Pentraxin 3 as a promising biomarker for cancer detection: a systematic review and meta-analysis

Background Mounting studies reported that circulating pentraxin 3 (PTX3) expression level was signicantly different between cancer patients and healthy groups, suggesting that PTX3 may be a potential biomarker for cancer detection. However, the results were inconsistent. In this paper, a systematic review and meta-analysis was performed to quantitatively assess the diagnostic value of PTX3 in cancer detection.Methods A comprehensive computerized literature search was conducted in Embase, PubMed, Cochrane Library, Web of Science, Chinese National Knowledge Infrastructure (CNKI) from inception to July 31, 2019. Eligible studies were identied and raw data were extracted. Diagnostic estimates were synthesized using STATA (version 12.0) and MetaDisc (version 1.4) statistical software.Results Overall, 9 studies from 8 citations with a total of 1408 cancer patiens and 3116 controls were included in this meta-analysis. The global sensitivity was 0.70 (95% condence interval (CI): 0.67 – 0.72), and the specicity was 0.77 (95% CI: 0.75 – 0.78). The pooled positive likelihood ratio (PLR), negative likelihood ratio (NLR), and the diagnostic odds ratio (DOR) were 2.86 (95% CI: 2.29 – 3.56), 0.40 (95% CI: 0.32 – 0.50) and 7.38 (95% CI: 5.05 – 10.78), respectively. The merged AUC was 0.80 (95% CI: 0.76 – 0.83).Conclusion The serum PTX3 appears to be a reliable biomarker for cancer detection though large-scale multicenter studies are needed.

To summarize the currently available data, a systematic review and meta-analysis was conducted to evaluate the diagnostic performance of PTX3 in cancer screening based on several clinical trials.

Methods
This study was conducted according to the Preferred Reporting Items for Systematic reviews and Metaanalysis (PRISMA) statement [20,21] .

Publication search strategy
A comprehensive literature search was carried out in PubMed, Embase, Cochrane Library, Web of Science, and China National Knowledge Infrastructure (CNKI) from inception up to July 31, 2019. The search words used for article identi cation included: i) "cancer" or "carcinoma" or "tumor" or "neoplasma"; ii) "pentraxin 3" or "PTX3"; iii) "diagnostic" or "diagnosis". Meanwhile, the literatures cited by all the suitable studies were scanned to acquire more relevant citations. Additionally, no language restrictions were applied in this systematic review and meta-analysis.

Study selection
Two researchers (Qindong Liang and Guangjie Zhang) carefully scanned all the searched papers and independently identi ed the articles suitable for the systematic review. The articles were considered eligible if they satis ed all the following criteria: 1) patients with cancer were con rmed by histopathological examination or endoscopy; 2) blood samples were collected before any operation; 3) level of pentraxin 3 was measured in serum, plasma or blood; 4) the study was a case-control, prospective or retrospective cohort design; 5) su cient data to draw 2 by 2 contingency tables. On the contrary, letters, commentary, abstracts presented in conferences, case reports and reviews were excluded. For duplicate publications adopting the same population, we selected the most recent article or the one with most informative data.
Quality assessment Study quality assessment was completed independently by the same two researchers according to the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) checklist [22]. Any disagreement was settled through thorough discussion, and when it was necessary, arbitration by a third researcher to reach an agreement.

Data extraction
Two investigators (Qindong Liang and Guangjie Zhang) meticulously scrutinized all the included articles and the relevant data of each enrolled studies was extracted. For each research, the following data were collected: 1) study characteristics (the rst author's name, year of publication, country, assay method, the number of cases and control groups); 2) the diagnostic data (such as sensitivity, speci city, cutoff value, etc.). The exact number of 2 by 2 contingency tables (true positive (TP), false positive (FP), true negative (TN) and false negative (FN)) were obtained directly or by recalculation.

Statistic analysis
The global diagnostic variables with corresponding 95% con dence intervals (CIs), like sensitivity, specifcity, PLR, NLR, and DOR were calculated using STATA (version 12.0) and Meta-Disc (version 1.4) software [23]. The SROC curve was drawn to evaluate the diagnostic performance and to calculate the area under curve (AUC). The pooled AUC was used to describe the overall accuracy as a potential summary of the SROC curve. Meanwhile, Spearman correlation test was performed to quantify the threshold effect. Heterogeneity induced by non-threshold effect between the included studies was measured by the Cochran Q method and the inconsistency index (I 2 ) test. P value < 0.1 and I 2 value > 50% indicated a signi cant heterogeneity existing in our systematic review. Moreover, meta-regression analysis was conducted to nd out the sources of heterogeneity. At last, publication bias was determined by Deek' regression test. P value < 0.05 was considered to be statistically signi cant.

Study selection
Running the systematic search, a total of 121 citations were found: 35 in PubMed, 61 in Embase, 0 in Cochrane Library, 11 in Web of Science, and 14 in CNKI, respectively. As depicted in the flowchart ( Figure  1), 20 duplicate records were removed. Furthermore, after scanning the titles and abstracts, 83 records were discarded because they were irrelevant studies, abstracts, books, letters, meetings, or comments. Subsequently, 18 studies were identi ed for full-text reading. After reading the full-text of the 18 studies, 10 studies were excluded for the following reasons: two were not about cancer [24,25], ve did not refer to circulating PTX3[26-30], two articles did not report the diagnostic performance of circulating PTX3 in cancer detection [31,32], one paper did not present su cient data [33]. Finally, eight papers involving nine studies were included in this meta-analysis [12][13][14][15][16][34][35][36].

Studies characteristics
The main features of the nine included studies, such as name of the rst author, year of publication, country, cancer type, number of patients and controls, sensitivity and speci city, are presented in Table 1 by descending order of publication year. The included studies enrolled a total of 1408 cancer patients and 3116 control individuals. Among these studies, six were conducted in China [12,15,[34][35][36], two in Italy [13,14], and one in USA [16]. Besides, six researches investigated the diagnostic value of PTX3 for lung cancer, and one for gastric cancer, colorectal cancer and prostate cancer, each. The publication date of the included studies rang from 2011 to 2019. All these studies used the method of enzyme linked immunosorbent assay (ELISA) to detect the serum level of PTX3.

Quality assessment
To assess the quality of these nine eligible studies, a graph of risk of bias and applicability concerns was drawn. As shown in Figure 2, only one study [13] conformed to all the quality assessment items, and the bias of the eligible studies was mainly focused on "patients selection" and "index test". Specially, in the domain of patient selection, 5 studies[12, 14-16] did not state whether the study avoided inappropriate exclusions. Meanwhile, in the domain of index test, 6 studies [14,15,[34][35][36] did not adopt blinded method and one study [12] was unclear if blinded method was used. Generally, all the included nine studies could be considered to be of upper quality.

Threshold effect analysis and Heterogeneity exploration
Spearman's rank correlation analysis was carried out to evaluate the heterogeneity caused by threshold effect between the included studies (run on MetaDisc, version 1.4). Spearman correlation coe cient was -0.233 and the P-value was 0.546 (p > 0.05), indicating that threshold effect was not the source of heterogeneity in this study. Nevertheless, heterogeneity indeed existed as the I 2 of SE and SP were 83% and 87.4%, separately. Thereupon, meta-regression analysis was conducted to explore the sources of heterogeneity (performed on MetaDisc, version 1.4). We considered etnnicity, assay kit, sample size, and cancer type as the most possible factors causing heterogeneity. However, no satisfactory clues were observed. The results of meta-regression analysis are listed in Table 2.

Publication bias
Deeks' funnel plot asymmetry analsysis (tested with STATA, version 12.0) was run to assess the potential bias between the nine eligible studies. As shown in Figure 7, no indication of publication bias was observed (t = 0.46; p > 0.05).

Discussion
Though traditional surgical specimens or biopsy tissues and imaging techniques are used in the cancer diagnosis, limitations still exist mainly due to the invasive procedures and the delayed re ection of tumor dynamic changes. Besides, cancer patients are usually found at the relatively late stages and the outcome is generally poor. Therefore, it is urgent to develop new methods with high performance for cancer diagnosis using minimally invasive procedures. PTX3, also named TSG14, has been known as a candidate novel marker of in ammation [37], especially cancer-related in ammation [38,39]. PTX3 involves in many crucial biological events, including cellular proliferation, angiogenesis, apoptosis, cancer cell invasion and metastasis [8,40]. Recent studies suggested PTX3 could be a potential serum biomarker for cancer screening [12][13][14][15][16]. However, to the best of our knowledge, there is no systematic evidence-based evaluation for PTX3 as a tumor marker in cancer detection. In this study, we performed a systematic review and meta-analysis to assess the diagnostic performance of circulating PTX3 in cancer diagnosis.
In this meta-analysis, a total of eight papers with 4524 individuals (1408 cancer patients and 3116 controls) were enrolled. Pooling the data from the included studies presented a combined SE of 0.70 (95% CI: 0.67 -0.72) and a combined SP of 0.77 (95% CI: 0.75 -0.78). Diagnostic odds ratio (DOR), a merged power of SE and SP, is a metric of overall diagnostic performance. A higher DOR, a better diagnostic accracy. The global DOR of PTX3 in this study is 7.38 (95% CI: 5.05 -10.78), implying that PTX3 was of good diagnostic performance in cancer screening. To further investigate the diagnostic performance, we drew the SROC curve and calculated the corresponding AUC. Generally speaking, an AUC value of 0.75 to 0.92 means a good overall diagnostic performance, 0.92 to 0.96 means very good , and 0.97 to 1 means excellent [41,42]. The AUC of SROC curve for PTX3 was 0.8 (95% CI: 0.76 -0.83), suggesting a good diagnostic accuracy. All of these results imply that PTX3 could be a novel promising diagnostic biomarker in cancer detection.
Our study has several advantages. Firstly, the computerised search did not involve any language restriction on the publications in order to enroll more suitable studies besides english version. Secondly, our study is the rst systematic review to verify the overall diagnostic accuracy of circulating PTX3 in cancer screening. Thirdly, PTX3 could be a potential biomarker for cancer with minimal invasiveness comparing to bioscopy and pathological examination.
However, our study also has some weaknesses. At rst, though meticulous comprehensive search was carried out in PubMed, Embase, Cochrane Library, Web of Science, China National Knowledge Infrastructure (CNKI), only nine eligible studies were included in this meta-analysis. Besides, substantial heterogeneity in sensitivity (I 2 = 83%) and speci city (I 2 = 87.4%) exists between the included studies.
According to the results of Spearman's rank correlation analysis (Spearman coe cient = -0.233, p value = 0.546 ), threshold effect was not the source of heterogeneity. Furthermore, we conducted meta-regression analysis to identify the factor(s) causing heterogeneity. Several covariates, like etnnicity, assay kit, sample size, and cancer type were adopted as the most possible source(s) of heterogeneity. Unfortunately, no statistical signi cant evidence was obtained. What's more, although there was no signi cant publication bias among the studies presented by Deeks' funnel plot asymmetry analsysis (Figure 7), some latent bias may still exists because positive results are likely to be reported.

Conclusions
In conclusion, this systematic review and meta-analysis provide convincing evidence that PTX3 is of considerable diagnostic accuracy and may be a potential biomarker in cancer detection. However, as the current diagnostic data are limited, more multicenter, well-designed high quality trials with large sample size should be done to strengthen the theory of PTX3 in cancer diagnosis.