Diagnostic valuation of MMP family members as biomarkers for pancreatic cancer: a systematic review and meta-analysis

ABSTRACT

Background

There are numerous controversies surrounding the diagnosis of matrix metalloproteinases (MMPs) in pancreatic cancer (PC). Consequently, this study conducts the first comprehensive meta-analysis evaluating the diagnostic performance of MMP family members in PC, highlight potential biomarkers, and investigate sources of heterogeneity.

Method

Articles on the MMP family and PC diagnosis were retrieved from PubMed and Web of Science. Diagnostic accuracy was assessed using sensitivity (SEN) and specificity (SPE). Threshold effects, heterogeneity (via meta-regression and subgroup analyses), and result robustness (via sensitivity analysis) were evaluated. Publication bias was examined using Deeks’ funnel plot.

Result

This meta-analysis included 48 studies from 23 articles, covering 12 types of MMP. The aggregated SEN, SPE, and area under the curve (AUC) for MMPs in diagnosing PC ranged from 0.61 to 0.76, 0.66 to 0.78, and 0.7273 to 0.8840, respectively. Meta-regression identified sample type as a major contributor to heterogeneity, with tissue-based MMPs showing markedly superior diagnostic performance compared to serum or plasma-based assays. Overall, MMPs showed potential diagnostic value, but their accuracy remains moderate and requires further validation in larger, well-designed studies.

Conclusion

MMPs represent potential candidates for PC diagnosis, but current evidence is insufficient for clinical application, requiring further prospective studies.

1. Introduction

As one of the cancers with the worst prognosis, the 5-year overall survival rate of PC is only 13% [1]. PC is an aggressive, highly malignant neoplasm characterized by rapid progression and early metastatic propensity, which is one of the key factors contributing to its poor prognosis and high mortality rate [2]. With the increase in incidence rate in recent years, PC has become the third leading cause of cancer death in Western countries. PC is projected to be the second leading cause of cancer deaths in the United States by 2030 [1,3]. Due to the absence of early symptoms, most patients are diagnosed at an advanced stage, resulting in missed opportunities for optimal treatment [4]. One of the biggest challenges in PC management is early detection in high-risk individuals (e.g., chronic pancreatitis patients) and early diagnosis in symptomatic patients [5]. Currently, effective early detection methods remain lacking. Ideal tumor markers exhibit high SEN in clinical examinations and can detect tumors early, making them crucial for early diagnosis [6]. In PC, CA19-9 is the only approved serum biomarker; however, its SEN and SPE are insufficient for early detection and diagnosis [7,8]. Accordingly, there is an urgent need to identify new biomarkers with higher SEN and SPE to enhance early diagnosis of PC.

The MMP family consists of 28 members, of which at least 23 are expressed in human tissues [9]. Based on their domain composition, MMPs are categorized into collagenase (MMP-1, MMP-8, MMP-13, MMP-18), gelatinase (MMP-2, MMP-9), stromelysins (MMP-3, MMP-10), matrix hemolysin (MMP-7, MMP-11, MMP-26), membrane-type MMPs (MMP-14, MMP-15, MMP-16, MMP-17, MMP-24, MMP-25) and other MMPs (MMP-12, MMP-19, MMP-20, MMP-21, MMP-23, MMP-27, MMP-28) [10]. MMPs facilitate angiogenesis, invasion and metastasis of tumor cells by exerting proteolytic activity and degrading physical barriers [9,11–14]. The expression levels of MMPs correlate with the progression and metastasis of various cancers, including breast cancer [15], ovarian cancer [16], colorectal cancer [17] and gastric cancer. Studies have identified increased expression levels of certain MMP family members in PC, suggesting their potential as diagnostic markers for PC [18–23]. However, the current findings regarding MMPs as diagnostic markers for PC remain inconsistent. Given the limitations of existing studies, we conducted this meta-analysis to summarize published clinical research and evaluate the diagnostic accuracy of the MMP family in differentiating of PC.

2. Methods

2.1. Search strategy

A systematic literature search was performed using electronic databases PubMed and Web of Science to identify eligible studies published up to 31 December 2022. The search strategy included the following terms: (MMP OR Matrix Metalloproteinase) AND (Pancreatic Neoplasm OR Pancreas Neoplasm OR Pancreatic Cancer OR Pancreas Cancer). After an initial screening of article abstracts and titles, full-text reviews were conducted for studies that met the inclusion criteria. Additionally, the reference lists of included articles were manually searched to identify other relevant studies. The literature search was conducted independently by two researchers, and any discrepancies were resolved through discussion.

2.2. Inclusion and exclusion criteria

Inclusion criteria: (1) PC was confirmed by histological examination; (2) The article was published in English; (3) Sufficient data were available to extract or calculate true positive (TP), false positive (FP), false negative (FN), and true negative (TN) values; (4) Samples were derived from tissue or blood; (5) MMPs reported in three or more studies were subjected to quantitative analysis, while those reported in fewer than three studies underwent qualitative analysis. Exclusion criteria: (1) Studies using samples other than tissue or blood; (2) Non-English publications; (3) Studies involving non-human subjects; (4) Identical publications, conference summaries, correspondence, comments, and expert opinions; (5) Articles without full-text availability.

2.3. Data extraction and quality evaluation

Two researchers independently extracted data from the included articles, including the first author, year of publication, country, MMP type, sample source, testing method, number of cases and controls, as well as true positive (TP), true negative (TN), false positive (FP), and false negative (FN) values. The two researchers conducted independent evaluations of the study’s quality using the QUADAS-2 tool [24].

2.4. Statistical analysis

All statistical analyses were performed using Meta-DiSc 1.4 and STATA 17.0. For each study, TP, FP, FN, and TN were extracted to compute SEN, SPE, positive likelihood ratio (PLR), negative likelihood ratio (NLR), and diagnostic odds ratio (DOR), along with their 95% confidence intervals (CI). To assess potential threshold effects, we applied Spearman rank correlation analysis. In cases where no threshold effect was identified, pooled diagnostic estimates were computed using either a fixed-effects or random-effects model, depending on the level of heterogeneity. Heterogeneity across studies was evaluated using the I² statistic, with I² > 50% indicating substantial heterogeneity. A summary receiver operating characteristic (SROC) curve was constructed, and the area under the curve (AUC) was used to assess overall diagnostic accuracy. Subgroup analyses were performed based on sample type (tissue vs. blood), ethnicity, and sample size to identify potential sources of heterogeneity. Where applicable, meta-regression was conducted to further explore covariates contributing to heterogeneity. Sensitivity analyses were used to evaluate the robustness of the pooled results. Publication bias was assessed using Deeks’ funnel plot asymmetry test, with a p < 0.05 indicating significant bias.

3. Results

3.1. Search results and research features

The flow chart of study selection is illustrated in Figure 1. After removing 487 duplicate articles, a total of 1984 articles were identified (641 from PubMed, 1342 from the Web of Science, and 1 from other sources). The remaining 1497 articles were screened based on titles and abstracts. Of these, 1453 articles were excluded, and the remaining 44 articles were reviewed in full. According to the exclusion criteria, 21 articles were excluded, of which 8 lacked sufficient data, 9 had no control group, 2 were unrelated to PC, and 2 used non-eligible sample sources. Ultimately, 23 articles were included in the meta-analysis [19–23,25–42].

Figure 1.

Flow chart of literature screening.

The main characteristics of the included articles are summarized in Table 1. A total of 48 studies from 23 articles investigated 12 types of MMPs: 12 studies each on MMP-2, MMP-9, 12, 8 on MMP-7, 3 on MMP-11 and MMP-12, 2 on MMP-1, MMP-3, and MMP-10, and 1 on MMP-8, MMP-14, MMP-21, and MMP-26. Among them, seven articles evaluated more than one MMP to diagnose PC. In the included studies, samples were obtained from either tissue or blood. All of the studies included in this review were published between 1998 and 2021.

Table 1.

Basic characteristics of the included research.

Author	Year	Country	MMPs	Specimen	Sample size		TP	FP	TN	FN
					Case	Control
Marta [16]	2019	Poland	MMP-2	Serum	92	134	43	44	90	49
Zhu [17]	2019	China	MMP-2	Tissue	67	20	40	6	14	27
Zhang [18]	2020	China	MMP-11	Serum	116	37	88	7	30	28
			MMP-11	Serum	116	18	93	5	13	23
Xu [19]	2013	China	MMP-2	Tissue	45	27	29	0	27	16
DRANKA-BOJAROWSKA [20]	2021	Poland	MMP-2	Serum	30	60	24	26	34	6
			MMP-9	Serum	30	60	26	26	34	4
Ville [23]	2007	Finland	MMP-21	Tissue	25	18	22	4	14	3
			MMP-26	Tissue	25	18	13	2	16	12
Nidhi [24]	2017	India	MMP-2	Plasma	127	50	84	17	33	43
MAIKEN [43]	2010	Denmark	MMP-9	Serum	51	52	30	34	18	21
Xiang [25]	2017	China	MMP-2	Tissue	30	17	19	0	17	11
			MMP-9	Tissue	30	17	17	0	17	13
Xu [26]	2015	China	MMP-9	Tissue	103	6	61	0	60	42
Barbara [27]	2009	Poland	MMP-9	Serum	78	115	50	45	70	28
Lucie [28]	2004	Britain	MMP-1	Tissue	45	10	29	3	7	16
			MMP-2	Tissue	75	10	50	7	3	25
			MMP-3	Tissue	45	10	40	10	0	5
			MMP-7	Tissue	45	10	41	2	8	4
			MMP-8	Tissue	45	10	33	6	4	12
			MMP-9	Tissue	45	10	19	0	10	26
			MMP-11	Tissue	75	10	52	2	8	23
			MMP-12	Tissue	45	10	20	10	0	25
			MMP-14	Tissue	75	10	44	4	6	31
Duan [29]	2013	China	MMP-9	Tissue	62	16	37	3	13	25
Nahla [30]	2020	Egypt	MMP-7	Serum	25	25	24	0	25	1
			MMP-9	Serum	25	25	17	5	20	8
			MMP-10	Serum	25	25	19	0	25	6
			MMP-12	Serum	25	25	16	0	25	9
Tian [31]	2008	China	MMP-9	Tissue	51	8	42	0	8	9
Li [32]	2005	China	MMP-7	Tissue	47	22	30	7	15	17
Christoph [33]	2014	Germany	MMP-1	Serum	48	44	32	14	30	16
			MMP-2	Serum	48	44	31	29	15	17
			MMP-3	Serum	48	44	33	15	29	15
			MMP-7	Serum	48	44	43	0	44	5
			MMP-9	Serum	48	44	37	21	23	11
			MMP-10	Serum	48	44	41	8	36	7
			MMP-12	Serum	48	44	48	0	44	0
Li [34]	2008	China	MMP-2	Tissue	40	20	22	3	17	18
Koshiba [35]	1998	Japan	MMP-2	Tissue	33	27	33	13	14	0
			MMP-9	Tissue	33	27	7	6	21	26
Katarzyna [36]	2016	Poland	MMP-2	Tissue	29	29	16	1	28	13
			MMP-7	Tissue	29	29	28	2	27	1
			MMP-9	Tissue	29	29	13	0	29	13
Zhai [37]	2015	China	MMP-2	Tissue	130	130	100	38	92	30
Hyung-Doo Park [38]	2012	South Korea	MMP-7	Serum	109	70	78	14	56	31
			MMP-7	Serum	139	146	90	31	115	49
Zhao [39]	2016	China	MMP-7	Tissue	81	81	45	44	37	36

Abbreviations: TP, true-positive; FP, false-positive; TN, true-negative; FN, false-negative.

3.2. Quality assessment

The quality of eligible studies was evaluated using the QUADAS-2 tool, as shown in Figure S1. Overall, the quality of the studies included was relatively good.

3.3. Meta-analysis of the diagnostic value of MMP-2 in PC

Twelve studies evaluated the diagnostic value of MMP-2 in PC. Substantial heterogeneity was found after testing the heterogeneity between studies (SEN and SPE I² were 81.2% and 87.3%, respectively). Therefore, a random-effects model was applied. The pooled estimates were as follows: SEN 0.67 (95% CI: 0.63–0.70), SPE 0.68 (95% CI: 0.64–0.72), PLR 2.02 (95% CI: 1.43–2.85), NLR 0.49 (95% CI: 0.38–0.64), and DOR 5.28 (95% CI: 2.70–10.33) (Figure 2). The AUC was 0.75 (Figure S8A). The above data demonstrated the relatively outstanding diagnostic value of MMP-2 in PC.

Figure 2.

Forest plots of sensitivity (A), specificity (B), positive likelihood ratios (C), negative likelihood ratios (D) and diagnostic odds ratio (E) for MMP-2 in the diagnosis of pancreatic cancer.

The Spearman correlation coefficient was 0.413 (p > 0.05), suggesting that the heterogeneity was not attributable to threshold effects. Subsequently, subgroup analysis and meta-regression were conducted based on ethnicity, sample type, and sample size to explore the sources of heterogeneity. As shown in Table S1, subgroup analysis revealed that the Asian population demonstrated higher SEN (0.71 vs 0.60), SPE (0.74 vs 0.61), and AUC (0.80 vs 0.66) than the Caucasian population. Regarding sample type, tissue-derived samples generated higher SEN (0.69 vs 0.63), SPE (0.76 vs 0.61), and AUC (0.81 vs 0.67) compared to blood-derived samples. However, meta-regression indicated that these factors were not statistically significant source of heterogeneity (Figure S2). This inconsistency between numerical subgroup differences and statistical significance is likely due to small subgroup sample sizes and residual confounding from unmeasured factors (e.g., regional assay standards), which limited the statistical power of meta-regression.

Sensitivity analysis showed that two studies had a substantial influence on the pooled results. After excluding these studies, the updated pooled estimates were as follows: SEN 0.65 (95% CI: 0.61–0.69), SPE 0.67 (95% CI: 0.63–0.71), PLR 1.92 (95% CI: 1.33–2.78), NLR 0.52 (95% CI: 0.40–0.68), and DOR 4.12 (95% CI: 2.16–7.88), showing no substantial deviation from the original estimates. Detailed results are shown in Figure S3. Therefore, the sensitivity analysis supports the overall robustness of the findings.

As shown in Figure 5(A), the p-value derived from Deeks’ funnel plot was 0.32. The symmetric distribution of the funnel plot suggests that the pooled data are reliable and that no significant publication bias is present.

Figure 5.

Deeks’ funnel diagram of MMP-2 (A), MMP-7 (B), MMP-9 (C), MMP-11 (D) in the diagnosis of pancreatic cancer.

3.4. Meta-analysis of the diagnostic value of MMP-7 in PC

Eight studies from seven articles evaluated the diagnostic value of MMP-7 in PC. The Spearman correlation coefficient was −0.814 (p < 0.05), indicating the presence of a threshold effect. The included studies applied heterogeneous cutoff definitions, which likely contributed to the considerable variability in diagnostic performance (SEN: 0.56–0.97; SPE: 0.46–1.00). (Figure S4). Lower thresholds generally yielded higher SEN but lower SPE, while stricter thresholds improved SPE at the expense of SEN. This explains much of the substantial heterogeneity (I² > 85%) observed in the pooled MMP-7 analysis. Meta-analysis of all eight studies showed that the aggregated SEN and SPE of MMP-7 for diagnosing PC were 0.72 (95% CI: 0.68–0.76) and 0.77 (95% CI: 0.72–0.81), respectively. The pooled PLR and NLR were 4.18 (95% CI: 1.96–8.94) and 0.28 (95% CI: 0.16–0.47), respectively (Figure S4). The AUC of the SROC curve was 0.88 (Figure S8B). Subgroup analysis showed that the I² for sensitivity decreased to 0% in the Caucasian population after stratification by ethnicity, suggesting that ethnicity might contribute to heterogeneity in sensitivity. The Caucasian population had higher SEN (0.92 vs 0.67) and SPE (0.95 vs 0.72) compared to the Asian population (Table S2). Due to the limited number of studies and the diversity of cutoff definitions, a sensitivity analysis restricted to standardized thresholds could not be conducted. Goodness-of-fit and bivariate normality analysis indicated that the bivariate mixed effect model was appropriate for selection, and no outliers were identified through impact analysis and outlier detection (Figure 3). As shown in Figure 5(B), the scatter points were distributed relatively symmetrically on both sides of the regression line, with no substantial rightward skew; the regression test (p = 0.14) did not suggest significant publication bias.

Figure 3.

Sensitivity analysis results of MMP-7 in diagnosing pancreatic cancer (a)goodness of fit, (b) bivariate normality, (c) influence analysis, and (d) outlier detection.

3.5. Meta-analysis of the diagnostic value of MMP-9 in PC

The accuracy of MMP-9 expression in diagnosing PC was analyzed in 12 studies, and a random-effects model was applied due to existing heterogeneity. As shown in Figure 4, the SEN and SPE of MMP-9 were 0.61 (95% CI: 0.57–0.65) and 0.66 (95% CI: 0.61–0.70), respectively. In addition, PLR, NLR, and DOR were 2.04 (95% CI: 1.37–3.04), 0.52 (95% CI: 0.41–0.67), and 5.19 (95% CI: 2.50–10.57), respectively. The SROC curve showed an AUC of 0.72.

Figure 4.

Forest plots of sensitivity (A), specificity (B), positive likelihood ratios (C), negative likelihood ratios (D) and diagnostic odds ratio (E) for MMP-9 in the diagnosis of pancreatic cancer.

The Spearman correlation coefficient was 0.413 (p = 0.183), indicating no significant threshold effect. Ethnicity, sample type, and sample size were included in both meta-regression and subgroup analysis (Table S3, Figure S5). Subgroup analysis revealed that the Asian population demonstrated a higher diagnostic value than the Caucasian population, with a SEN of 0.56 (95% CI: 0.53–0.65), SPE of 0.88 (95% CI: 0.78–0.94), DOR of 9.39 (95% CI: 1.71–51.52), and AUC of 0.86. Regarding sample type, a significant difference was observed between subgroups: tissue-derived samples showed a higher SPE of 0.92 (95% CI: 0.85–0.92) but a lower SEN of 0.56 (95% CI: 0.50–0.61). Notably, after stratifying by sample size, the I² value for sensitivity decreased to 0% in subgroups with a sample size greater than 100. This suggests that sample size may contribute to heterogeneity in SEN. In addition, meta-regression analysis indicated that sample type was a significant contributor to heterogeneity in SPE. Sensitivity analysis showed no substantial change in results after excluding a single study, confirming robustness (Figure S6). Deeks’ funnel plot was used to assess publication bias, and the results (Figure 5(C)) showed an evident asymmetry in the distribution of studies around the regression line. The regression test yielded a statistically significant p-value (p = 0.03), indicating the presence of potential publication bias. The skewed pattern, with smaller studies favoring higher diagnostic odds ratios, suggests that positive findings may have been preferentially published or selectively reported.

3.6. Meta-analysis of the diagnostic value of MMP-11 in PC

Three studies investigated the diagnostic accuracy of MMP-11 in PC. As no significant heterogeneity was observed, a fixed-effects model was used to summarize SEN, SPE, PLR, NLR, and DOR of 0.76 (95% CI: 0.71–0.81), 0.78 (95% CI: 0.68–0.88), 3.46 (95% CI: 2.17–5.52), 0.31 (95% CI: 0.24–0.39), and 11.63 (95% CI: 6.18–21.87), respectively (Figure S7). The area under the SROC curve was 0.80 (Figure S8D). Deeks’ funnel plot for MMP-11 included only three studies. The scatter points were distributed closely along the regression line, and the regression asymmetry test yielded a p-value of 0.13, suggesting no statistically significant publication bias (Figure 5(D)). However, due to the small number of included studies, the test may have limited power to detect asymmetry. Therefore, the absence of publication bias should be interpreted with caution.

3.7. The diagnostic value of other MMP family members for PC

Three studies investigated the diagnostic value of MMP-12, while two articles investigated MMP-1, MMP-3, and MMP-10, and one article evaluated the diagnostic value of MMP-8, MMP-14, MMP-21, and MMP-26 in PC. The small number of studies precluded quantitative meta-analysis, and MMP-12 analysis was further compromised by extreme SPE variability: one of the three studies reported SPE = 0, while the other two showed SPE = 1. This outlier caused marked heterogeneity, rendering pooled results highly unreliable. A study from the UK reported the expression of MMP-8 and MMP-14 in PC, showing that MMP-8 was expressed in 73.33% of PC patients and 60% of healthy individuals, with SEN and SPE of 73.33% and 40%, respectively. MMP-14 was expressed in 58.67% of PC patients and 40% of healthy individuals, with SEN and SPE of 58.67% and 60%, respectively. A Finnish study performed immunohistochemical analysis of MMP-21 and MMP-26 in 25 PC tissues and 18 non-tumor tissues, revealing increased expression of MMP-21 and MMP-26 in PC patients. MMP-21 was detected in 22 of 25 PC samples and 4 of 18 control samples, with SEN and SPE of 88.00% and 77.78%, respectively. MMP-26 was detected in 13 of 25 cases of adenocarcinoma and 2 of 18 control samples, with SEN and SPE of 52.00% and 88.89%, respectively. Two studies on MMP-1 reported SEN and SPE of 64.44% and 67%, and 70% and 68%, respectively. Two studies on MMP-3, conducted in the UK and Germany, reported SENs of 88.89% and 69%, respectively. Two studies evaluated MMP-10 expression levels in serum samples from PC patients and healthy individuals. One study from Egypt showed that the SEN and SPE of MMP-10 were 76% and 100%, respectively. Another study from the UK, involving 48 PC patients and 44 healthy individuals, showed the SEN and SPE of MMP-10 were 85% and 82%, respectively.

4. Discussion

As a highly lethal malignant tumor, PC is projected to become the second leading cause of cancer-related deaths in Western countries within the next decade. The absence of effective early detection tools poses significant challenges to its early diagnosis and evaluation of PC. Currently, commonly used biomarkers for PC are mainly derived from germline mutations in susceptible genes, tissue biopsies and surgical specimens, liquid biopsies (including blood, urine, saliva, and pancreatic fluid), and imaging-based omics biomarkers [44]. The diagnostic value of these biomarkers remains limited. For example, CA19-9 is the most widely used biomarker for PC. However, some studies have reported that within 0–6 months prior to the diagnosis of PDAC, the SEN and SPE of CA19-9 are 0.60 and 0.99, respectively [45]. The SEN of CA19-9 is 0.68 within 1 year before diagnosis and decreases to 0.53 within 2 years before diagnosis [46]. Carcinoembryonic antigen (CEA), the second most commonly used biomarker for PC, has been shown to have a pooled diagnostic SEN of only 0.45 [47]. Given the limited SEN or SPE of currently available biomarkers, an increasing number of novel and alternative biomarkers are being studied for their diagnostic potential. The relationship between MMP family members and various tumors (including PC) has been extensively reported, and their involvement in multiple tumorigenic processes has been well established. The processes of metastasis and local tumor invasion largely depend on the degradation of the extracellular matrix (ECM). The tumor microenvironment, especially the ECM, is considered a key contributor to cancer progression. MMPs are classified as the main factors leading to the degradation of ECM components. Huang Xiwen [48] revealed that the diagnostic SEN of the potential serum biomarkers MMP-7 and MMP-9 in colorectal cancer patients was 87.4% and 82%, respectively. Liu Xu et al. [49] found that serum MMP-9 levels in non-small cell lung cancer (NSCLC) than in those with benign lung diseases. MMP-9 was identified as a potential diagnostic biomarker for NSCLC, with a SEN of 87.1%, SPE of 80.56%, and an AUC of 0.89. Weaver et al. [50] have confirmed that MMP-12 plays a protective role in regulating T cell function, which could potentially be utilized to enhance the effectiveness of immunotherapy strategies, offering new treatment options for pancreatic cancer patients. Numerous studies have investigated the diagnostic value of the MMP family in PC; however, their findings remain inconsistent. Therefore, we conducted this meta-analysis to systematically evaluate the diagnostic accuracy of MMPs in PC. To our knowledge, this is the first meta-analysis to independently assess the diagnostic performance of MMPs in PC.

Through searching multiple databases, 48 studies on the diagnostic value of MMP in PC were ultimately collected, involving a total of 12 MMPs. Among these studies, quantitative analyses were performed for MMP-2, MMP-7, MMP-9, MMP-11, each of which was reported in at least three studies. Notably, MMP-12 results are highly unstable. This instability stems from an outlier study (SPE = 0), whereas the other two studies reported SPE = 1. In addition, the diagnostic performance of MMP-2 in PC was represented by a pooled SEN of 0.67 (95% CI: 0.63–0.70) and SPE of 0.68 (95% CI: 0.64–0.72). The comprehensive PLR, NLR, and DOR were 2.02 (95% CI: 1.43–2.85), 0.49 (95% CI: 0.38–0.64), and 5.28 (95% CI: 2.70–10.33), respectively. The AUC was 0.75. Due to significant heterogeneity, subgroup analysis and meta-regression were conducted. However, heterogeneity was not caused by ethnicity, sample types, or sample size. Notably, “non-significant meta-regression” does not negate numerical subgroup differences: meta-regression tests overall contributions to heterogeneity, while subgroup results reflect within-group performance. Small subgroup sizes reduced statistical power, and unmeasured factors likely masked true associations. The aggregated SEN and SPE of MMP-7 for diagnosing PC were 0.72 (95% CI: 0.68–0.76) and 0.77 (95% CI: 0.72–0.81), respectively; the summary PLR and NLR were 4.18 (95% CI: 1.96–8.94) and 0.28 (95% CI: 0.16–0.47), respectively; the AUC of the SROC curve was 0.88. The Spearman correlation coefficient was −0.814 (p < 0.05), indicating that the heterogeneity was attributable to threshold effects. Subgroup analysis further suggested that ethnicity may be a contributing factor to the observed heterogeneity. The pooled SEN, SPE, PLR, NLR, and DOR of MMP-9 were 0.61 (95% CI: 0.57–0.65), 0.66 (95% CI: 0.61–0.70), 2.04 (95% CI: 1.37–3.04), 0.52 (95% CI: 0.41–0.67), and 5.19 (95% CI: 2.50–10.57), respectively; AUC was 0.72. Meta-regression analysis suggested that sample type may be a potential source of heterogeneity. Sensitivity analysis showed no significant change in the pooled results after excluding studies with a potential impact, confirming the robustness of the meta-analysis findings. An ideal AUC should approach 1, with a minimum value of 0.7 to indicate its clinical utility as a biomarker. Overall, MMP family members such as MMP-2, MMP-7, and MMP-9 demonstrate moderate diagnostic performance in PC, with pooled AUC values ranging from 0.72 to 0.88 in our meta-analysis. In clinical practice, although CA19-9 remains the cornerstone biomarker, it exhibits limited sensitivity in detecting early-stage PC. Combining MMPs with CA19-9 has the potential to enhance overall diagnostic accuracy. Yet limited data precluded head-to-head meta-analysis or comparative ROC analysis: only 3 MMP-7 studies measured CA19-9/CEA, without standardized combined metrics. Thus, MMPs’ incremental value remains unproven and requires future prospective studies. Yablecovitch et al. [51] demonstrated that the combination of MMP-7, SDC1, and CA19-9 achieved a perfect area under the ROC curve (AUC = 1.0, p < 0.001) with 100% SEN for distinguishing PDAC. Such biomarker panels hold potential to rule out PDAC in suspected cases due to their high SEN. By integrating biomarker weights and inter-relationships, these models may improve early detection rates and patient outcomes in PC. Furthermore, several studies have reported the diagnostic value of MMP-1, MMP-3, MMP-10, MMP-8, MMP-14, MMP-21, and MMP-26 in PC. However, due to the limited number of studies (only one or two per marker), a meta-analysis could not be performed. Nevertheless, the reported SEN and SPE of these MMPs still indicate potential diagnostic value for PC (sensitivity: 52.00%-88.89%; specificity: 0.00%-100.00%). Among them, MMP-10 demonstrated particularly promising diagnostic performance, with reported SEN of 76% and 85%, and SPE of 100% and 82%, respectively.

The expression of MMPs in PC varies depending on their type, ethnicity, and sample source. Hofmann et al. [52,53] utilized DNA microarray technology to identify differentially expressed genes between cancerous and adjacent non-tumor tissues of NSCLC. The results revealed MMP-12 expression in cancerous tissues was 4.1-fold higher than in adjacent non-tumor tissues. Gene expression analysis showed that MMP-12 was not expressed in normal tissues, whereas immunohistochemical staining confirmed its presence exclusively in tumor cells. Christoph [36] analyzed compartment-specific expression profiles of multiple MMPs using multiple ELISA assays and found that MMP-12 could serve as an ideal diagnostic predictor (AUC = 1, SEN = 100%, SPE = 100%). In addition, MMP-2, MMP-7, and MMP-9 have long been considered as potential diagnostic biomarkers for PC [30,54]. Due to the ability of MMP-2 to degrade the extracellular matrix, reduce tumor cell adhesion, and promote neovascularization, elevated MMP-2 expression is strongly associated with increased risk of distant metastasis, thereby accelerating tumor progression. The nerve growth factor in PC can stimulate the expression and activation of MMP-2, and act on the p44/42 MAPK pathway through TRK-A receptor, thereby enhancing the neurotropic invasion of PC cells [55]. MMP-7, also known as matrilysin, is a zinc-dependent endopeptidase, primarily expressed in epithelial cells of glandular tissues, and derived from the extracellular matrix [56]. In pancreatic metaplasia models, MMP-7 expression is upregulated and has been associated with increased tumor size and metastasis [57,58]. MMP-9 is recognized as a critical regulator that facilitates the secretion of proangiogenic factors, including vascular endothelial growth factor (VEGF), thereby promoting tumor-associated neovascularization [59,60]. PC has low vascularity, which constitutes a quite difficult tumor angiogenesis environment, resulting in a higher demand for MMPs (such as MMP-9). Studies have confirmed that MMP-9 is positively expressed in both the cytoplasm of tumor cells and the surrounding stromal matrix in PDAC patients [33]. Furthermore, meta-regression and subgroup analysis indicate that both MMP-9 and MMP-2 exhibit reliable diagnostic performance in cancer detection when assessed using either tissue or blood samples, with tissue samples showing superior accuracy. Similarly, studies have shown that the expression levels of most angiogenic cytokines and MMPs are significantly increased in PC tissues, surrounding stromal matrices, and serum. Although some researchers have found that the expression of most angiogenic cytokines and MMPs is higher in the serum than in tissue, the overall diagnostic performance tends to be superior when tissue-based assays are used [36]. Moreover, neither ethnicity nor sample size had a significant impact on diagnosis performance. This heterogeneity may be attributed to differences in clinical application and underlying biological characteristics between tissue-based and serum-based MMP assays. Tissue-based assays require the collection of tumor samples via surgical excision or needle biopsy. These methods enable direct evaluation of MMP expression in the tumor microenvironment, offering pathological insights into ECM degradation and tumor invasion. While informative for understanding tumor pathogenesis and prognosis, tissue sampling is invasive and may cause patient discomfort or complications. In contrast, serum-based assays are noninvasive and suitable for large-scale screening and longitudinal monitoring. Serum MMPs, released from tumor cells and adjacent tissues into the bloodstream, reflect tumor burden and systemic effects. Dynamic changes in serum MMP levels, especially MMP-9 May 2001aid early detection, treatment monitoring, and recurrence prediction. However, serum levels can be influenced by inflammation, trauma, or other nonmalignant conditions, reducing diagnostic specificity. Moreover, serum MMPs may not fully reflect local tumor biology, limiting their sensitivity for small or early-stage tumors.

At the same time, this meta-analysis has several limitations. First, some included original studies lacked a blinded design, which may have affected diagnosis accuracy. Second, substantial methodological heterogeneity existed across studies, including variable cutoff values, control group selection, detection techniques, and MMP expression differences across sample sources; this potentially led to inconsistent conclusions, and full identification of heterogeneity sources remained elusive despite subgroup analysis, meta-regression, and sensitivity analysis. Third, the limited number of studies on certain MMP types precluded further subgroup or quantitative analysis. Fourth, the absence of head-to-head comparison data with CA19-9/CEA meant we could not definitively confirm MMPs’ incremental clinical value over existing tests. Thus, the diagnostic performance and clinical value of the MMP family in PC should be interpreted cautiously, and more high-quality, standardized studies are needed to validate our findings.

5. Conclusions

Overall, members of the MMP family hold promise as novel diagnostic biomarkers for PC, but their diagnostic accuracy remains moderate, and no single MMP shows definitive superiority. MMP-7 and MMP-9 exhibit relatively stable performance and may serve as complementary tools to existing biomarkers. In the future, larger-scale, multi-center clinical studies will be essential to further validate the diagnostic utility of MMPs in PC.

Funding Statement

This work was supported by the Key Research Program of Higher Education in Henan Province [No. 22A330003].

Article highlights

• This meta-analysis systematically evaluated the diagnostic accuracy of 12 MMPs in pancreatic cancer across 48 studies.

• Among MMP family members, MMP-7 and MMP-9 exhibited relatively stable diagnostic performance, with pooled AUC values of 0.88 and 0.72, respectively, suggesting their potential as candidate biomarkers.

• Pooled sensitivity and specificity of MMPs ranged from 0.61–0.76 and 0.66–0.78, respectively, across studies.

• Meta-regression identified sample type as the major contributor to heterogeneity among included studies.

• Tissue-based assays showed significantly higher diagnostic accuracy than serum/plasma-based assays.

• Subgroup analysis revealed better performance of MMPs in Asian populations compared to Caucasian populations.

• MMP-based diagnostic models may complement CA19-9, especially in cases with borderline biomarker levels.

• The study emphasizes the need for large-scale, multi-center prospective studies – especially head-to-head comparisons with CA19-9 and development of multi-marker models – to validate MMPs’ clinical utility.

Author contribution

Mengyuan Lu conceived and designed the study, and was primarily responsible for manuscript writing and revision.

Donglin Jiang, Jiaxin Zheng and Ling Liu performed data extraction, quality assessment, and statistical analysis.

Yajing Shen, Tiandong Li and Peng Wang prepared the figures and contributed to data visualization and result interpretation.

Jianxiang Shi and Keyan Wang assisted with methodology design, supervised data verification.

Hua Ye supervised the overall study, provided critical revisions, and approved the final version of the manuscript.

All authors have read and approved the final manuscript.

Disclosure statement

The authors have no relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.

Reviewer Disclosure

Peer reviewers on this manuscript have no relevant financial or other relationships to disclose

Writing Disclosure

No writing assistance was utilized in the production of this manuscript.

Ethical conduct of research

The authors state that they have obtained appropriate institutional review board approval () and/or have followed the principles outlined in the Declaration of Helsinki for all human or animal experimental investigations. In addition, for investigations involving human subjects, informed consent has been obtained from the participants involved.

Data availability statement

All the data presented in this study are available in this article.

AI declaration

No AI-based tools or technologies were used in the generation, editing, or proofing of this manuscript.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/17520363.2025.2571394