]Non-Hispanic White789,452/130,9111,203,431/1,537,0470.60[0.58; 0.61]<0.000181.8%[0.56; 0.64]Hispanic413,613/129,57792,744/1,400,5051.75[1.55; 1.98]<0.000197.3%[1.14; 2.69]Non-Hispanic Asian/Pacific Islander53,983/118,65634,732/1,340,8331.26[1.16; 1.36]< 0.000185.7%[1.03; 1.53]Any other group61,986/129,73017,063/1,402,1281.54[1.38; 1.71]< 0.000175.8%[1.16; 2.03]Genetic syndrome32/18718/2,1211.15[0.26; 5.06]0.84900%,[0.05; 29.58]History of chronic pancreatitis523/7371,070/37,8600.83[0.53; 1.30]0.41572.7%[O.44; 1.56]Family history of pancreatic cancer961/1,0001,912/40,8451.58[0.74; 3.39]0.236168.9%[O.15; 17.23]Family history of any cancer5114/2881,736/3,3400.91[0.68; 1.22.0.534531.2%[0.60; 1.38]First degree family history of cancer488/300292/1,1270.97[0.69; 1.35]0.85410%[0.56; 1.66]Personal history of previous cancer4148/1,8342,499/11,3540.30[0.26; 0.36]< 0.00019.5%[0.23; 0.40]History of diabetes798/5751,664/5,0910.46[0.18; 1.17]0.104591.6%[0.02; 11.01]Previous history of cholecystectomy213/60114/6021.17[0.61; 2.24]0.64540%,[0.02; 80.16]Diagnosis of a multiple malignancy211/6441,330/36,9080.31[0.17; 0.57]0.00010%[0.01; 15.70]History of previous radiotherapy2254/1,21623,196/135,2531.31[1.14; 1.50]0.00020%[0.53; 3.23]Abbreviations, EOPC, early-onset pancreatic cancer; LOPC, late-onset pancreatic cancer; BMI, body mass index; OR, odds ratio; MD, mean difference; CI, confidence interval; I2, heterogeneity statistic. Pooled effect estimates of risk factors in early-onset pancreatic cancer versus late-onset pancreatic cancer Abbreviations, EOPC, early-onset pancreatic cancer; LOPC, late-onset pancreatic cancer; BMI, body mass index; OR, odds ratio; MD, mean difference; CI, confidence interval; I2, heterogeneity statistic. Discussion We conducted this analysis to characterize the clinicopathological features of EOPC, driven by its rising incidence and significant societal burden.33,34,35,36,37,38 A precise understanding of how EOPC differs from LOPC is consequential to tailoring therapeutic strategies and developing early diagnostic pathways for younger patients. While previous reviews have focused on survival and risk factors, a comprehensive synthesis of its clinical and pathological presentation has been lacking. EOPC patients are more often male, current smokers, of non-White ethnicity, with a history of prior radiotherapy. LOPC is associated with never, or ex-smoking, status and a history of prior or multiple malignancies. Notably, no significant differences were observed regarding BMI, genetic syndromes, family cancer history, alcohol excess, or chronic pancreatitis. When characterizing anatomical distribution, a clear distinction must be made between absolute tumor frequency and relative age-based associations. While the pancreatic head is the most common primary tumor site for EOPC patients,39,40,41,42 our meta-analysis demonstrates that older age is even more strongly associated with this location (OR, 0.96, 95% CI, [0.94; 0.97]). More importantly, this relative head-preference in LOPC remains significant across all age-cutoff subgroups and after the exclusion of the highly influential Huang dataset. These findings challenge earlier discordant reports of a head predilection favoring EOPC over LOPC16,38 and highlight the impact of age definitions on study outcomes. For instance, the association between younger age and head tumors reported by Artinyan et al. illustrates how broad thresholds (≤66 years) misattribute the head-predominant distribution of LOPC to early-onset disease.43 Conversely, smaller studies with strictly defined EOPC (<40 years), such as Lüttges et al., found no such association, aligning with our sensitivity-adjusted results.44 Beyond the pancreatic head, our pooled analysis initially suggested a significant relative association between EOPC and tail tumors (OR, 1.16, 95% CI, [1.13; 1.18]). However, sensitivity analysis established that this was entirely contingent on a single, highly influential registry study (Huang 2022). Upon its exclusion, the association lost significance (OR, 1.06, 95% CI, [0.97; 1.16]) and heterogeneity was substantially reduced, directly challenging the concept of a definitive tail predilection in younger patients. Analyses of other sites further highlight the impact of study influence and age definitions. For body tumors, a relative LOPC association emerged only at narrower cutoffs (≤40 or ≤45 years) and reversed when a broader one was applied (<60 years). Similarly, the trend of whole-pancreas involvement in EOPC was highly sensitive to a cluster of influential studies, while no discernible age-based differences were found for uncinate or combined body/tail tumors. Ultimately, these outcomes confirm a strong relative head-predilection among older cohorts, while establishing that apparent age-specific associations at other anatomical sites remain highly susceptible to registry bias and cohort selection. Regarding tumor resectability, no significant difference was observed between EOPC and LOPC. However, Rashad (2024, Egypt) showed a significantly higher rate of tumor resectability in EOPC.20 In contrast, Takeda (2022, Japan) showed a lower rate of tumor resectability in EOPC.12 This exhibits substantial heterogeneity in tumor resectability. The observed heterogeneity in tumor resectability is likely driven by the geographical location of included patients and differences in the definitions used. Rashad et al. used a local, multidisciplinary team-based approach, classifying patients as resectable, borderline resectable, or metastatic according to institutional policies, which introduces potential subjectivity and variability.20 In contrast, Takeda et al. applied detailed, guideline-based criteria that defined resectable, borderline resectable, and locally advanced PC based on precise tumor-vessel relationships, providing a more objective and reproducible assessment.12 These discrepancies in classification likely contributed to the opposing effect directions in the individual studies and the high overall heterogeneity. Initial analysis of the surgical resection rate showed no statistically significant difference between EOPC and LOPC. However, Ntala (2018, RLH, UK) and Ma (2022, China) were identified as outlier and influential analysis studies, and the exclusion of both studies revealed a significant association, with EOPC patients demonstrating a higher surgical resection rate compared to LOPC.19,31 This shift suggests that the original non-significant finding was likely influenced by outlier studies with distinct clinical populations and methodological limitations, rather than reflecting a true absence of association. The excluded studies differed significantly from the rest of the included research. Both were small, single-center retrospective studies with a limited sample size and possible selection bias.19,31 The other focused only on patients with both PC and type 2 diabetes, which does not reflect the broader PC population.31 Clinically, it makes sense that surgical resection is more likely in EOPC. Younger patients usually have better overall health, fewer other illnesses, and can better handle major pancreatic surgery, so they are more often considered for aggressive surgical treatment. Clinicians may also choose more aggressive treatments for younger people because they have a longer life expectancy and may benefit more from surgery that aims to cure. Several factors contributed to the differences seen in the surgical resection studies. First, the data sources varied. Large registry-based studies included most patients, while smaller single-center studies offered more detailed but less widely applicable information. Second, the criteria for selecting patients were not consistent. Some studies looked at all PC patients, while others focused only on those who could have surgery. For instance, some analyses excluded patients with advanced disease or those who did not undergo surgery, while others included all patients. Third, the definition of early-onset disease varied across studies, with some using an age cutoff of 45 years or younger and others using 50 years or younger. A lower age limit identifies a smaller and possibly more distinct group, while a broader definition encompasses a more diverse population. Although subgroup analysis did not reveal a significant difference between age cutoffs, these differences likely contribute to the inconsistency between studies. Analysis of secondary endpoints revealed abdominal pain as the most significant symptom of EOPC (OR, 1.72, 95% CI, [1.48; 1.99]). This strong association persisted across influence diagnostics and aligned with multiple cohorts such as Jiang, Raissouni, and Tingstedt that report abdominal pain in 62%, 87.5%, and 91% of younger cohorts, respectively.40,45,46 He et al. noted a similar trend (younger 32% versus. older 22%).47 Conversely, LOPC displayed a non-significant trend toward higher jaundice presentation rates (OR, 0.84, 95% CI, [0.69; 1.01]). Given LOPC’s established head predominance, this trend is topographically expected, as head tumors frequently cause early biliary obstruction.48,49 In contrast, EOPC’s relative lack of early-warning signs inherently delays diagnosis, allowing progression to severe pain driven by perineural invasion and retroperitoneal metastasis.50,51 In terms of systemic symptoms, diabetes mellitus and cachexia rates initially showed no significant overall differences between EOPC and LOPC. However, subgroup analyses indicated that the youngest subgroup (≤40 years) was significantly less likely to present with cachexia (OR, 0.36, 95% CI, [0.19; 0.70]). Furthermore, omitting one highly heterogeneous dataset (Ma 2022) shifted the pooled estimate to reveal significantly increased cachexia in LOPC (OR, 0.61, 95% CI, [0.45; 0.82]). This suggests that weight loss is more characteristic of LOPC, likely exacerbated by age-related frailty or the higher incidence of obstructive head tumors. Nonetheless, the sensitivity of this finding warrants careful interpretation. Pathologically, EOPC exhibits aggressive features, most notably a consistent association with high-grade histology. This is evidenced by a significant link to undifferentiated tumors across all age subgroups (OR, 1.25), as well as poorly differentiated tumors (OR, 1.13), specifically when broader age definitions (<55, <60 years) are applied. In contrast, initial estimates found no significant differences regarding moderately or well-differentiated tumors, with significant associations only emerging at these broader cutoffs (<55, <60). EOPC pathology, therefore, follows an age gradient: undifferentiated tumors hallmark the very young, as Bergmann et al. demonstrated, while broader definitions produce a histological profile increasingly similar to that of older populations.52 The aggressive biological phenotype of EOPC is characterized by higher rates of lymph node involvement and distant metastasis at diagnosis, along with a lack of significant differences in perineural or vascular invasion between age groups. He et al. similarly found comparable rates of microvascular invasion (32% in young versus 42% in old, p=0.1) and perineural invasion (64% versus 73%, p=0.09).47 Despite non-significant nodal differences in that specific study, heightened propensity for nodal involvement remains consistently documented39,40,42,53,54 and is supported by our finding of a significant association between EOPC and positive nodal status (OR, 1.31, 95% CI, [1.27; 1.35]). Moreover, node-negative disease was significantly less common in the youngest EOPC cohort (≤40 years) (OR, 0.53, 95% CI, [0.47; 0.60]). Ultimately, the aggressive phenotype of EOPC appears to favor systemic spread rather than isolated local invasion. The Lüttges et al. series, which reports comparable pathological features across age groups and no distinctive abdominal pain, represents an outlier in this regard, an outcome likely attributable to its limited sample size and consequent lack of statistical power.44 EOPC exhibits a distinct clinicopathological profile characterized by abdominal pain at presentation, poorer differentiation, and a greater likelihood of lymph node positivity. This aggressive pathology highlights the need for high clinical suspicion and biologically informed management. Future studies must adopt stratified age groupings to enable personalized therapeutic strategies. Limitations of the study First, the absence of a standardized EOPC definition contributed to statistical heterogeneity and complicated cross-study interpretations.55 Second, most included studies were retrospective and registry-based, with inherent risks of coding inaccuracies, limited symptom documentation, and limited detail on pathological and staging variables. Third, surgical cohorts from tertiary centers are prone to selection bias toward resectable disease and younger, fitter patients, complicating generalizability. Additionally, it is difficult to interpret tumor resectability and surgical resection rate outcomes because studies often use different definitions. Few studies use standardized radiologic or multidisciplinary criteria to define resectability. Most use surgical resection as a substitute, which may not accurately reflect the true extent of tumor resectability. Surgical resection rates also depend on the selection of patients, the operating procedures of each institution, and the availability of surgery to patients. These factors can introduce selection and treatment bias. A major problem in current research on tumor resectability is the lack of standardized, widely accepted criteria for determining resectability. Without these, it is hard to compare results across studies or combine data for analysis. Nonetheless, this study has notable strengths. We provide one of the most comprehensive analyses of EOPC clinicopathological features to date, incorporating a large and diverse pooled population. Our systematic use of age subgrouping, sensitivity analyses, and influence diagnostics provides a framework for understanding conflicting results in prior literature, distinguishing significant associations from those that are mostly driven by individual datasets, broad age definitions, or measurement bias. Implications and future research Unexplained abdominal pain in adults under 45 should raise clinical suspicion and warrant timely investigation. Given the high frequency of nodal spread, meticulous staging, including comprehensive lymph node assessment, is especially critical. Future research must move beyond a binary EOPC/LOPC dichotomy. We strongly recommend granular age stratifications (e.g., <40, 40–49, 50–59) and uniform reporting of tumor location, staging, and symptomatology. Prospective studies incorporating molecular profiling should prioritize these cohorts to determine correlating biological drivers. In this broader context, supported by external literature, genetic counseling and testing should be offered to patients under the age of 45. Understanding the molecular basis of poor differentiation and early metastatic spread may enable novel targeted treatments for this growing population. In addition to tumor resectability, future investigations should adopt guideline-consistent definitions to ensure and improve comparability. Resource availability Lead contact Further information and requests for resources should be directed to and will be fulfilled by the lead contact, Ahmad Omar Saleh (ahmadomar17713@gmail.com). Materials availability This study is a meta-analysis and does not involve the use of reagents or the generation of new materials. Data and code availability •All data analyzed in this study were extracted from previously published articles and are included within the main manuscript and supplementary materials. The extracted datasets are provided as an Excel file, while the statistical outputs supporting the findings are available in the main text and supplementary materials. All data analyzed in this study were extracted from previously published articles and are included within the main manuscript and supplementary materials. The extracted datasets are provided as an Excel file, while the statistical outputs supporting the findings are available in the main text and supplementary materials. •The statistical code used for data synthesis and meta-analysis is available from the corresponding author upon reasonable request. No custom software or proprietary algorithms were developed in this study. The statistical code used for data synthesis and meta-analysis is available from the corresponding author upon reasonable request. No custom software or proprietary algorithms were developed in this study. •This study did not generate new biological materials, reagents, or experimental models. Additional study materials and supplementary information are available within the article and its supplementary files. This study did not generate new biological materials, reagents, or experimental models. Additional study materials and supplementary information are available within the article and its supplementary files. Author contributions Conceptualization, M.A. and R.A.; methodology, A.O.S.; software, T.M.A., H.M.A.-s.’a., and A.O.S.; validation, M.A.; formal analysis, A.O.S.; investigation, H.M.A.-s.’a., T.M.A., J.H., R.R., R.H., and R.Alk.; data curation, A.O.S. and R.A.; resources, R.A. and A.O.S.; writing – original draft, R.E., R.A., and A.O.S.; writing – review and editing, Z.H. and M.A.; visualization, A.O.S. and R.A.; supervision, M.A.; project administration, R.A. Declaration of interests The authors declare no competing interests. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERSoftware and algorithmsSoftware: EndNote X9Clarivate Analyticshttps://endnote.com/downloadsSoftware: R-StudioR Foundation for Statistical Computinghttps://www.R-project.org/RayyanQatar Computing Research Institutehttps://www.rayyan.aiDeposited dataExtracted study datasetThis paperTable S4Database: PubMedPublished literaturehttps://www.ncbi.nlm.nih.gov/pubmedDatabase: ScopusPublished literaturehttps://www.scopus.com/pages/home#basicDatabase: Cochrane LibraryPublished literaturehttps://www.cochranelibrary.comDatabase: EmbasePublished literaturehttps://www.embase.comDatabase: Web of SciencePublished literaturehttps://www.webofscience.comData and Code AvailabilityThis studyNo new data or code generated Experimental model and study participant details Not applicable. This study is a systematic review and meta-analysis and does not involve human participants, animals, or in vitro experimental models. Method details Protocol registration This systematic review and meta-analysis were conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines.56 The study protocol was pre -registered with PROSPERO (CRD420251146596). Data sources & search strategy A comprehensive literature search was conducted in PubMed, Scopus, Web of Science (WOS), and the Cochrane Central Register of Controlled Trials (CENTRAL) from inception to August 2025, while Embase was searched from inception to November 2024. The search was conducted without applying any filters or restrictions to include all published articles up to that date. Keywords and MeSH terms were applied, including: “Early -onset,” “Early onset,” “Early -onset pancreatic cancer,” “EOPC,” “Late-onset,” “Late onset,” “Late-onset pancreatic cancer,” “LOPC,” and “Pancreatic Neoplasms.” We customized the search strategy to meet the requirements of various databases. The full search strategies for each are presented in Table S3. Study selection and eligibility criteria Eligible studies described the clinical and pathological features of patients with pancreatic neoplasms, comparing EOPC and LOPC. The primary outcome was tumor location (head, uncinate process, body, and tail). Secondary outcomes included tumor resectability and surgical resection rate. Because reporting varied across studies, these were analyzed as separate outcomes. True tumor resectability was defined as the proportion of patients considered resectable or borderline resectable based on radiologic or multidisciplinary assessment according to each study’s criteria. Surgical resection rate was defined as the proportion of all patients who underwent curative-intent surgery (R0/R1), regardless of their initial radiologic resectability. Additional secondary outcomes included presenting symptoms (jaundice, cachexia or weight loss, abdominal pain, or incidental detection after diabetes mellitus diagnosis), differentiation grade, perineural invasion, perivascular invasion, and lymph node status. All missing data were recorded as reported by each study. We included observational retrospective studies. While case reports, case series, conference abstracts, reviews, editorials, studies lacking relevant outcomes, and non-human studies were excluded. Two independent reviewers (R.K. and J.H.) screened all titles and abstracts using Rayyan.ai57 and reviewed the full texts. Disagreements on the selection process were resolved by consulting a third reviewer (R.A.). Data collection process Two authors (T.A. and H.A.) independently extracted the data from all the relevant included studies. Any discrepancies that arose during data collection between the two authors were discussed and resolved in consensus with a third author (R.A.). All included papers were accessed online via open or institutional access; no authors were contacted. Data were extracted from the main manuscripts, tables, graphs, and the authors’ supplementary materials into a uniform data extraction Excel sheet. The extracted data included a summary of each study: study ID, year of publication, country of origin, study design, and pancreatic cancer type. Baseline characteristics of participants were also recorded, including age, sex, the number of patients with EOPC and LOPC, and the total sample size. The outcome data were extracted into event and total counts for each group. Quality assessment Risk of bias in the included studies was assessed independently by two authors (T.A. and H.A.) using the NIH Quality Assessment Tool for Cross-sectional and Cohort Studies,58 which consists of 14 questions. The final assessment was based on responses to these questions and the authors’ judgment. Disagreements between reviewers were resolved by discussion and consensus with another author (A.O.S). Quantification and statistical analysis Meta-analysis was conducted using R-Studio.59 Odds ratios (ORs), their corresponding 95% confidence intervals (CIs), and prediction intervals (PIs) were estimated using a generalized linear mixed model (GLMM). When studies reported variability measures other than standard deviation (e.g., 95% confidence intervals, median with range, or interquartile range), the corresponding mean and standard deviation were estimated using established formulas as recommended in the Cochrane Handbook for Systematic Reviews of Interventions and Wan et al.‘s formula.60 Subgroup analysis was conducted using different diagnostic age cutoffs for early-onset pancreatic cancer (EOPC): ≤40 years, ≤45 years, ≤50 years, <55 years, and <60 years. Influence diagnostics assessed how individual studies affected the pooled effect size. Influence plots using Cook’s distance, DFBETAS, and hat values identified studies with disproportionate influence. Any study flagged as potentially influential was examined further by re-running the meta-analysis with the study excluded. We compared changes in pooled effect estimates and heterogeneity statistics (I2, τ2, and the Q statistic) to determine whether the results remained robust after removing influential studies. Heterogeneity was assessed using the I2 statistic and reported P value.61,62 The random effects model was used in all analyses, regardless of heterogeneity, as recent evidence suggests it provides a more robust outcome than the alternative fixed-effects models.63 Published: August 21, 2026 Footnotes Supplemental information can be found online at https://doi.org/10.1016/j.isci.2026.116643. Supplemental information Document S1. Figures S1–S18, Tables S1–S3, and Data S1, and Supplemental References Table S4. Extracted clinicopathological and outcome data from studies included in the meta-analysis References Associated Data Supplementary Materials Document S1. Figures S1–S18, Tables S1–S3, and Data S1, and Supplemental References Table S4. Extracted clinicopathological and outcome data from studies included in the meta-analysis Data Availability Statement •All data analyzed in this study were extracted from previously published articles and are included within the main manuscript and supplementary materials. The extracted datasets are provided as an Excel file, while the statistical outputs supporting the findings are available in the main text and supplementary materials. All data analyzed in this study were extracted from previously published articles and are included within the main manuscript and supplementary materials. The extracted datasets are provided as an Excel file, while the statistical outputs supporting the findings are available in the main text and supplementary materials. •The statistical code used for data synthesis and meta-analysis is available from the corresponding author upon reasonable request. No custom software or proprietary algorithms were developed in this study. The statistical code used for data synthesis and meta-analysis is available from the corresponding author upon reasonable request. No custom software or proprietary algorithms were developed in this study. •This study did not generate new biological materials, reagents, or experimental models. Additional study materials and supplementary information are available within the article and its supplementary files. This study did not generate new biological materials, reagents, or experimental models. Additional study materials and supplementary information are available within the article and its supplementary files.