Salivary metabolites for pancreatic cancer diagnosis: a systematic review and meta-analysis Authors: Maryam Koopaie, Mahnaz Fatahzadeh, Sajad Kolahdooz Published: 2025 ABSTRACT AbstractBackgroundPancreatic cancer (PC), a malignancy with poor prognosis, demands innovative diagnostic strategies for early detection. Current methods, such as endoscopic ultrasound and blood-based biomarkers, face limitations in accessibility, invasiveness, and accuracy, particularly for early-stage disease. Saliva, a non-invasive biofluid enriched with metabolites and proteins, offers promise as a diagnostic medium.MethodsA systematic search was conducted across databases, including MEDLINE (PubMed), Scopus, Web of Science, LIVIVO, Embase, Cochrane Library, Ovid, and the Google Scholar search engine. The search had no restrictions on language or publication date. The database search was performed in September 2024 and updated in December 2024 using PRISMA-DTA guidelines. Methodological quality was assessed using Quality Assessment Tool for Diagnostic Accuracy Studies-2 (QUADAS-2). Statistical analysis was performed, and pooled sensitivity, specificity, and diagnostic odds ratio (DOR) calculated.ResultsThis systematic review and meta-analysis include five studies published between 2010 and 2024. 7,799 records were screened, yielding five studies (4,328 subjects: 854 patients with PC, 3,474 controls) for inclusion in our analysis. Pooled sensitivity and specificity for salivary metabolites were 0.784 (95% CI: 0.769–0.797) and 0.793 (95% CI: 0.764–0.819), respectively, with a diagnostic odds ratio (DOR) of 15.79 (95% CI: 12.61–19.76) and an AUC of 0.867. Multi-metabolite panels significantly outperformed single biomarkers (DOR: 101.68, 95% CI: 24.79-417.08 vs. 14.29, 95% CI: 11.59-17.62), underscoring the value of combinatorial approaches. Subgroup analyses revealed pathway-specific variations, with nervous system-related pathways showing the highest DOR (24.58, 95% CI: 12.41-48.69) and polyamine pathways the lowest (8.96, 95% CI: 5.36-14.98).ConclusionThis study highlights saliva’s potential as a complementary tool in PC diagnosis and in bridging the gap between experimental biomarkers and clinical utility.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12876-025-04483-w.Keywords: Pancreatic neoplasms, Metabolite, Protein, Saliva, Diagnosis, Biomarker, Systematic review, Meta-analysis Background Pancreatic cancer (PC) is a devastating malignancy known for its aggressive progression, poor prognosis, and alarmingly low survival rate, which makes it one of the most challenging cancers in modern oncology [1]. PC progresses asymptomatically in its early stages, leading to late diagnosis and advanced disease at presentation [2]. Although PC ranks 12th in terms of incidence globally, it ranks 6th in mortality, highlighting its high lethality and the critical importance of early detection in reducing its global mortality rate [3]. Pancreatic cancer remains one of the deadliest malignancies, with a five-year survival rate below 5% for metastatic disease and around 13% across all stages in the United States population [4]. This bleak outlook emphasizes the necessity for innovative diagnostic strategies to detect the PC at earlier, more treatable stages [5]. Conventional diagnostic approaches, including imaging modalities such as endoscopic ultrasound, are effective but often costly, invasive, and inaccessible to many populations [6]. Additionally, commonly used blood-based tumor markers like Carbohydrate antigen 19-9 (CA19-9) have limitations, particularly in detecting early-stage disease or distinguishing PC from benign conditions such as chronic pancreatitis (CP) [7, 8]. These limitations have spurred interest in developing alternative, non-invasive diagnostic tools that are reliable, cost-effective, and easily deployable in diverse healthcare settings. In recent years, saliva has gained considerable attention as a promising diagnostic medium for various diseases, including cancer [9]. Saliva is a readily accessible, non-invasive biological fluid containing a rich repertoire of metabolites, proteins, DNA, RNA, and other biomolecules that reflect systemic and local physiological and pathological processes [10]. In contrast to blood sampling, saliva collection offers a non-invasive approach that does not necessitate trained individuals, making it advantageous for large-scale population screening initiatives [11]. Furthermore, saliva exhibits stability when stored under standard conditions, enhancing its applicability in diagnostic processes, particularly in environments with limited resources [12]. Emerging research has highlighted the potential of salivary biomarkers for early PC detection [13]. The diagnostic significance of genetic and microbial biomarkers in the saliva of patients with PC has been examined [14, 15]. Salivary diagnostics have evolved beyond oral diseases to become a transformative liquid biopsy for systemic cancers [16], including anatomically distant malignancies [17] such as pancreatic [18], gastric [19], lung [20], and breast carcinomas [21, 22]. Particularly relevant to PC, where late diagnosis remains a significant barrier, studies have shown that salivary metabolites have potential in discriminating between cases and healthy controls [23]. Saliva as a compliant, scalable diagnostic medium that meets criteria for minimally invasive precision oncology, particularly in resource-limited settings where endoscopic surveillance remains impractical [24]. The interaction effects of genetic, microbial, protein-based, and metabolomics biomarkers in PC reflect the multifaceted nature of this disease [25–27]. Metabolomics biomarkers capture downstream effects, such as changes in protein expression and metabolic reprogramming [28, 29]. PC alters glycolytic pathways, lipids, and amino acids, releasing specific metabolites into the bloodstream, which are transferred to saliva through salivary glands [30, 31]. Besides, systemic inflammation caused by cancer oxidative stress also triggers inflammatory metabolites, and oxidative stress markers, among others, appear in saliva [32]. Changes due to the malignancy within the oral microbiome also contribute to variations in salivary metabolite levels [33, 34]. Integrating biomarker types and developments in metabolomics and proteomics have improved the reliability of saliva as a promising diagnostic medium [35]. The pancreas and salivary glands share notable anatomical and functional similarities, including the presence of acini, ductal systems, hormonal and neural regulation, and secretory cells [36–38]. These similarities extend to the secretion of proteins and metabolites, highlighting their potential utility in disease diagnosis [39, 40]. Additionally, shared diseases between the pancreas and salivary glands underscore the close relationship between these two organs in physiological and pathological processes [41]. A saliva-based diagnostic tool would not only improve accessibility and compliance but also reduce the need for invasive and costly procedures, enabling the identification of PC at earlier stages and potentially improving survival outcomes. Despite these advances, significant barriers remain to the clinical implementation of salivary biomarkers. Variability in study designs, patient populations, analytical techniques, and validation protocols has led to inconsistent findings. Additionally, the lack of robust evidence synthesis has hindered the translation of promising biomarkers into clinically actionable diagnostic tools. While individual studies provide valuable insights, their limited scope and variability often prevent a comprehensive evaluation of their clinical utility. Therefore, a systematic review of salivary metabolomics biomarkers for PC diagnosis is long overdue, and to our knowledge, no diagnostic systematic review has yet been undertaken on this subject. This systematic review and meta-analysis aim to synthesize the available evidence on salivary biomarkers for diagnosis of PC and bridge the gap between experimental findings and clinical practice by comprehensively. The objectives are to identify the most robust and reproducible salivary biomarkers and assess their diagnostic performance through pooled sensitivity, specificity, and odds ratios. Methods Protocol and registration The PRISMA-DTA methodology [42] (Supplementary File 1) and followed the methodological guidance provided in the Cochrane Handbook [43] were utilized to conduct this systematic review and meta-analysis. The details of this process have been documented on the PROSPERO website (CRD420251022734), https://www.crd.york.ac.uk/PROSPERO/view/CRD420251022734. Research design and strategy for literature search Keywords were derived from key concepts: pancreatic neoplasms, metabolites, proteins, saliva, and the mouth. We searched for relevant keywords across several academic databases, including MEDLINE, Scopus, Web of Science, LIVIVO, Embase, Cochrane Library, Ovid, and Google Scholar search engines. The search strategy for the PubMed database was developed using Medical Subject Headings (MeSH) terminology. The database search, which had no restrictions on language or publication date, was conducted in September 2024 and updated in December 2024. Supplementary File 2 provides the specific search strategy used. The studies retrieved from the search were incorporated into EndNote X 8.1 for managing references. Duplicate records were identified and excluded during this phase of the process (Fig. 1, PRISMA flow diagram based on Page et al. [44]). Flow diagram based on the PRISMA framework, showing the procedure of record identification, evaluation, and inclusion Inclusion and exclusion criteria Studies that met the inclusion criteria were systematically reviewed. The studies focused on identifying and analyzing salivary metabolites and proteins of patients with PC. This systematic review excluded evidence sources such as in vivo and in vitro research, literature reviews (including systematic review and meta-analyses), clinical trials, case reports and series, along with editorials and letters. The reasons for excluding studies from the systematic review were insufficient evidence to use pathology as the diagnostic gold standard of PC, and lack of sufficient data to calculate diagnostic statistics. The meta-analysis involved exclusively those records that met the inclusion criteria and reported data on true positive (TP), false negative (FN), true negative (TN), and false positive (FP). Selection and screening of studies A two-stage screening process was used to identify eligible records. Two authors independently assessed the titles and abstracts of retrieved articles. Disagreements were resolved through discussion with a third party to reach consensus. Subsequently, individual authors evaluated the full text, and only those deemed relevant by authors were considered in the systematic review. Any remaining inconsistencies were resolved through comprehensive discussion among the authors until a consensus was reached. Data extraction Relevant data, including authors’ names, year of publication, country, of, sampling method, control group, type of biomarker, type of saliva, method for biomarker detection, statistical model, TP, FP, TN, and FN, for each individuals study were extracted from each eligible study. In cases where the confusion matrix was not directly reported within the paper, the data necessary for its construction were gleaned through digitizer software, enabling extracting relevant information from the source material (GetData Graph Digitizer 2.24). The computation of sensitivity and specificity was performed via Youden’s index [45]. Research quality Quality Assessment Tool for Diagnostic Accuracy Studies-2 (QUADAS-2) was used to assess the quality of the diagnostic accuracy studies included in this review [46]. To ensure inter-rater reliability, two authors independently evaluated the quality of the records and subsequently discussed their assigned scores for eligible studies to reach an agreement. Data and statistical analysis Diagnostic test characteristics and diagnostic odds ratio (DOR) were computed with the split component synthesis method, with 95% confidence interval (95% CI). The I-squared (I2) statistic was also employed to quantify potential heterogeneity across studies [47]. STATA V17.0 (Stata Corp), R package (V4.2.0), Meta-DiSc 1.4 (Madrid, Spain), MetaDTA: Diagnostic Test Accuracy Meta-Analysis v2.1.1 (https://crsu.shinyapps.io/MetaDTA/) [48], and GraphPad Prism 9.5.0 (GraphPad Software) were applied in the analytical process. To compare the effects of control group type (healthy vs. non-cancerous individuals) with those of PC patients, subgroup analyses were performed. A meta-analysis based on the random-effects model was conducted to combine the study results and derive overall estimates of diagnostic test characteristics and DOR estimates. The analysis involved plotting the area under the curve (AUC) and the summary receiver operating characteristic (SROC) curve. In this meta-analysis, we used the likelihood matrix of meta-analysis to synthesize the findings from individual studies and summarize the ability of a test to revise the prior probability of disease. Deeks' funnel plot test was performed to evaluate publication bias among the eligible studies. Clinical utility was quantified using standardized net benefit (sNB); (range: 0 (no utility) to 1 (maximum utility)). sNB curves were plotted against disease threshold probabilities to compare biomarker performance with CA19-9. Threshold probabilities (0–100%) were used to model clinical scenarios with justified interventions. Results Study selection A detailed flow diagram outlining the selection process of studies is presented in Fig. 1. The initial comprehensive search of the databases resulted in 7799 studies. Duplicate entries (n=3497) and non-related studies (n=2388) were identified and removed. A further 1735 records were excluded after screening of their abstracts. Full texts of the remaining 179 publications were assessed, and 174 studies were excluded (wrong publication type (n = 54), wrong population (n = 46), wrong in material and methods (n=45), wrong objective (n=22), wrong outcome (n=9)) at this stage. As illustrated in Fig. 1, five studies met the inclusion criteria of the systematic review [18, 49–52], and in the final stage, four studies qualified for inclusion in the meta-analysis [18, 49, 50, 52]. Study characteristics This review involved five studies published between 2010 and 2024 which compared salivary metabolomics and proteomics of patients with PC to those of controls (Table 1). Table 1Data summary and study attributes extracted from the selected eligible studies for systematic review and meta-analysisNo.AuthorsYearCountryMedical CenterType of salivaTemperature of saliva storageBiomarker typeSingle/Multiple biomarkersMeasurement methodModelValidation methodType of control groupPCa casesHealthy controlsCPb controls1Sugimoto et al. [49]2010JapanUCLA Medical Centerunstimulated saliva-metaboliteSingle (36 biomarkers, Supplementary File 3) and Multiple biomarkers (5 biomarkers) include: Phenylalanine, Tryptophan, Ethanolamine, Carnitine, C7H12N2O3.Capillary electrophoresis-time of flight mass spectrometry(CE-TOF-MS)Artificial neural networks (ANNs)tenfold cross-validationHealthy18872Asai et al. [18]2018Japanunstimulated saliva−80°CmetaboliteMultiple biomarkers (4 biomarkers) include: alanine, N1-acetylspermidine, 2-oxobutyrate, and 2-hydroxybutyrate.Capillary electrophoresis-mass spectrometry (CE-MS)Multiple logistic regression (MLR)k-fold cross-validation and resamplingHealthy+ CP3926143Tomita et al. [50]2018Japanunstimulated saliva−80°CmetaboliteSingle (3 biomarkers in two different storage conditions) include:N1-acetylspermineN1-acetylspermidinespermineLiquid chromatography-mass spectrometry(LC-MS)Random noise simulation-Healthy+ CP4039 healthy and CP controls4Deutsch et al. [51]2020Israelunstimulated saliva−80°CproteinsMultiple biomarkers (5 biomarkers) include: (Cytokeratin-14 (P02533), Lactoperoxidase (P22079), Cytokeratin-16 (P08730), Cytokeratin-17 (Q04695), and Peptidyl-prolyl cis-trans isomerase B (P23284)).LC-MS/MSc, LF-qMS d, 2DE eStatistical analysis-Healthy+ CP1516 healthy and CP controls5Ruiz‑Barrios et al. [52]2024United Mexican Statessaliva−80°CmetaboliteSingle (spermine)HPLCf with fluorescence detectionStatistical analysis-Healthy1111aPC: Pancreatic cancerbCP: Chronic pancreatitiscLC-MS/MS: Liquid chromatography-mass spectrometry/mass spectrometrydLF-qMS: Label-free quantitative mass spectrometrye2DE: 2D Gel electrophoresisfHPLC: High-performance liquid chromatography Data summary and study attributes extracted from the selected eligible studies for systematic review and meta-analysis Capillary electrophoresis-time of flight mass spectrometry (CE-TOF-MS) Single (3 biomarkers in two different storage conditions) include: N1-acetylspermine N1-acetylspermidine spermine Liquid chromatography-mass spectrometry (LC-MS) aPC: Pancreatic cancer bCP: Chronic pancreatitis cLC-MS/MS: Liquid chromatography-mass spectrometry/mass spectrometry dLF-qMS: Label-free quantitative mass spectrometry e2DE: 2D Gel electrophoresis fHPLC: High-performance liquid chromatography In this systematic review of five studies, four were related to salivary metabolites for PC diagnosis [18, 49, 50, 52]. These four studies were assessed in the meta-analysis, while only one study was related to salivary proteomics for diagnosing PC [51]. A total of forty-three study units were involved in our meta-analysis. These study units comprised 4328 subjects (854 patients with PC and 3474 noncancerous controls) (Supplementary File 4). In addition to healthy controls, three of the five studies involved in the systematic review used patients with CP as controls [18, 50, 51] (Table 1). In three studies, more than one salivary biomarker (multiple) was used in the analyses [18, 49, 51], whereas in two studies, only one type of salivary metabolite was used [50, 52]. In four studies, mass spectrometry methods were used to identify salivary metabolites in patients with PC [18, 49–51]; two studies used chromatography, while only one study utilized the high-performance liquid chromatography (HPLC) method for this purpose [52]. Additionally, only two studies applied internal validation to confirm and evaluate the results [18, 49] (Table 1). Quality assessment analysis Studies identified as having a high risk of bias presented concerns primarily in the domains of flow and timing [49] and patient selection [50]. The high risk in flow and timing reflects the potential inconsistencies in patient follow-up and timing of the index and reference tests. For Patient Selection, high risk arises due to potential biases in how patients were included in the studies. The index test domain often showed unclear risk of bias [18, 49]; conversely, the reference standard domain predominantly demonstrated low risk of bias [50]. Concerning applicability, studies categorized as low risk mainly reflected appropriate and valid patient selection criteria [18, 49, 50, 52]. However, the unclear risk for applicability was related reference standard domain [50] highlighting ambiguities in how the diagnostic benchmarks align with real-world scenarios. The index test domain was showed concerning applicability [49] (Fig. 2 and Supplementary File 5). Methodological quality assessment using Quality Assessment Tool for Diagnostic Accuracy Studies-2 (QUADAS-2). A Proportion of studies with low, high, or unclear risk of bias across QUADAS-2 domains. B Proportion of studies with low, high, or unclear concerns regarding applicability Salivary metabolomics diagnostic accuracy The overall specificity and sensitivity were 0.793 (95% CI: 0.764–0.819) and 0.784 (95% CI: 0.769–0.797), respectively (Supplementary File 6, Figure S1). The combined PLR and NLR were 3.82 (95% CI: 3.31–4.41) and 0.30 (95% CI: 0.26–0.34), respectively (Supplementary File 6, Figure S2). The overall DOR was 15.79 (95% CI: 12.61–19.76). A forest plot of pooled DOR based on single or multiple-salivary metabolite biomarkers is presented in Fig. 3. Forest plot of overall DOR (ascending order) based on single or multiple-salivary metabolite biomarkers for PC diagnosis SROC curves were used to evaluate the overall DOR associated with the salivary metabolites. This resulted in an AUC of 0.867 (Fig. 4). SROC curve of salivary metabolite biomarkers for PC diagnosis Figure S3 of Supplementary File 6 depicts evaluation of publication bias using Deeks' funnel plot method in the meta-analysis of the salivary metabolomics for PC diagnosis. The funnel plot shows the relationship between the DOR and the inverse of the square root of the effective sample size for each included study unit and illustrates no significant evidence of publication bias (p=0.750). A Spearman's correlation coefficient of 0.37 (p=0.016) was observed which suggests a statistically significant, moderately positive association between biomarker levels and disease severity on diagnostic accuracy. The I2 heterogeneity of DOR was 14.9% (p=0.083), suggesting negligible heterogeneity in DOR. The estimate of between-study variance, suggested minimal statistical differences across studies. Subgroup analysis Subgroup analysis were conducted based on the biomarker number and chemical structure and metabolic pathways (Supplementary File 7). Biomarker number The pooled DOR of the single metabolite (14.29, 95% CI: 11.59–17.62.59.62) was lower that of using multiple metabolites (101.68, 95% CI: 24.79–417.08.79.08). The heterogeneity chi-squared statistic was 34.65 in the single metabolite with a p-value of 0.669, indicating no statistically significant heterogeneity among the 40 included studies (p>0.05). The inconsistency (I2) was 0.0%, which reflected perfect homogeneity and all observed variability is attributed to random error rather than true differences between studies. The heterogeneity chi-squared statistic was 3.35 when using multi metabolite with a p-value of 0.187. This indicated no statistically significant heterogeneity among the included studies (p>0.05). However, the inconsistency (I2) of 40.3% suggests moderate heterogeneity in the data. Chemical structure and metabolic pathways In this study, signaling pathways related to metabolism and cellular regulation were classified into six main categories. This classification was based on biological function, role in diseases, and association with cellular processes. The categories are as follows: Polyamines Polyamines Tissue repair and collagen signaling pathways Tissue repair and collagen signaling pathways Detoxification and xenobiotic metabolism Detoxification and xenobiotic metabolism Specialized and regulatory signaling pathways Specialized and regulatory signaling pathways Amino acid and nitrogen metabolism Amino acid and nitrogen metabolism Nervous system and muscle function signaling pathways Nervous system and muscle function signaling pathways Forest plot of subgroup meta-analysis based on the biomarker pathway was shown in Figure S4 of Supplementary File 6. Nervous system and muscle function signaling pathways (24.58, 95% CI: 12.41–48.69.41.69) exhibited the highest DOR, while Polyamine signaling pathways (8.96, 95% CI: 5.36–14.98.36.98) resulted in the lowest DOR. The meta-regression analysis using inverse variance weights revealed a statistically significant association for the Pathway variable (p = 0.047), indicating a relative decrease in DOR (0.88, 95% CI: 0.78–1.00.78.00). In this meta-analysis, the investigated metabolic biomarkers were categorized into four groups based on sensitivity and specificity: Group 1 (sensitivity: 0.6–0.8, specificity: 0.6–0.8), Group 2 (sensitivity: 0.8–1.0, specificity: 0.6–0.8), Group 3 (sensitivity: 0.6–0.8, specificity: 0.8–1.0), and Group 4 (sensitivity: 0.8–1.0, specificity: 0.8–1.0). Comparisons with CA19-9 (Sensitivity = 0.72, specificity = 0.86) [60] revealed that biomarker selection based on disease threshold probability could significantly improve the sNB, which quantifies clinical utility ranging from 0 (no utility) to 1 (maximum utility), with negative values indicating potential harm (Fig. 5). For instance, in Group 2, the study by Tomita et al. [50] demonstrated that the metabolite spermine outperformed CA19-9 at threshold probabilities below 40%, reducing unnecessary interventions. This finding was corroborated in Group 4 by Ruiz-Barrios et al. [52], where spermine exhibited superior performance even at threshold probabilities up to 70%. In Group 3, Asai et al. [18] reported that a combination of four biomarkers achieved higher diagnostic accuracy than CA19-9. Furthermore, in Group 4, Sugimoto et al. [49] showed that a panel of 5 biomarkers surpassed CA19-9 at threshold probabilities exceeding 0.8. These findings suggest that multiple-biomarker panels enhance DOR and improve cost-effectiveness by reducing reliance on invasive procedures like biopsies. The horizontal axis represents the threshold probability of disease, which indicates the minimum disease risk a patient must have before specific therapeutic interventions are considered. The vertical axis represents the standardized net benefit (sNB). The higher the curve of a biomarker, the greater its clinical utility. Below, the biomarkers from the current meta-analysis are compared with the results of the meta-analysis of the protein biomarker CA19-9 in the serum of patients with PC [60] as reference line, serving as a basis for the diagnostic clinical value of these metabolomics biomarkers. A Group 1 (sensitivity 0.6–0.8, specificity 0.6–0.8), B Group 2 (sensitivity 0.8–1.0, specificity 0.6–0.8), C Group 3 (sensitivity 0.6–0.8, specificity 0.8–1.0), D Group 4 (sensitivity 0.8–1.0, specificity 0.8–1.0) Discussion The findings of this study underscore the potential of salivary metabolomics as a non-invasive diagnostic tool and a promising form of liquid biopsy for PC. The pooled sensitivity (0.784) and specificity (0.793) suggest moderate diagnostic accuracy, comparable to established biomarkers such as CA19-9 but with the added advantage of non-invasiveness. The DOR of 15.788 further highlights the discriminative power of salivary metabolites in distinguishing patients with PC from controls. The salivary metabolomics demonstrated good discriminative ability (AUC=0.867) but moderate diagnostic accuracy in terms of sensitivity (0.784) and specificity (0.793), suggesting its potential utility as an adjunctive screening tool rather than a standalone diagnostic method for PC. Multiple-metabolite panels demonstrated excellent diagnostic accuracy (DOR=101.68) compared to single metabolites (DOR=14.29), aligning with the multifactorial nature of PC pathogenesis and the need for combinatorial biomarker approaches. These results are consistent with emerging evidence that saliva, as a "mirror of systemic health," reflects metabolic perturbations caused by malignancies, including PC [15, 53, 54]. The observed heterogeneity in DOR across metabolic pathways (nervous system and muscle function signaling pathways: DOR = 24.58 vs. polyamine signaling pathways: DOR = 8.96) provides mechanistic insights into PC biology. Pathways linked to nervous system regulation may reflect tumor-induced systemic metabolic reprogramming or paraneoplastic effects [55, 56]. These findings align with studies linking polyamines to cell proliferation and differentiation in PC [57, 58]. However, the limited studies and geographic concentration (Japan) introduce potential bias, emphasizing the need for multi-center validation. Clinically, salivary biomarkers' moderate sensitivity/specificity suggests they may serve as a screening adjunct rather than a standalone diagnostic tool. However, their non-invasive nature and stability under standard storage conditions make them ideal for large-scale surveillance in high-risk populations, such as those with hereditary pancreatitis or familial PC syndromes. The superior performance of multiple-metabolite panels underscores the importance of integrating metabolites to enhance diagnostic precision [28, 59]. Notably, the sole proteomics study in this review reported promising results [56], highlighting the need for further exploration of salivary proteins in PC detection. Limitations of this study include heterogeneity in control groups and variability in metabolomics platforms. Additionally, the lack of standardized protocols for saliva collection (stimulated vs. unstimulated) may have influenced biomarker stability. Publication bias (higher likelihood of publication and access for studies with statistically significant outcomes) may have influenced the meta-analysis results [61, 62]. It is recommended that future research focus on standardizing methodologies, validating findings across diverse cohorts, and exploring the biological interactions between salivary biomarkers and pancreatic tumorigenesis to enhance diagnostic accuracy and clinical applications Conclusion Salivary metabolomics represents a promising, non-invasive adjunct for PC diagnosis, particularly when leveraging multiple-metabolite panels. While it currently lacks the standalone accuracy to replace conventional methods, its integration into screening algorithms could reduce unnecessary invasive procedures and accelerate early detection. Future work must address technical and clinical heterogeneity to unlock its full potential in precision oncology. Supplementary Information Supplementary Material 1. Supplementary Material 2. Supplementary Material 3. Supplementary Material 4. Supplementary Material 5. Supplementary Material 6. Supplementary Material 7. Acknowledgments Not applicable Abbreviations 2D Gel electrophoresis 95% Confidence interval Artificial neural networks Area under the curve Carbohydrate antigen 19-9 Capillary electrophoresis-mass spectrometry Capillary electrophoresis-time of flight mass spectrometry Chronic pancreatitis Diagnostic odds ratio False positive False negative High-performance liquid chromatography Hierarchical summary receiver characteristic operator curve I-square Liquid chromatography-mass spectrometry Liquid chromatography-mass spectrometry / mass spectrometry Label-free quantitative mass spectrometry Medical subject headings Multiple logistic regression Negative likelihood ratio Pancreatic cancer Positive likelihood ratio Quality Assessment Tool for Diagnostic Accuracy Studies Standardized net benefit Summary receiver operating characteristic True positive True negative Authors’ contributions MK, and SK contributed to the conception and design of the work. MK and SK independently evaluated the literature. MK and SK reviewed the articles based on the inclusion criteria by reviewing the titles, abstracts, and full texts of each study. MK and SK contributed to the acquisition and analysis of the data. MK drafted the manuscript. MF and SK revised the manuscript. All the authors have read and approved the final version of the submission. Funding Not applicable. Data availability The datasets used and/or analyzed during the current study are available in the Supplementary Files and Figures, and more required data will also be available from the corresponding author upon reasonable request Declarations Ethics approval and consent to participate This study is a systematic review and meta-analysis that involved analysis of data from previously published studies. No new data were collected from human participants by the authors. All studies included in this review had obtained informed consent from their participants, as stated in their original publications. Consent for publication All authors read and approved the final manuscript. Competing interests The authors declare no competing interests. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1. Supplementary Material 2. Supplementary Material 3. Supplementary Material 4. Supplementary Material 5. Supplementary Material 6. Supplementary Material 7. Data Availability Statement The datasets used and/or analyzed during the current study are available in the Supplementary Files and Figures, and more required data will also be available from the corresponding author upon reasonable request