Prognostic value of YKL-40 in solid tumors: a meta-analysis of 41 cohort studies

We conducted an extensive literature search for relevant studies from the Embase (from 1974 to March 08, 2019), Medline (from 1966 to March 08, 2019) and Web of Science databases (from 1985 to March 08, 2019). The search strategy included the following keywords: “YKL-40”, “YKL40”, “CHI3L1”, “chitinase-3-like-1”, “GP-39”, “glycoprotein-39”, “CGP-39”, “cartilage glycoprotein-39”, “hCGP-39”, “human cartilage glycoprotein-39”, “tumor”, “neoplasm” and “cancer”. Reports in English were eligible for inclusion. Furthermore, we manually reviewed the relevant articles to implement our search.

Studies were included in the meta-analysis according to the following criteria: (1) patients with a diagnosis of solid tumor that was confirmed through histopathologic examinations; (2) sufficient data were provided to determine an estimate of the hazard ratio (HR) for OS and disease-free survival (DFS)/progression-free survival (PFS); (3) more than 30 patients were enrolled in each study; and (4) cohort studies published in English. When the same patient population was used in multiple publications, only the latest was included in the meta-analysis. Reviews and comments were excluded.

Two investigators independently assessed the quality of the eligible studies. The Newcastle–Ottawa Scale (NOS) was used to assess study quality. The NOS is based on three parameters of quality: selection (0–4 points), comparability (0–2 points), and outcome assessment (0–3 points). The scores ranged from zero points (worst) to nine points (best). Disagreements on the quality assessment were resolved by discussion. We also conducted sensitivity analyses to assess the stability of the results.

We extracted the following information from each study: author’s name, publication year, patients’ country, cancer type, number of patients, tumor stage, metastasis status, treatment methods, YKL-40 cutoff value, specimen type, and HR with 95% CI for DFS, PFS and OS. If the HRs and 95% CIs were not directly collected from the original article, they were calculated by the method of Parmar et al. [12].

The pooled HR or odds ratio (OR) was initially calculated by a fixed effect model [13]. If the I² statistic was more than 30% or the fixed effects P value for the I² statistic was less than 0.10, indicating significant heterogeneity across studies, then a random effect model was applied to calculate the pooled HR or OR [14]. To explore the potential causes of heterogeneity, subgroup analyses and meta-regression analyses were conducted (HR was considered to be associated with covariates when the P value was less than 0.05). To evaluate the stability of the results, we conducted a sensitivity analysis by omitting each study in turn. The scope of this analysis was to evaluate the influence of a single study on the overall outcome. To investigate publication bias in the literature, we performed a visual inspection of the funnel plot symmetry and Begg’s regression and Egger’s linear regression tests (P < 0.05 was statistically significant) [15, 16]. The nonparametric trim-and-fill method was performed to further assess the potential effect of publication bias [17]. All statistical analyses were performed using the Meta package in R software (version 3.5.1, The R Project for Statistical Computing).

The flowchart of the study selection process is shown in Fig. 1. The literature search yielded a total of 1440 records. Of these articles, 502 studies were retrieved for detailed evaluation. In total, 461 studies were further excluded because of there was YKL-40 protein in the tumor tissue, the study was meta-analysis, the study involved leukemia or lymphoma, there were multiple publications, there was insufficient data, there was no survival data or the data was not relevant to tumor tissues. Finally, 41 articles [9, 18‐57] met the selection criteria after reviewing the abstracts or full texts.

The characteristics of the eligible studies are summarized in Tables 1 and 2. The sample size in each study ranged from 37 to 1432 patients, and a total of 7762 patients were included in the meta-analysis. These studies were published from 1995 and 2019. 32 studies were from Caucasian countries [9, 18‐32, 35‐37, 41, 42, 44, 45, 47‐49, 51‐56], and the other 9 studies were from Asian countries [33, 34, 38‐40, 43, 46, 50, 57]. The proportion of stage IV(D) or stage III(C) disease and the proportion of metastatic or extended disease were used as the proportion of advanced stage disease in 20 studies [18‐20, 25, 26, 28, 29, 31, 33‐35, 37, 38, 40, 43, 45, 48‐50, 57] and 10 studies [21, 23, 24, 32, 44, 46, 47, 52, 54, 56], respectively. The proportion of advanced stage disease was not reported in the other 11 studies [9, 22, 27, 30, 36, 39, 41, 42, 51, 53, 55]. Johansen et al. [27] investigated the mid-therapy value of serum/plasma YKL-40, and Bernardi et al. [37] investigated the value of the serum/plasma YKL-40 ratio [(1 week value-baseline value)/baseline value]. The other studys investigated the baseline value of serum/plasma YKL-40 in solid tumors. The cutoff value of serum/plasma YKL-40 ranged from 44.6 to 247 µg/L in 24 studies [9, 18‐21, 24‐26, 33‐36, 39, 41‐43, 45, 46, 48, 50, 53‐56]. In total, 6 studies [27, 30, 40, 44, 47, 51] analyzed serum/plasma YKL-40 as a continuous variable, and the cutoff value was not reported in 11 studies [22, 23, 28, 29, 31, 32, 37, 38, 49, 52, 57]. A total of 39 studies [18‐54, 56, 57] reported HRs with 95% CIs for OS or DFS/PFS directly. The HR with a 95% CI for OS or DFS/PFS was estimated from Kaplan–Meier curves in the other 2 studies [9, 55]. The quality assessment of the trials is shown in Table 2.

The HRs for OS were available in 38 studies [9, 18‐32, 34‐42, 44‐56], and 2 HRs were extracted from 2 studies each because 2 cohorts were used in these studies. The forest plot of all studies is provided in Fig. 2. As heterogeneity among studies clearly existed (P < 0.01, I² = 82%), a random-effects model was applied. The pooled HR showed that elevated serum/plasma YKL-40 was significantly associated with poor OS (HR, 1.44; 95% CI 1.33–1.56).

To explore potential causes of heterogeneity, we performed meta-regression analyses using the following covariates: ethnicity, publication year, YKL-40 cutoff value, sample size, proportion of advanced stage disease, and specimen type, and treatment method(surgery vs. other treatment methods). The results indicated that ethnicity (P = 0.5611; Table 3), publication year (P = 0.4102), YKL-40 cutoff value (P = 0.5199), sample size (P = 0.3790), proportion of advanced stage disease (P = 0.2221), specimen type (P = 0.9164) and treatment method (0.7215) did not contribute to the cause of heterogeneity.

A visual inspection of the funnel plot revealed asymmetry. This was confirmed by Egger’s test (P < 0.01), although Begg’s test did not indicate statistical significance (P = 0.244). For this reason, we performed a trim-and-fill analysis and found that 19 studies might be missing (Fig. 3). When these studies were published, the adjusted HR was 1.13 (95% CI 1.05–1.22; P < 0.01, I² = 85%; Additional file 1: Figure S1), and the results continued to show a statistically significant association between serum/plasma YKL-40 and OS. The sensitivity analysis indicated that no individual studies significantly affected the overall outcome and demonstrated the stability of the results (Additional file 2: Figure S2).

A total of 10 studies comprising 2865 patients reported 12 HRs for OS in gastrointestinal tumors [18, 30, 39‐41, 44, 47, 50, 52, 55]. Overall, elevated serum/plasma YKL-40 was associated with poor OS (HR, 1.37; 95% CI 1.18–1.58; Fig. 4). As heterogeneity existed among studies (P < 0.01, I² = 66%), a random-effects model was applied.

In the subgroup analysis based on ethnicity, there was a significant relationship between elevated serum/plasma YKL-40 and poor OS in both the “Caucasian countries” subgroup and the “Asian countries” subgroup (HR, 1.28; 95% CI 1.11–1.48 and HR, 1.87; 95% CI 1.41–2.48, respectively; Additional file 3: Figure S3a). Statistical heterogeneity was significant in the “Caucasian countries” subgroup, whereas it was not significant in the “Asian countries” subgroup (I² = 66%, P < 0.01 and I² = 0%, P = 0.80, respectively).

In addition, we investigated the prognostic role of serum/plasma YKL-40 with respect to OS for patients who received different treatment methods. Patients received surgery alone in some studies and other treatment methods in the other studies. There was a significant relationship between elevated serum/plasma YKL-40 and poor OS in both the “surgery” subgroup and the “other treatment methods” subgroup (HR, 1.31; 95% CI 1.05–1.63 and HR, 1.43; 95% CI 1.14–1.80, respectively; Additional file 3: Figure S3b). Statistical heterogeneity was significant in the “other treatment methods” subgroup, whereas it was not significant in the “surgery” subgroup (I² = 79%, P < 0.01 and I² = 44%, P = 0.11, respectively).

To explore potential causes of heterogeneity, we performed meta-regression analyses using the following covariates: ethnicity, publication year, YKL-40 cutoff value, sample size, proportion of advanced stage disease, specimen type, and treatment method. The results indicated that the only explanatory variable that influenced HR was ethnicity (P = 0.0407, Table 4) and that publication year (P = 0.5750), YKL-40 cutoff value (P = 0.0908), sample size (P = 0.6562), proportion of advanced stage disease (P = 0.4457), specimen type (P = 0.4700), and treatment method (P = 0.6596) did not contribute to the cause of heterogeneity.

A visual inspection of the funnel plot did not reveal asymmetry (Additional file 4: Figure S4). This was confirmed by Egger’s test (P = 0.1129) and Begg’s test (P = 0.337). The sensitivity analysis indicated that no individual studies significantly affected the overall outcomes and demonstrated the stability of the results (Additional file 5: Figure S5).

As shown in Fig. 5, the prognostic effect of serum/plasma YKL-40 was highest in ovarian cancer (HR, 2.27; 95% CI 1.69–3.06; P = 0.68, I² = 0%; Fig. 6a), followed by melanoma (HR, 1.77; 95% CI 1.18–2.67; P = 0.03, I² = 65%; Fig. 6b), lung cancer (HR, 1.73; 95% CI 1.35–2.23; P = 0.42, I² = 0%; Fig. 6c), urologic neoplasms (HR, 1.61; 95% CI 1.08–2.40; P < 0.01, I² = 81%; Fig. 6d) and glioblastoma (HR, 1.23; 95% CI 1.07–1.42; P = 0.35, I² = 4%; Fig. 6e); in contrast, the prognostic effect of serum/plasma YKL-40 was not statistically significant in breast cancer (HR, 1.07; 95% CI 0.98–1.17; P < 0.01, I² = 83%; Fig. 6f).

In total, 10 HRs for DFS/PFS were available in 10 studies [21, 33, 38, 42‐44, 46, 49, 55, 57]. The estimated pooled HR for these studies showed a high risk of disease progression in patients with elevated serum/plasma YKL-40 (HR, 1.11; 95% CI 1.02–1.22; Additional file 6: Figure S6). As heterogeneity existed among studies (P < 0.01, I² = 68%), a random-effects model was applied.

To explore potential causes of heterogeneity, we performed meta-regression analyses using the following covariates: ethnicity, publication year, YKL-40 cutoff value, sample size, proportion of advanced stage disease, and specimen type and treatment method. The results indicated that the only explanatory variable that influenced HR was the YKL-40 cutoff value (P = 0.0017; Additional file 7: Table S1) and that ethnicity (P = 0.9445), publication year (P = 0.6929), sample size (P = 0.0538), proportion of advanced stage disease (P = 0.8162), and specimen type (P = 0.4427) and treatment method (P = 0.3932) did not contribute to the cause of heterogeneity.

Several studies investigated the associations between YKL-40 and clinicopathological parameters. Eight studies reported related data. Of these, 4 studies reported the association between YKL-40 and tumor stage (III-IVvs.I-II, C-D vs. A-B or extended vs. limited); 3 studies reported the association between YKL-40 and metastasis status (lymph node or liver metastasis vs. no metastasis); and 1 study reported both. Pooled outcome from five studies demonstrated a strong association between YKL-40 and clinical stage(OR, 1.47; 95% CI 1.02–2.12; Additional file 8: Figure S7a). Unfortunately, similar association was not observed between YKL-40 and metastasis status (OR, 2.14; 95% CI 0.89–5.14; Additional file 8: Figure S7b) in 4 studies.