The prognostic role of lymphocyte-to-monocyte ratio in patients with resectable pancreatic cancer: a systematic review and meta-analysis

Literature search

This systematic analysis adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines (Page et al., 2021) and relied on data registered in PROSPERO under the reference CRD42023461260. The PRISMA 2020 checklist can be found within Table S1.

In September 2023, a systematic literature search of English-language publications was conducted using PubMed, Embase, Cochrane, and Web of Science to investigate the prognostic role of the lymphocyte-to-monocyte ratio in resectable pancreatic cancer. Detailed search methods are outlined in Table S2.

Additionally, a manual review of the reference lists of all eligible studies was performed. Two investigators independently conducted searches for nested reports within primary studies (Haipeng Li and Shang Peng). If contrasting opinions arose, the two investigators discussed until a consensus was reached. In cases where a consensus was not reached, a third investigator (Ran An) was included to facilitate resolution.

Identification of eligible studies

The final analysis included studies that fulfilled the following criteria: (1) were randomized controlled trials, cohort, or case-control studies; (2) involved adult patients with pancreatic cancer; (3) reported LMR; (4) involved patients eligible for surgical removal; and (5) had sufficient data available to calculate HR.

The following types of publications were excluded: reviews, letters, editorial comments, case reports, conference abstracts, pediatric articles, unpublished articles, and articles not in English. Additionally, patients with advanced pancreatic cancer were excluded as surgical removal is not feasible for this patient population. For studies that provided LMR data, a reciprocal transformation method was applied, as necessary.

Data extraction

Data extraction was independently conducted by two investigators. In cases of discordance, a third investigator was consulted to make a final determination. The following data was extracted from the studies included in the analysis: primary author, publication year, study duration, study location, sample size, participant age, TNM stage, LMR threshold, body mass index (BMI), NOS, and patient eligibility for surgical removal. In instances where the study reported continuous variables as a median with a range or interquartile range, a validated mathematical approach was applied to compute the mean ± standard deviation (Wan et al., 2014; Luo et al., 2018). For the studies with absent or unreported data, the investigators proactively reached out to the corresponding authors for the provision of complete data if it was accessible. Detailed information can be found in Table 1.

Quality assessment

The quality of the incorporated studies was evaluated using the Newcastle-Ottawa Scale (NOS) (Gurol-Urganci et al., 2013), with studies receiving seven to nine points considered to be high-quality studies and four to six points considered to be moderate-quality studies (Kim et al., 2019). Additionally, the level of evidence for each study was assessed following the guidelines established by the Oxford Centre for Evidence-Based Medicine Levels of Evidence Working Group. Two investigators independently appraised the quality of each eligible study and assigned it an evidence level. Any disparities were resolved through discussion between the two investigators. Detailed NOS results can be found in Table S3.

Statistical analysis

An evidence synthesis was performed using Review Manager version 5.3, developed by the Cochrane Collaboration in Oxford, UK. Hazard ratio (HR) was used for comparing continuous and dichotomous variables. All metrics were presented with corresponding 95% confidence intervals (CIs). The chi-squared test (Cochran’s Q) and the inconsistency index (I²) were used to assess the heterogeneity of the included studies (Higgins & Thompson, 2002). Significant heterogeneity was indicated by an χ p-value < 0.05 or I² > 50%. In cases where significant heterogeneity was identified, a random-effects model was applied for the estimation of the combined HR; otherwise, a fixed-effect model was applied. One-way sensitivity analyses were also performed to gauge the impact of individual studies on the combined results for outcomes characterized by significant heterogeneity. Additionally, for outcomes that encompassed ten or more included studies, Egger’s regression tests were performed using Stata version 15.0 (Stata Corp, College Station, TX, USA). A p-value < 0.05 was considered statistically significant publication bias (Egger et al., 1997).

Literature search and study characteristics

The systematic search and selection process is presented as a flowchart in Fig. 1. The systematic literature search yielded a total of 355 relevant articles, with contributions from various sources including PubMed (n = 103), Embase (n = 143), Cochrane (n = 3), and Web of Science (n = 106). After duplicate publications were removed, a total of 224 titles and abstracts were reviewed for inclusion. Ultimately, 14 full-text articles, encompassing data from 4,019 patients, were included in the final pooled analysis and among them, the study by Ueberroth et al. adjusted for sex, age, and stage at diagnosis, while the remaining studies did not make adjustments (Stotz et al., 2015; Li et al., 2016; Sierzega et al., 2017; Kawai et al., 2019; Onoe et al., 2019; Pointer et al., 2020; Takano et al., 2020; Takeuchi et al., 2020; Zhou et al., 2020; Fang et al., 2021; Markus et al., 2021; Ueberroth et al., 2021; Kubota et al., 2022; Neumann et al., 2023). It is noteworthy that all 14 included studies used a retrospective cohort study design. Of the final 14 articles, eight originated from Asia, four from Europe, and two from the Americas. LMR values ranged between 1.6–3.3, most of the thresholds were around three, and the NOS scores ranged between 6–8 points. LMR was a robust predictor of OS, demonstrating strong prognostic significance (HR = 0.55, 95% CI [0.44–0.69], I² = 79%, P < 0.00001). This predictive significance extended to various types of pancreatic cancer, such as pancreatic ductal adenocarcinoma (HR = 0.73, 95% CI [0.57–0.93], I² = 46%, P = 0.01), pancreatic neuroendocrine neoplasms (HR = 0.81, 95% CI [0.66–0.99], P = 0.04) and other subtypes (HR = 0.40, 95% CI [0.22–0.72], I² = 89%, P < 0.00001), but not to pancreatic head cancer (HR = 0.46, 95% CI [0.16–1.13], I² = 59%, P = 0.12). LMR retained its predictive value across different regions, including Asia (HR = 0.62, 95% CI [0.47–0.76], I² = 68%, P < 0.0001), Europe (HR = 0.78, 95% CI [0.67–0.91], I² = 0%, P = 0.002), and the Americas (HR = 0.14, 95% CI [0.08–0.24], I² = 0%, P < 0.00001). Notably, both LMR cut-off values greater than or equal to three (HR = 0.62, 95% CI [0.47–0.82], I² = 67%, P = 0.0009) and less than three (HR = 0.47, 95% CI [0.32–0.69], I² = 85%, P = 0.0001) exhibited prognostic significance. After categorizing the 14 studies based on NOS scores as ≥7 (HR = 0.66, 95%CI [0.54–0.80], I² = 53%, P = 0.04) and <7 (HR = 0.41, CI [0.23–0.72], I² = 89%, P < 0.00001), the results showed that LMR had a good predictive value in both subgroups. The sensitivity analysis for OS confirmed the strong predictive value of LMR, whereas LMR did not exhibit predictive significance for RFS (HR = 0.35, 95% CI [0.09–1.32], I² = 95%, P = 0.12). The subgroup analysis results of LMR for resectable pancreatic cancer are shown in Table 2.

Overall survival

Overall Survival (OS) data was synthesized from a total of 14 studies, encompassing 4,019 patients. The pooled analysis revealed a statistically significant result, indicating that LMR plays predictive role in individuals diagnosed with resectable pancreatic cancer. When a patient’s LMR surpassed the designated threshold, it was associated with a more positive prognosis, leading to an extended OS. Conversely, when a patient’s LMR value fell below the LMR threshold, the patient’s prognosis tended to be less favorable, leading to a shorter OS (HR = 0.55, 95% CI [0.44–0.69], I² = 79%, P < 0.00001; Fig. 2).

Subgroup analysis of OS: population

To further assess the prognostic significance of LMR for overall survival in various subtypes of pancreatic cancer, the 14 articles were divided into four different subtypes of pancreatic cancer for detailed investigation: pancreatic ductal adenocarcinoma, pancreatic neuroendocrine neoplasms, pancreatic head cancer, and all other types of pancreatic cancer (Fig. 3). In the pancreatic ductal adenocarcinoma group, five studies were included with statistically significant results (HR = 0.73, 95% CI [0.57–0.93], I² = 46%, P = 0.01). There were two studies in the pancreatic head cancer group, and the results were not statistically significant (HR = 0.46, 95% CI [0.16–1.13], I² = 59%, P = 0.12). The pancreatic neuroendocrine neoplasms group only included one study, but the results were statistically significant (HR = 0.81, 95% CI [0.66–0.99], P = 0.04). Finally, the group of all other types of pancreatic cancer included six studies with statistically significant results (HR = 0.40, 95% CI [0.22–0.72], I² = 89%, P < 0.00001). The Egger’s test indicated publication bias within the group of pancreatic ductal adenocarcinoma (P = 0.041) but not in the group of other types of pancreatic cancer (P = 0.076). The sub-group analysis findings indicate that LMR exhibits a robust predictive capacity for OS in pancreatic ductal adenocarcinoma, pancreatic neuroendocrine neoplasms, and other types of pancreatic cancer. However, the results for pancreatic head cancer were not statistically significant.

Subgroup analysis of OS: region

A regional subgroup analysis was also performed based on the OS data from the 14 articles (Fig. 4). The studies were categorized into three major regions: Asia, the Americas, and Europe. There were eight articles from Asia, four articles from Europe, and two articles from the Americas. LMR showed prognostic significance in all three regional subgroup analyses: Asia (HR = 0.62, 95% CI [0.47–0.76], I² = 68%, P < 0.0001), the Americas (HR = 0.78, 95% CI [0.67–0.91], I² = 0%, P = 0.002), and Europe (HR = 0.78, 95% CI [0.67–0.91], I² = 0%, P = 0.002). The Egger’s test revealed publication bias in the Asia subgroup (P = 0.005) but not in the Europe subgroup (P = 0.242). This indicates that LMR plays a significant predictive role in pancreatic cancer patients across these three continents.

LMR threshold

LMR values were then extracted from the 14 articles and a threshold was set at three. This threshold value was chosen because eight of the 14 articles had LMR values greater than or equal to three, while six articles had LMR values less than three, making the groups close to equal. Separate analyses were performed for the eight articles and the six articles.

The results of these analyses showed that both the group with LMR values greater than or equal to three (HR = 0.62, 95% CI [0.47–0.82], I² = 67%, P = 0.0009) and the group with LMR values less than three (HR = 0.47, 95% CI [0.32–0.69], I² = 85%, P = 0.0001) had significant statistical significance. The Egger’s test results demonstrated publication bias both in the group with LMR values greater than or equal to 3 (P = 0.042) and in the group with LMR values less than 3 (P = 0.036).These findings imply that establishing an LMR threshold of three is a reliable way to effectively forecast the prognosis of patients with pancreatic cancer (Fig. 5).

NOS

In order to ensure a comprehensive analysis and maintain the integrity and representativeness of the data, we chose to include six studies of moderate quality. According to the Newcastle-Ottawa Scale (NOS) scores, five of these studies scored 6 points, while one study scored 5 points. To ensure the reliability of the moderate-quality studies, we categorized the articles based on Newcastle-Ottawa Scale (NOS) scores. Studies with NOS scores of 7 or higher were classified separately from those with scores below 7. Subgroup analyses were then conducted to compare the outcomes between these two groups.

The results revealed that among the 14 selected studies, both those with NOS scores greater than or equal to 7 (HR = 0.66, 95% CI [0.54–0.80], I² = 53%, P = 0.04) and those with scores less than 7 (HR = 0.41, CI [0.23–0.72], I² = 89%, P < 0.00001) demonstrated statistically significant findings. The Egger’s test results showed no publication bias in the group with NOS scores less than 7 (P = 0.127), while publication bias was detected in the subgroup with NOS scores greater than or equal to 7 (P = 0.002). These findings suggests that all 14 studies included in our analysis were reliable (Fig. 6).

RFS

Among the 14 included articles, three provided HR along with corresponding confidence intervals for RFS. These data were extracted for analysis, and the results indicated that the p-values associated with RFS did not achieve statistical significance (HR = 0.35, 95% CI [0.09–1.32], I² = 95%, P = 0.12). This result suggests that LMR may not be a robust predictor of RFS in this context (Fig. 7). However, it is important to acknowledge that the absence of statistical significance in these studies may stem from limited datasets, and there could be some uncertainty in the statistical outcomes. Notably, an Egger’s test did not raise significant concerns (p = 0.507).

Sensitivity analysis

A univariate sensitivity analysis was performed for OS by systematically excluding individual studies to assess the impact of each study on the prognostic significance of LMR (Fig. 8). The outcomes of the sensitivity analysis indicated that, regardless of the exclusion of any specific study, the statistical significance of LMR in predicting OS remained stable and unaltered. However, the results also indicated that (Ueberroth et al., 2021) may have some bias towards the upper confidence interval. This study tried to compare the high and low levels of LMR and NLR in Black and non-Black patients, which may have affected the confidence interval. However, the results of an Egger’s test showed a high degree of publication bias (p = 0.003).