Beginning in May 2025, a comprehensive electronic search was conducted across PubMed, EMBASE, and the Cochrane Library databases. The search utilized keywords such as “Neutrophil-to-Eosinophil Ratio” and “Neutrophil/Eosinophil Ratio.” The complete search syntax is available in the Supplementary Material . In addition to database retrieval, grey literature was reviewed via Google Scholar, and the reference lists of all eligible studies were manually screened for additional sources.

Studies were included if they met the following criteria: (1) enrolled patients diagnosed with cancer; (2) involved treatment with ICIs; (3) stratified patients into high and low NER groups; and (4) reported at least one relevant clinical endpoint—namely, overall survival (OS), PFS, objective response rate (ORR), or disease control rate (DCR). Studies were excluded if they were conference abstracts or commentary articles. When multiple publications reported on overlapping patient cohorts, only the version with the most complete dataset and rigorous methodology was included (15).

Key information was systematically extracted from each eligible study, including the first author’s name, year of publication, study period, geographic location, tumor classification, treatment strategy, sample size, participant demographics (such as age and sex), and the cutoff value. When available, hazard ratios (HRs) from multivariate analyses were preferred over those from univariate models (16).

The methodological quality of the included observational studies was assessed using the Newcastle–Ottawa Scale (NOS) for cohort studies. This scale evaluates three broad domains: (1) Selection of study groups (up to 4 points), including representativeness of the exposed cohort, selection of the non-exposed cohort, ascertainment of exposure, and demonstration that outcome of interest was not present at the start of the study; (2) Comparability of cohorts based on design or analysis (up to 2 points); and (3) Outcome assessment (up to 3 points), including assessment of outcome, follow-up duration, and adequacy of follow-up. Studies scoring more than six points were considered high quality. All steps were performed independently by two reviewers. Any discrepancies between reviewers were resolved through consultation with the senior author.

This study received approval from the institutional ethics committee. Owing to its retrospective design, the requirement for informed consent was waived. A historical cohort analysis was conducted involving patients diagnosed with hepatocellular carcinoma (HCC) who underwent treatment with ICIs combined with anti-angiogenic agents at our center between Mar 2019 and May 2023. Inclusion criteria mandated at least one measurable tumor lesion as defined by RECIST version 1.1 (17).

Clinical and demographic information was extracted from electronic medical records and included patient age, gender, Eastern Cooperative Oncology Group performance status (ECOG PS), underlying hepatitis type, presence of cirrhosis, Barcelona Clinic Liver Cancer (BCLC) stage, Child–Pugh score, number of lesions, macrovascular invasion status, treatment line, modified albumin–bilirubin (mALBI) grade, and serum alpha-fetoprotein (AFP) levels. Tumor response and progression were evaluated according to RECIST version 1.1 criteria. Follow-up imaging via CT was routinely conducted at intervals of one to three months after treatment initiation. PFS was defined as the time from the first dose of immune checkpoint blockade to radiographic progression or death, while OS was measured from treatment initiation to death from any cause.

Categorical data were expressed as absolute frequencies accompanied by their respective percentages. Survival outcomes across different subgroups were evaluated using the Kaplan–Meier estimator in conjunction with the Cox proportional hazards regression model. Meta-analytical computations were performed with Stata version 18.0, and the results were graphically summarized using forest plots. To quantify heterogeneity across included studies, both the I² statistic and Cochran’s Q test were applied. A heterogeneity level was considered significant if the I² exceeded 50% or the corresponding p-value was below 0.1 (18). When substantial variability was detected, the DerSimonian–Laird random-effects model was employed; otherwise, a fixed-effect model using the Inverse Variance method was applied.

Publication bias was investigated through Begg’s and Egger’s statistical tests (19). Sensitivity analyses were also conducted by sequentially omitting individual studies to assess the influence of each on the pooled HRs and overall effect estimates (20). Furthermore, subgroup analyses were conducted by stratifying data according to NER cutoff values and the type of Cox regression model used. A two-tailed p-value of less than 0.05 was considered to indicate statistical significance.

An initial search across the databases, supplemented by manual screening of reference lists, yielded a total of 208 potentially relevant records. Following the removal of 54 duplicate entries, 125 studies were excluded after evaluation of titles and abstracts, as they did not meet the predefined inclusion criteria. A full-text assessment of the remaining 32 articles resulted in the exclusion of 20 papers that failed to satisfy the eligibility standards. Consequently, 12 studies were ultimately included in the meta-analysis (10–14, 21–27) ( Figure 1 ).

Table 1 provides an overview of the key characteristics of the included studies. In total, 1,716 individuals were enrolled, with sample sizes ranging from 21 to 401 per study. Of the 12 studies, five were conducted in the USA and two in Japan. Five studies involved patients with renal cell carcinoma, two with urothelial carcinoma, and two were pan-cancer studies. All studies employed a retrospective design. Based on the NOS, quality scores ranged from 6 to 8, indicating a low risk of bias ( Table 1 ).

This meta-analysis incorporated 12 qualified studies involving a total of 1,716 patients to assess the prognostic relevance of the NER on OS in individuals receiving ICI therapy. The aggregated HR indicated a significant association between higher NER levels and poorer OS outcomes (HR = 1.82, 95% CI: 1.57–2.11, p < 0.001; Figure 2A ). Between-study variability was negligible, as reflected by Cochran’s Q test and an I² value (I² = 0, p = 0.444), supporting the use of a fixed-effects model.

Robustness of the pooled results was validated through sensitivity analysis, which involved the stepwise exclusion of each individual study. The overall estimates for OS remained consistent throughout this process ( Figure 2B ). Furthermore, assessments for publication bias using Begg’s and Egger’s tests revealed no statistically significant evidence of bias (Begg’s p = 0.118; Egger’s p = 0.108).

Subgroup analyses demonstrated that both univariate (HR = 1.71, 95% CI: 1.42–2.08, p < 0.001) and multivariate regression approaches (HR = 1.98, 95% CI: 1.57–2.49, p < 0.001) consistently revealed a statistically significant link between elevated NER values and reduced OS ( Figure 3 ). Moreover, this inverse association remained evident when the NER threshold exceeded 30 (HR = 1.91, 95% CI: 1.56–2.34, p < 0.001) or fell within the 20–30 range (HR = 1.81, 95% CI: 1.38–2.36, p < 0.001, Supplementary Figure S1 ). In contrast, when the cut-off point was below 20, NER failed to show prognostic value in predicting OS among cancer patients (HR = 1.67, 95% CI: 0.73–3.81, p = 0.227, Supplementary Figure S1 ).

This meta-analysis included nine eligible studies encompassing 1,504 patients to investigate the prognostic impact of the NER on PFS among individuals treated with ICIs. The pooled HR indicated a strong correlation between elevated NER and unfavorable PFS outcomes (HR = 1.62, 95% CI: 1.34–2.97, p < 0.001; Figure 4A ). Substantial heterogeneity was observed among the studies (I² = 41.8%, p = 0.089), warranting the use of a random-effects model.

Sensitivity analysis, performed through sequential removal of each study, confirmed the stability of the combined PFS estimates ( Figure 4B ). In addition, assessments for potential publication bias using Begg’s and Egger’s tests did not indicate statistical evidence of asymmetry (Begg’s p = 0.129; Egger’s p = 0.145).

Subgroup analyses further supported the association between high baseline NER and reduced PFS, with consistent results observed in both univariate (HR = 1.90, 95% CI: 1.32–2.75, p < 0.001) and multivariate models (HR = 1.46, 95% CI: 1.20–1.79, p < 0.001, Figure 5 ). Notably, this negative prognostic relationship was maintained when the NER cut-off was greater than 30 (HR = 1.66, 95% CI: 1.31–2.10, p < 0.001) or within the 20–30 range (HR = 1.62, 95% CI: 1.07–2.45, p < 0.001, Supplementary Figure S2 ).

Our study further investigated the association between NER and ORR, incorporating data from six studies involving a total of 973 cancer patients. As no significant heterogeneity was observed across these studies, a fixed-effect model was applied (I² = 42.4%, p = 0.123). The meta-analysis demonstrated that patients with elevated NER had a significantly lower ORR compared to those in the low-NER group (OR = 0.50, 95% CI: 0.37–0.68, p < 0.001; Figure 6A ).

Sensitivity analysis confirmed the robustness of this finding, as sequential exclusion of individual studies did not materially alter the overall effect estimate ( Figure 6B ). Assessments for publication bias using Begg’s and Egger’s tests revealed no statistically significant evidence of bias (Begg’s p = 0.328; Egger’s p = 0.428). Moreover, subgroup analysis based on different NER cut-off values consistently supported the observed association, indicating that the inverse relationship between NER and ORR remained stable across various threshold definitions ( Supplementary Figure S3 ).

The relationship between NER and DCR among cancer patients was analyzed based on four studies comprising 759 individuals. As the analysis revealed no significant heterogeneity (I² = 0, p = 0.586), a fixed-effects model was deemed appropriate. The aggregated findings demonstrated that higher NER levels were significantly correlated with a lower DCR compared to patients with lower NER values (OR = 0.44, 95% CI: 0.31–0.61, p < 0.001; Figure 7A ).

To assess the reliability of these results, sensitivity analyses were performed by systematically excluding each study. The consistency of the effect estimates across iterations confirmed the robustness and stability of the pooled outcome for DCR ( Figure 7B ).

In view of the limited literature addressing the prognostic relevance of the NER in hepatocellular carcinoma (HCC), we conducted an analysis using patient data from our institution to enhance current insights into NER as a prognostic biomarker in oncology.

Supplementary Table 1 presents the baseline demographic and clinical profiles of the 67 patients with HCC included in our cohort. The median age was 58.2 years, with an age range spanning from 40.2 to 81.23 years. A predominance of male participants was observed, accounting for 67.16% (n = 45) of the total. Regarding performance status, 62.69% (n = 42) had an ECOG PS score of 0, while the remaining 37.31% (n = 25) had a score of 1. Chronic viral hepatitis was documented in 77.61% (n = 52) of the cohort, and cirrhosis of the liver was present in 67.16% (n = 45). According to the BCLC staging system, 5.97% (n = 4) were categorized as early stage, 41.79% (n = 28) as intermediate stage, and 52.24% (n = 35) as advanced stage. Microvascular invasion was identified in 35.82% (n = 24) of patients, and AFP levels exceeded 400 ng/mL in 58.21% (n = 39) of cases.

Patients were stratified into two subgroups according to the median baseline NER threshold. Kaplan–Meier analysis demonstrated that individuals with higher NER exhibited markedly reduced OS (p = 0.006) as well as PFS (p = 0.033) when compared to those with lower NER values ( Figure 8 ).