Tumor Budding and Poor Prognosis in Oral Cancer: A Systematic Review and Meta-Analysis

Background: Tumor budding (TB) has been investigated in several types of solid tumors. In oral cancer, studies show its association with survival. However, for its implementation in routine histological analyses, results with a high certainty of evidence are needed. Therefore, the aim of this systematic review is to explore the association between tumor budding and overall survival (OS), disease-specific survival (DSS), and disease-free survival (DFS) in oral cancer. Methods: A search was performed in Embase, PubMed, Scopus, Livivo, Web of Science, and Google Scholar. We adopted the following inclusion criteria: studies that evaluate tumor budding in oral cancer, that investigate survival, and presenting cohort design. We excluded reviews and studies without hazard-ratio (HR) data. Results: This systematic review included 22 studies and showed an association between TB and survival. High-grade TB is associated with a worse OS in univariate analysis (HR = 3.11; 95% CI: 2.06-4.69, p<0.01) and multivariate analysis (HR = 2.62; 95% CI: 1.64-4.20, p<0.01); with a poorer DSS in univariate (HR = 2.43; 95% CI: 1.94-3.03, p<0.01) and multivariate analysis (HR = 2.01; 95% CI: 1.43-2.83, p< 0.01); and with a worse DFS in univariate (HR = 1.94; 95% CI: 1.44-2.62, p<0.01) and multivariate analysis (HR = 2.15; 95% CI: 1.31-3.53, p< 0.01). Sensitivity analysis showed that the results are robust, and no significant publication bias was identified in univariate analysis for DFS (Egger’s test: p = 0.94). The certainty of the evidence was graded as low or very low. Conclusion: Our findings indicate that TB is an independent prognostic factor of OS, DSS, and DFS in oral cancer. However, further studies are needed to increase the certainty of the evidence.


Introduction
Oral cancer is frequent between head and neck cancers, and oral squamous cell carcinoma (OSCC) is the most common subtype. This subtype counts for about 90% of oral cancers. Patients with this malignant neoplasm show high morbidity and poor prognosis, with 5-year survival of approximately 50% (Jensen et al., 2015;Bjerkli et al., In recent years, researchers have suggested a prognostic role of tumor budding (TB) in cancer. It is defined as the presence of single cancer cell(s) or cluster(s) of less than five cancer cells at the invasive front. TB represents the dissociation of invasive cancer cells from the main tumor mass (Almagunsh et al., 2018). The importance of TB in cancer prognosis has been studied widely in colorectal cancer (Rogers et al., 2016;Lugli et al., 2017). It was included as an additional prognostic factor in the TNM and World Health Organization (WHO) classification schemes only for colorectal cancer (in 2017 and 2019, respectively) (Lugli et al., 2021). Since then, TB has been investigated in lung , breast (Huang et al., 2022), endometrial (Ocal and Guzelis, 2022), esophageal (Almangush et al., 2016), gastric (Xue et al., 2023), pancreaticobiliary (Karamitopoulou et al., 2018), and head and neck cancers (Zhu et al., 2018). In addition, studies have shown the association between this parameter and survival in oral cancer (Wang et al., 2011;Sakata et al., 2018;Ebihara et al., 2019;Xie et al., 2019;Yu et al., 2019;Lugli et al., 2021).
TB can be analyzed through H&E staining, which makes this tool promising and completely applicable in daily histopathological practice. However, for a histopathological parameter to be included in routine histological analyses, studies must show that this parameter is a good prognostic factor with a high certainty of evidence. Therefore, this systematic review explores the association between tumor budding and survival in oral cancer.

Materials and Methods
This review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Page et al., 2021), and the protocol was registered in the International Prospective Register of Systematic Reviews (PROSPERO) under the registration number CRD42021275388.

Eligibility criteria
The focused question was: Is there an association between tumor budding and the prognosis of oral cancer patients? For the construction of the question, we used the PECOS strategy: 1. We established the following inclusion criteria: studies published in the Roman alphabet; that evaluate tumor budding in oral cancer patients; that investigate OS, DSS, or DFS; and cohort studies. There was no restriction regarding the time of publication. The following exclusion criteria were used: reviews, series and case reports, case-control, cross-sectional studies, in vitro and animal models, and studies that did not report hazard ratio (HR).

Information sources
The searches were performed in the following databases: PubMed, Embase, Web of Science, Scopus, and Livivo. An additional gray literature search was conducted on Google Scholar (the first 100 records were selected). The software EndNote Web was used to collect references and exclude duplicates. We conducted database searches on August 25, 2021; and updated it on November 16, 2022.

Search strategy
The following search strategy was used: ("Mouth Neoplasms" OR "Mouth Neoplasm" OR "Oral Neoplasm" OR "Oral Neoplasms" OR "Cancer of Mouth" OR "Mouth Cancers" OR "Oral Cancer" OR "Oral Cancers" OR "Cancer of the Mouth" OR "Mouth Cancer" OR "Oral Tongue Squamous Cell Carcinoma" OR "Oral Squamous Cell Carcinoma" OR "Oral Cavity Squamous Cell Carcinoma" OR "Oral Squamous Cell Carcinomas" OR "Squamous Cell Carcinoma of the Mouth") AND ("tumor budding" OR "tumor-budding" OR "tumour budding" OR budding) AND (Survival OR Prognosis OR Prognoses OR "Prognostic Factors" OR "Prognostic Factor" OR "overall survival" OR "hazard ratio" OR "disease-free survival" OR "Lymphatic Metastasis" OR "Lymphatic Metastases" OR "Lymph Node Metastasis" OR "Lymph Node Metastases"). Adaptations were performed in the present search strategy according to the characteristics of each database (Supplementary Table 1).

Study selection
Study selection was performed in two phases, using Rayyan software (Ouzzani et al., 2016). In the first phase, two independent reviewers (DFGO and RVO) screened titles and abstracts of articles. In the second phase, full-text articles were reviewed. The conflicts were solved with a third reviewer (SGF).

Data collection process
Two reviewers (DFGO and RVO) performed the data collection in an independent way. The conflicts were solved with a third reviewer (SGF). We collect data about author, publication year, country, staining method, gender, age, sample size, tumor site, histologic type, differentiation, TNM, clinical stage, follow-up, the cutoff point for high-grade tumor budding, HR, and confidence interval (OS, DSS, and DFS).

Risk of bias
The risk of bias assessment was conducted by two independent reviewers (DFGO and RVO), using the Joanna Briggs Institute critical appraisal checklist for cohort studies (Moola et al., 2020). A third examiner (SGO) resolved any disagreement. The overall risk of bias was judged according to the following: "high risk" when the study reached up to 49% score "yes"; "moderate" when the study reached 50%-69% score "yes"; and "low risk" if "yes" scores were more than 70% (Polmann et al., 2020). The risk of bias figure was generated using the statistical software R version 4.0.5. 21 studies evaluated squamous cell carcinoma (SCC) (95.45%). Most SCC was moderately differentiated (53.37%). Regarding tumor budding, 59.12% of patients had a low grade.

Risk of Bias
The principal methodological difficulty identified was related to the follow-up. However, overall, all studies were classified as low risk of bias (Supplementary Figure 1).

Synthesis of results Overall survival
Meta-analyses for OS are presented in Figure 2. High-grade TB was associated with a worse OS in univariate analysis (HR = 3.11; 95% CI: 2.06-4.69, p<0.01) and multivariate analysis (HR = 2.62; 95% CI: 1.64-4.20, p<0.01). The test for heterogeneity finds significant heterogeneity (I 2 = 73%, p < 0.01). In addition, we performed subgroup analyzes which showed reduced statistical heterogeneity when the HR was pooled by continent, country, cut-off, oral subsite, and staining method (Supplementary Table 3). To evaluate the stability of this meta-analysis, a sensitivity analysis was performed. The results demonstrated no significant changes in HR values (Supplementary Figure 2). Publication bias was not assessed, as less than ten studies were included in each meta-analysis.

Disease-free survival
Forest plots for DFS are presented in Figure 4. The pooled effect size estimate indicated that high-grade TB was significantly associated with a worse DFS in univariate (HR = 1.94; 95% CI: 1.44-2.62, p < 0.01) and multivariate analyses (HR = 2.15; 95% CI: 1.31-3.53, p<0.01). Sensitivity analysis showed that the results were stable (Supplementary Figure 4) and there was no significant publication bias in univariate analysis (Supplementary Figure 5, Egger's test: p = 0.94). However, the results had high heterogeneity. We also performed subgroup analyzes which demonstrated reduced statistical heterogeneity when the HR was pooled by continent, staining method, cut-off, oral subsite, country, and clinical stage (Supplementary Table 4).

Summary measures and synthesis of the results
We evaluated OS, DSS, and DFS in this systematic review. Therefore, HR was the effect measure adopted. For this purpose, we performed a meta-analysis of the pre-calculated data. We used the statistical software R version 4.0.5 (meta package) (Balduzzi et al., 2019). We adopted the random model as the included studies were from different populations. To pool the data, we used the inverse variance method and Restricted maximumlikelihood to estimate the tau² value. I², Q test, and tau² were investigated to evaluate the statistical heterogeneity. To explore the heterogeneity sources, subgroup analysis was performed based on the continent, country, cut-off, oral subsite, staining method, and clinical stage. In addition, sensitivity analysis was conducted by omitting one study at a time to evaluate the robustness of the pooled HR. We also investigated publication bias in the meta-analysis for DFS (univariate analysis) because the number of studies was more than 10. So, we used the funnel plot and Egger test. All the statistical tests were two-sided with a significance level of 5% (p<0.05).

Certainty of evidence
The certainty of evidence was evaluated through the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) . The GRADE categorizes evidence into 4 levels: high, moderate, low, and very low. Cohort studies are initially classified in low evidence, and some factors can downgrade the certainty of evidence (studies limitations, inconsistencies of the results, imprecision, indirect evidence, and publication bias) and others can upgrade the certainty of evidence (large effects, dose-response gradient, and effect of plausible residual confounding). A summary of the findings table was produced using the GRADEpro (McMaster University and Evidence Prime Inc.).

Selection of the Studies
A total of 559 registers were retrieved from databases and gray literature. After excluding duplicates, 373 studies were screened, including 75 records. Of these articles, 13 studies were conference abstracts (reports not retrieved). Therefore, 62 full-text articles were reviewed, including a total of 22 studies in this systematic review (Figure 1).

Studies Characteristics
This systematic review included 22 studies, conducted in China (5), Japan (5), Brazil (2), Denmark (2), Norway (1), USA (2), India (1), South Korea (2), Italy (1), and one with samples from two countries (Brazil and Finland). These were published between 2011 and 2022, including a total of 3504 patients with a mean age of 58.94 years (Table 1, Table 2, and Supplementary Table 2). The mean follow-up was 60.14 months, and 61.81% of patients were male. Most tumors were found in the tongue and mainly classified in T2, N0, and M0. Concerning the histologic type, one study analyzed salivary duct carcinoma, and      Table 3 presents the GRADE evaluation. The certainty of the evidence for overall survival and disease-specific survival was graded "very low" due to publication bias. While for disease-free survival the certainty was graded as "low".

Discussion
Oral cancer is one of the most common malignancies worldwide (Karjol et al., 2020). Of these, 90% are squamous cell carcinomas, which shows a 5-year overall survival rate of 50% and less than 30% in advanced stages (Dolens et al., 2021). Regarding the staging system, the TNM system is the most used classification. Of note, studies show patients in similar stages may have different prognoses (Xie et al., 2019;Kang et al., 2021). Therefore, new biomarkers and histopathological parameters are needed to predict survival in oral cancer.
In 2017, two new histopathological parameters were included in the TNM system. The T category incorporated the depth of invasion, and the N category included the extranodal extension (Dourado et al., 2020;Dolens et al., 2021). Tumor budding is a new histopathological feature that has been investigated in recent years. It is defined as a single isolated cancer cell or small cluster composed of fewer than five cells at the invasive front, representing epithelial-mesenchymal transition (Xie et al., 2019;Dourado et al., 2020). It indicates the dissociation of invasive cancer cells from the tumor mass and has been reported in many cancers as a promising prognostic parameter (Almangush et al., 2018;Zhu et al., 2018;Karjol et al., 2020). Systematic reviews with meta-analyses show that high-grade TB is significantly associated with poor prognosis in patients with gastric cancer, gynecologic cancer, lung cancer, metastatic colorectal cancer, and head and neck squamous cell carcinoma (Zhu et al., 2018;Guo et al., 2019;Ailia et al., 2022;Thakur et al., 2022;Qu et al., 2023).
This systematic review and meta-analysis included 22 studies that investigated the prognostic role of TB in oral cancer. According to our analyses, this histopathological feature was significantly correlated with oral cancer. Therefore, high-grade tumor budding is an independent predictor of overall survival, disease-specific survival, and disease-free survival in this cancer. To the best of our knowledge, this is the first systematic review that included all subtypes of oral cancer in the meta-analyses and included the largest number of studies.
In our review, all studies were cohort types. This design is susceptible to confounding bias, which can be controlled by multivariate analysis. Thus, other strengths of this review are the pooled HRs from multivariate analyses. This effect estimate was mainly adjusted for age, gender, tumor size, lymph node invasion, and clinical stage, showing that tumor budding is an independent prognostic factor in oral cancer. In addition, we conducted a sensitivity analysis, suggesting the robustness of the findings.
This study has some limitations. Firstly, there is no common consensus on the assessment method. Most studies evaluated tumor buds using H&E staining. This parameter easily adapts to routine histopathologic examination without requiring additional expensive techniques (Wang et al., 2011). However, tumor buds may sometimes be hard to identify. In this case, a cytokeratin stain may be used to confirm the suspected tumor bud (Studer et al., 2021). Another limitation is the different cutoff points, which contributed to the detection of statistical heterogeneity. Subsequently, we removed the heterogeneity when grouping studies with the same cutoff point (5 buds).
There is consensus on the evaluation of tumor budding only in colorectal cancer. In 2016, a scoring method for tumor budding in colorectal cancer was proposed by the International Tumor Budding Consensus Conference (ITBCC), contributing to the implementation of the assessment of this histopathological parameter in routine practice, clinical trials, and meta-analyses (Studer et al., 2021). The main recommendations according to ITBCC were to count tumor buds in the hotspot (20x objective), using a three-tier system (0-4 buds, 5-9 buds, and 10 or more buds) (Lugli et al., 2017).
In this study, the certainty of the evidence was graded as very low for OS and DSS due to publication bias. Conversely, certainty was graded as low for DFS. Recently, the certainty was graded as very low in a systematic review that investigated the association between tumor budding and DFS in oral SCC (Dolens et al., 2021). Here, the researchers only included five studies in the meta-analysis. We included twelve studies, and no publication bias was identified. Therefore, there was no downgrade. We did not upgrade for moderate certainty because the HR did not greater than 2 .
Further studies are needed to assess the prognostic role of tumor budding and allow for an increase in evidence. These studies need to standardize tumor budding  Table 3. GRADE Summary of Findings * Publication bias was suspected as less than ten studies were included in each meta-analysis; HR: Hazard Ratio; CI: confidence interval assessment, present HR data for each subsite of the oral cavity and perform multivariate analysis. We recommend prospective cohorts as the design, and further systematic reviews with meta-analyses should be conducted. Thus, we will have certainty about the prognostic role of tumor budding in oral cancer, more precise risk stratification, and better clinical decision-making.
In conclusion, This systematic review and meta-analysis showed an association between tumor budding and the prognosis of oral cancer patients. We found that high-grade tumor budding is an independent prognostic parameter for overall survival, disease-specific survival, and diseasefree survival. However, the certainty of the evidence was graded as low or very low. Another point identified in this research was the variability of evaluation methods for this histopathological parameter. Therefore, more studies with standardized tumor budding assessments are necessary to increase the level of evidence.