Prognostic impact of pretreatment lymphocyte-to-monocyte ratio in advanced epithelial cancers: a meta-analysis

This meta-analysis was conducted in line with the preferred reporting items for systematic reviews and meta-analyses (PRISMA) statement [39]. Studies were identified by searching databases including PubMed, EMBASE, Web of science and Cochrane Library up to June 2018 without language restrictions. The full search strategies are presented in Additional file 1: Table S1. The reference lists of the previously published meta-analyses were also manually reviewed until no additional potential articles could be identified.

The identified studies were selected by two independent reviewers (Mao and Chen). First, the titles and abstracts were screened to assess study the eligibility, and then the full text was reviewed. Any disagreement was resolved by discussion or by a third reviewer (Duan) to reach a consensus. Studies meeting that met the following criteria were included: (1) Studies involving individuals with advanced-stage epithelial tumors and concerning the prognostic value of the pretreatment LMR. The definition of “advanced stage” was derived from the original research from which we extracted data. The timing of assessment of LMR was set at baseline before any treatment was initiated. (2) Studies providing the hazard ratio (HR) with a 95% confidence interval (CI) for overall survival (OS) or progression-free survival (PFS), or indirect information such as Kaplan–Meier curves used to estimate survival data on the basis of the methods previously described [2, 40‐42]. (3) If the same population was included in two or more studies, only the one study with the largest sample size or the latest information was included. (4) The full text was available. The exclusion criteria were as follows: (1) Non-human research; (2) Case reports, reviews, comments, editorials, letters or conference abstracts; (3) Patients with mesenchymal tumors or hematologic malignancies; (4) Insufficient data for estimating a HR and 95% CI; (5) LMR included only as a continuous variable rather than a dichotomized variable.

Two reviewers (Mao and Chen) independently carried out the data extraction from the eligible studies. The following information was recorded for each study: first author’s name, year of publication, research region, inclusion period, study design, number of patients, patient age, tumor type, tumor stage, treatment, cut-off value of LMR, time of LMR assessment, follow-up period, study endpoints, analysis of hazard ratios and adjustment variables. The individual HR (with the corresponding 95% CIs) in the studies was also extracted for OS and PFS to assess the therapeutic efficacy. The HRs were preferentially extracted from multivariate analyses. Any discrepancies between reviewers were resolved by consensus. As the previous studies reported [2, 43, 44], a set of modified predefined criteria was applied to assess the risk of bias of the included studies. The modified predefined criteria are shown in Additional file 1: Table S3. Studies with a score of 7 or higher were defined as high-quality, and those with a score whereas scores of less than 7 were considered low-quality.

General data were analyzed using Statistical Package for Social Sciences (SPSS) software (version 21.0 for Windows). STATA 12.0 software (StatCorp, College Station, TX, USA) was used to conduct the meta-analysis. Cochran’s Q test and Higgins I-squared statistic were used to test the heterogeneity of different studies. A P value of less than 0.1 was considered significant. I²> 50% was deemed to show substantial heterogeneity [45]. When the heterogeneity was significant, a random-effect model was applied; otherwise, a fixed-effect model was used. Summary HRs were calculated according to the appropriate model depending on the heterogeneity of the included studies. The reasons for inter-study heterogeneity were explored using subgroup analysis. Sensitivity analysis was also conducted by omission of each single study to evaluate the stability of the results. Publication bias was assessed using funnel plots, Begg’s and Egger’s tests [46, 47]. When publication bias was suggested, Duval and Tweedie trim-and-fill methods were applied for the number of missing studies, and the pooled estimate was recalculated to adjust the primary results [48]. All statistical tests were two-sided, and statistical significance was defined as P less than 0.05.

The flow chart of the literature search is shown in Fig. 1. In summary, our search strategy identified 1613 studies after searching the relevant online databases. We excluded 223 duplicate records from the initial studies. After screening the title and abstracts of 1390 studies, 1171 studies were removed, and another 181 articles were excluded after the assessment of full text. Finally, 35 studies [3‐11, 15‐38, 49, 50] met our inclusion criteria that were selected for the present meta-analysis.

These studies included a total of 8984 patients with a median age of 60.6 years and a median follow-up period of 26.8 months. Table 1 and Additional file 1: Table S2 provide the basic and summarized characteristics of the identified studies that met the inclusion criteria. In summary, all studies had a retrospective study design and were published between 2014 and 2018. 12 different kinds of epithelial tumors were included in these studies, and the median number of patients was 177. Colorectal cancer and lung cancer were the two main types of cancers. The main therapeutic strategies included chemotherapy, surgery and combined therapy. 28 studies were conducted in Asia, 5 in Europe and 2 in America and others. The association between pretreatment LMR and OS was investigated in all the included studies, among which 9 also investigated the association between pretreatment LMR and PFS as well. The median cut-off value for LMR was 3.23. Most of the included studies (32/35) used multivariate analysis method to adjust covariates when analyzing the prognostic value of LMR. According to the risk assessment scale, 3 studies had quality scores less than 7, the other 32 had a score more than 7 (Additional file 1: Table S3).

35 studies with 8984 individuals were included in the analysis of pretreatment LMR and OS. Figure 2a indicates that a higher pretreatment LMR was associated with improved OS (HR = 0.578, 95% CI 0.522–0.641, P < 0.001). Given that the test for heterogeneity was significant (Q =113.56, P < 0.001, I² = 70.1%), a random-effect model was used. Subgroup analyses were applied to explore potential sources of heterogeneity among several related clinical features for OS (Table 2). The pooled HRs of most subgroups were markedly changed in subgroup analyses. The subgroup analysis by tumor types showed a higher pretreatment LMR was significantly associated with better OS in colorectal cancer (n = 13, HR = 0.579, 95% CI 0.516–0.650, I² = 0%), lung cancer (n = 5, HR = 0.594, 95% CI 0.435–0.811, I² = 85.5%), pancreatic cancer (n = 5, HR = 0.588, 95% CI 0.407–0.851, I² = 67.9%), gastric cancer (n = 2, HR = 0.664, 95% CI 0.523–0.843, I² = 0%), nasopharyngeal carcinoma (n = 2, HR = 0.479, 95% CI 0.406–0.566, I² = 0%), renal cancer (n = 2, HR = 0.827, 95% CI 0.755–0.906, I² = 0%), cervical carcinoma (n = 1, HR = 0.337, 95% CI 0.164–0.691), ovarian cancer (n = 1, HR = 0.615, 95% CI 0.527–0.718), esophageal cancer (n = 1, HR = 0.495, 95% CI 0.315–0.778) and head and neck cancer (n = 1, HR = 0.28, 95% CI 0.168–0.466), but not breast cancer (n = 1, HR = 0.47, 95% CI 0.171–1.295, P = 0.144) and hepatocellular carcinoma (n = 1, HR = 0.73, 95% CI 0.399–1.336, P  = 0.308). To be noted, the subgroup analysis by different therapeutic strategies indicated that a higher pretreatment LMR was associated with improved OS in chemotherapy (n = 10, HR = 0.592, 95% CI 0.518–0.676, P < 0.001), surgery (n = 10, HR = 0.683, 95% CI 0.579–0.807, P < 0.001), combined therapy (n = 11, HR = 0.507, 95% CI 0.442–0.582, P < 0.001) which consists of surgery and (neo)adjuvant therapy. The cut-off values of LMR in the studies ranged from 2.35 to 5.46. After stratifying the cut-off values of LMR into two subgroups, < 3.0 and ≥ 3.0, we noted that the level of statistical heterogeneity (< 3.0, I² = 16%; ≥ 3.0, I² = 69.8%) was reduced, while the pooled HRs were not significantly altered. The reduction in statistical heterogeneity was also realized after adjusting research region (Asia, I² = 74.9%; Europe, I² =  0%; America and others, I² = 0%), number of cases (< 200, I² = 63.2%; > 200, I² = 41.3%), therapeutic strategies (Chemotherapy, I² = 31.9%; Molecular targeted, I² = 93%; Surgery, I² = 54.3%; Combined therapy, I² = 37.9%; others, I² = 0%) and follow-up period (≤ 33, I² = 67.8%; > 33, I² = 15.6%; NR, I² = 81.1%). Meanwhile, the subgroup analysis by publication year, initial inclusion period, median age, quality score and analysis of HR indicated that a high pretreatment LMR was consistently associated with superior OS.

Sensitivity analysis on the stability of the OS subset indicated that omitting any single study did not significantly affect the pooled HRs (Fig. 3a). As shown in Additional file 2: Figure S1A, the asymmetrical funnel plot suggested that there could be publication bias. It was further confirmed with Egger’s test (Begg’s test, P = 0.334; Egger’s test, P < 0.001). The adjusted random effects pooled HRs of 0.578 (95% CI 0.522–0.641), obtained using the trim-and-fill method, which was consistent with our primary analysis (Additional file 1: Table S4). The funnel plot adjusted with trim-and-fill methods was shown in Additional file 2: Figure S1B.

Nine studies with 2694 individuals were included in the analysis of pretreatment LMR and PFS. Figure 2b demonstrates that a high pretreatment LMR was associated with longer PFS (HR = 0.598, 95% CI 0.465–0.768, P < 0.001). Since the test for heterogeneity was significant (Q = 72.92, P < 0.001, I² = 89.0%), a random-effect model was used. Table 3 gives the results of subgroup analyses on potential sources of heterogeneity among several related clinical features of the included studies for PFS. The subgroup analysis by tumor types indicated that a higher pretreatment LMR was significantly associated with better PFS in colorectal cancer (n = 2, HR = 0.695, 95% CI 0.562–0.861, I² = 0%), cervical carcinoma (n = 1, HR = 0.239, 95% CI 0.151–0.379), ovarian cancer (n = 1, HR = 0.581, 95% CI 0.508–0.664) and esophageal cancer (n = 1, HR = 0.461, 95% CI 0.31–0.685) but not lung cancer (n = 4, HR = 0.738, 95% CI 0.54–1.007, I² = 80.00%, P = 0.056). A higher pretreatment LMR was proved to be associated with improved PFS in the subgroup analysis by different therapeutic strategies including chemotherapy (n = 4, HR = 0.62, 95% CI 0.558–0.688, P < 0.001) and combined therapy (n = 3, HR = 0.415, 95% CI 0.241–0.716, P = 0.002). The cut-off values of LMR ranged from 3.11 to 5.28 in different studies. The pooled HRs were not significantly altered by stratifying the cut-off values of LMR into 2 subgroups: ≤ 4.0 and > 4.0, which decreased the level of statistical heterogeneity (≤ 4.0, I² =  0%; > 4.0, I² = 92.2%) nonetheless. It was noted that the significant difference was altered in subgroup analysis by number of cases (< 200, P = 0.075), therapeutic strategies (molecular targeted, P = 0.384), follow-up period (NR, P = 0.356), quality score (< 7, P = 0.356) and analysis of hazard ratio (Univariate, P = 0.061). Sensitivity analysis further confirmed that omitting any single study did not significantly affect the pooled HRs, exhibiting good stability of PFS subset (Fig. 3b).