The Potential microRNA Prognostic Signature in HNSCCs: A Systematic Review

Head and neck squamous cell carcinomas (HNSCCs) are often diagnosed at advanced stages, incurring significant high mortality and morbidity. Several microRNAs (miRs) have been identified as pivotal players in the onset and advancement of HNSCCs, operating as either oncogenes or tumor suppressors. Distinctive miR patterns identified in tumor samples, as well as in serum, plasma, or saliva, from patients have significant clinical potential for use in the diagnosis and prognosis of HNSCCs and as potential therapeutic targets. The aim of this study was to identify previous systematic reviews with meta-analysis data and clinical trials that showed the most promising miRs in HNSCCs, enclosing them into a biomolecular signature to test the prognostic value on a cohort of HNSCC patients according to The Cancer Genome Atlas (TCGA). Three electronic databases (PubMed, Scopus, and Science Direct) and one registry (the Cochrane Library) were investigated, and a combination of keywords such as “signature microRNA OR miR” AND “HNSCC OR LSCC OR OSCC OR oral cancer” were searched. In total, 15 systematic literature reviews and 76 prognostic clinical reports were identified for the study design and inclusion process. All survival index data were extracted, and the three miRs (miR-21, miR-155, and miR-375) most investigated and presenting the largest number of patients included in the studies were selected in a molecular biosignature. The difference between high and low tissue expression levels of miR-21, miR-155, and miR-375 for OS had an HR = 1.28, with 95% CI: [0.95, 1.72]. In conclusion, the current evidence suggests that miRNAs have potential prognostic value to serve as screening tools for clinical practice in HNSCC follow-up and treatment. Further large-scale cohort studies focusing on these miRNAs are recommended to verify the clinical utility of these markers individually and/or in combination.


Introduction
Among the main tumors of the head and neck region, oral squamous cell carcinomas (OSCCs) represent the sixth malignant tumor in global incidence, with about 700,000 thousand new cases each year [1].
The risk factors most associated with the onset of head and neck squamous cell carcinoma (HNSCC) are alcohol and the consumption of smoked or chewed tobacco, and for laryngeal squamous cell carcinoma (LSCC), positivity to HPV subtypes 16 and 18 was considered a risk factor but with a favorable prognosis [2].
Non-Coding RNA 2023, 9, 54 2 of 32 Survival at 5 years after diagnosis remains very low, as only one of two patients survives, and surgical resective therapy can be very debilitating, with a worsening of the quality of life, difficulty in swallowing and speech, and in general due to a perceived deterioration in the relationship with other people [3].
The identification of survival prognostic biomarkers remains a very open topic: In fact, the ability to predict a disease by estimating the clinical trend and survival time remains one of the diagnostic and prognostic objectives to be achieved.In recent decades, several prognostic biomarkers have been investigated in an attempt to create a predictive survival biomolecular signature.
Among the widely studied prognostic and diagnostic biomarkers associated with head and neck cancers, we have the non-coding sequences of RNA messenger (mRNA), and among these, microRNAs (miRNA/miRs) [4].The latter group is a class of mature, noncoding, single-stranded RNAs with 21-23 nucleotides, which were proposed as promising biomarkers for patients with cancer diagnosis and follow-up [3,5].Some previous systematic literature reviews have tried to identify individual miRs, aggregating the prognostic survival data of multiple studies, obtaining promising results only in some cases for HNSCCs, such as in the cases of miR-31 [6], miR-21 [7], miR-155 [8], and miR-195 [9].
Other studies tried to identify a biomolecular signature by aggregating miRNAs in HNSCC tissue expression, exploring their use as potential biomarkers for cancer detection and/or prognosis [10,11].
This systematic review aims to identify retrospective and prospective clinical studies investigating the prognostic value of miR expression in HNSCC patients, as well as including data from previous systematic reviews with meta-analyses.From these studies, we selected the most promising miRs, inserting them into a biomolecular signature to test the prognostic value on a cohort of HNSCC patients according to the "The Cancer Genome Atlas" (TCGA) [12].

Study Selection
The following research question guided the selection of the studies: Are there biomolecular signatures consisting of non-coding mRNA sequences (especially miRs) in the scientific literature, whose differential expression in HNSCC tumor tissues was indicative of a different prognosis in patient survival?
The research phase was carried out by consulting and extracting the bibliographic references on three databases, SCOPUS (2455), Science Direct (1367), PubMed (2505), and on a Cochrane Library register (5), providing a total of 6332 articles.
Filters were applied on PubMed and Scopus to selectively include literature reviews and meta-analyses, together with clinical studies.Subsequently, the bibliographic references of Scopus and PubMed were reported on EndNote X8, and the duplicates were removed, while further overlapping of the references were manually removed.The articles obtained were selected by reading the abstract and the title; this phase was also performed for Science Direct and the Cochrane Library, and the articles selected from these two sources were added to those chosen from PubMed and Scopus, and thus 117 potentially eligible records were obtained.
A further search of the gray literature (Google Scholar and Open Gray) and previous systematic reviews did not identify additional manuscripts for inclusion in the present systematic review (Figure 1).Records were independently screened by two authors (M.D. and A.B.), while dubious situations were addressed at the end of the selection by involving a third author (F.S.) to resolve potential conflicts.
In total, 76 clinical studies and 15 systematic reviews were included at the end of the inclusion process.We designed our strategy to be optimized for a sensitive and broad search, and the results of this selection are reported in a flowchart (Figure 1).
On average, the selected reviews included many studies (≅9.1), with a range from 1 to 36, and the number of included patients ranged from 80 to 1200.Although the systematic review included HNSCCs, two reviews involved only OSCCs, and one study only covered LSCCs.The most reviewed prognostic index was the HR of OS (across different miR tissue expression levels), with thirteen reviews, followed by DFS (six studies), RFS (three studies), CSS (two studies), and PFS (one study).Only one review evaluated the RR of OS.
On average, the selected reviews included many studies ( ∼ =9.1), with a range from 1 to 36, and the number of included patients ranged from 80 to 1200.Although the systematic review included HNSCCs, two reviews involved only OSCCs, and one study only covered LSCCs.The most reviewed prognostic index was the HR of OS (across different miR tissue expression levels), with thirteen reviews, followed by DFS (six studies), RFS (three studies), CSS (two studies), and PFS (one study).Only one review evaluated the RR of OS.
The microRNAs most reviewed and included in the signatures were miR-21 (in eight revisions including four as a single miR), miR-155 (four times, three of which were within a signature with multiple miRs), and miR-375 (two within a signature).In particular, miR-21 was the most studied and (taken individually) presented an HR of OS ranging from 1.29 [7] to 1.81 [16].
The total number of included patients affected by HNSCCs was 6848, with 3295 cases definitely identified as OSCCs and at least 1493 presenting a localization to the tongue, while LSCC was present in 2179 patients.
In miR-21, the HR of OS between high and low expression levels ranged from 5.31 95% CI: [1.39-20.38][24] to 1.1302 95% CI: [0.34-3.757][98], and in poor OS, it was upregulated, similar to miR-155, while in miR-375, the HR of OS between low and high expression levels ranged from 12.8 95% CI: [3.4-48.6][57] to 1.32 95% CI: [0.76-2.27][53], and in case of low survival, it was downregulated.All data related to the clinical studies, as well as the survival data extrapolated from the Kaplan-Meier survival curves, are extensively reported in Tables 2 and 3.
Analyzing the studies and the systematic reviews performed on the prognostic biomarkers of survival, it becomes clearly evident that the miR that has been most investigated and provides the greatest number of data is miR-21, with 19 clinical studies, followed by miR-155 (8 studies) and miR-375 (7 studies).The other miRs have fewer studies with fewer patients included than miR-21 (1262 patients), miR-155 (706 patients), and miR-375 (572 patients).
For this reason, we decided to use a different cohort (TCGA), which includes about 512 patients, to verify whether a biosignature with a high expression of these three miRs in tumor tissues was correlated with low survival, and significant results were obtained for miR-21 and miR-155, while for miR-375, low survival was associated with low expression.Table 2. Data extracted from the 76 studies included, providing information regarding the type of tumor, the location of the tumor, the number of patients with data concerning the average age, the average or maximum follow-up, gender, and the common risk factors in the patients are reported to be smoking, alcohol, and HPV positivity; TNM (T: tumor size; N: regional lymph nodes; M: distant metastasis); pTNM, pathological TNM staging; cTNM, clinical TNM staging; N/A, not available; Ma (male); Fe (female); R (range); y (years); smoking (Sm); alcohol (Alc); SEM (standard error mean); PS (prospective study); RT (retrospective study); HPSCC (hypopharyngeal squamous cell carcinoma); OTSCC (oral tongue squamous cell carcinoma); BOTSCC (base of tongue squamous cell carcinoma); NPC (nasopharyngeal carcinoma).Data are not reported in a clear and explicit manner;\ data not present.Using the extracted data shown in Tables 1 and 2, the three main miRs investigated in the literature (miR-21, miR-155, and miR-375), whose altered expression was investigated in the prognosis of survival in HNSCC patients and included in molecular biosignatures, were then selected.The evaluation was performed through the Kaplan-Meier plotter database portal (https://kmplot.com/analysis/,accessed on 10 May 2023) [100], and HR data were extracted.
The difference between high and low tissue expression levels of the miRs taken into consideration presents an HR of OS = 1.28 95% CI: [0.95, 1.72], log-rank p = 0.1.Moreover, the Kaplan-Meier survival curve generated using the portal is depicted in Figure 2, and the considered follow-up period was 60 months.The median survival in the cohort of patients with low expression was 58.73, while that in patients with high expression was 46.47.
The cut-off value between high and low miR expression levels was automatically generated through the portal (Figure 3), and the cut-off values and the related p-values are present in the the Supplementary Materials (S1).
The portal to generate and display the Kaplan-Meier plot is used to establish a cut-off value and assign samples to one of the two cohorts, using the best available cut-off value.
To find the best cut-off, the process is repeated using the values of the input variables from the lowest quartile to the upper quartile, and the Cox regression for each setting is calculated [101].
The most significant cut-off value was employed as the optimal threshold to segregate the input data into two groups.Subsequently, the system presents a straightforward visual representation of this analysis, displaying the p-values obtained concerning the selected cut-off values.
In cases where the generated cut-off values were ambiguous (e.g., multiple cut-off values resulted in very low p-values), the value corresponding to the highest hazard ratio was selected.
The calculation of multiple cut-off values led to the generation of multiple assumptions.Hence, in this setup, the FDR was automatically computed using the Benjamini-Hochberg method to correct for multiple hypothesis testing [102].
Spearman's correlation and Pearson's correlation between the expression values of the different miRs investigated (Tables 4 and 5) were also calculated.
The difference between high and low tissue expression levels of the miRs taken into consideration presents an HR of OS = 1.28 95% CI: [0.95, 1.72], log-rank p = 0.1.Moreover the Kaplan-Meier survival curve generated using the portal is depicted in Figure 2, and the considered follow-up period was 60 months.The median survival in the cohort of patients with low expression was 58.73, while that in patients with high expression was 46.47.The cut-off value between high and low miR expression levels was automatically generated through the portal (Figure 3), and the cut-off values and the related p-values are present in the the Supplementary Materials (S1).
The portal to generate and display the Kaplan-Meier plot is used to establish a cutoff value and assign samples to one of the two cohorts, using the best available cut-off value.
To find the best cut-off, the process is repeated using the values of the input variables from the lowest quartile to the upper quartile, and the Cox regression for each setting is calculated [101].
The most significant cut-off value was employed as the optimal threshold to segregate the input data into two groups.Subsequently, the system presents a straightforward visual representation of this analysis, displaying the p-values obtained concerning the selected cut-off values.
In cases where the generated cut-off values were ambiguous (e.g., multiple cut-off values resulted in very low p-values), the value corresponding to the highest hazard ratio was selected.
The calculation of multiple cut-off values led to the generation of multiple assumptions.
Hence, in this setup, the FDR was automatically computed using the Benjamini-Hochberg method to correct for multiple hypothesis testing [102].
Spearman's correlation and Pearson's correlation between the expression values of the different miRs investigated (Tables 4 and 5) were also calculated.
All reported data can be reproduced via the Kaplan-Meier plotter portal [101].All reported data can be reproduced via the Kaplan-Meier plotter portal [101].Furthermore, additional tests were performed on miR-21, and for the main downregulated miRs in the literature, the extrapolated data are described in Figure 4.  Furthermore, the main results related to the resistance to chemotherapy and radiotherapy administered to patients, in support of resective surgical treatment and in relation to the altered expression of microRNAs, were extracted from the studies, as shown in Table 6.Furthermore, the main results related to the resistance to chemotherapy and radiotherapy administered to patients, in support of resective surgical treatment and in relation to the altered expression of microRNAs, were extracted from the studies, as shown in Table 6.

Risk of Bias
The risk of bias for systematic reviews was determined using the ROBIS tool, and for each factor, it was evaluated as "low", "high", or "unclear".The three phases of the evaluation process were as follows: Phase 1: the evaluation of the relevance of the research question (PICO); Phase 2: the identification of critical points of the review process; and Phase 3: the evaluation of the overall risk of bias of the review.All data related to the risk of bias are reported in Table 7.The main critical issues related to the individual revisions are as follows: Irimie-Aghiorghiese et al., 2019 [15]: Study eligibility criteria (?): The protocol number with which the systematic review was registered was not reported.Identification and selection of studies (?): The selection was performed only on two databases (PubMed and Embase), and the number of authors who conducted the research was not specified, nor were the start or end dates in which the review was conducted.Lubov et al., 2017 [16]: Study eligibility criteria (?): The protocol number with which the systematic review was registered was not reported.Identification and selection of studies (?): The number of authors who selected the articles and the start or end dates of the review were not reported.Data collection and study appraisal (?): The number of authors who performed the data extraction and the methods of data extraction were not stated.The manuscript is both a systematic review and a retrospective study of 100 patients.Xie and Wu, 2017 [17]: Study eligibility criteria (?): The protocol number with which the systematic review was registered was not reported.
Wang et al., 2019 [18]: Study eligibility criteria (?): The protocol number with which the systematic review was registered was not reported.Li et al., 2019 [19]: Study eligibility criteria (?): The protocol number with which the systematic review was registered was not reported.Identification and selection of studies (?): The start or end dates of the review were not specified.Data collection and study appraisal (?): The risk of bias was not formally assessed using an appropriate scale or tool.A bioinformatic analysis was also performed.Huang et al., 2021 [21]: Study eligibility criteria (?): The protocol number with which the systematic review was registered was not reported.Identification and selection of studies (?): The start and end dates of the review were not specified.Jamali et al., 2015 [22]: Study eligibility criteria (?): The protocol number with which the systematic review was registered was not reported.Troiano et al., 2018 [23]: Study eligibility criteria (?): The protocol number with which the systematic review was registered was not reported.Dioguardi et al., 2023 [9]: Synthesis and findings (?): The obtained results were excessively emphasized in the conclusions.
The risk of bias for prognostic studies was assessed using the parameters derived from REMARK.According to the REMARK guidelines, a score ranging from 0 to 3 was considered for each factor (Table 8).

Protocol
The planning of the systematic review was implemented following the guidelines described in the Cochrane Handbook for Systematic Reviews of Interventions.The drafting of the review manuscript followed the recommendations of PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analysis) [103], and the protocol was registered on PROSPERO before carrying out the selection of articles and was registered with the registration number CRD42023400856.

Eligibility Criteria
The search was directed towards the identification of retrospective or prospective clinical studies and bibliographic sources that reported systematic reviews of the literature regarding the role of non-coding RNAs and in particular of miRs, reporting prognostic data of survival in patients with HNSCCs associated with altered expression of a single miR or a signature of miR.
The exclusion criteria were the exclusion of all clinical trials and systematic reviews reporting no data on the use or detection of a molecular biosignature consisting of miRs in HNSCCs, all literature reviews (considered as bibliographic sources only), and studies that did not have an abstract in English.
Thus, the reporting data of all clinical trials and meta-analyses on a biomolecular signature consisting of miRs that is prognostic of survival in HNSCCs were considered potentially eligible.
The systematic review involved two reviewers (M.D. and A.B.) and followed the following stages: 1.
Choice of reviewers (M.D. and A.B.) and a third reviewer (F.S.) as a supervisor in case of conflict regarding the studies to be included, choice of outcomes to identify, choice of databases and k words used, choice of criteria of admissibility, choice of data to be extracted and methods of synthesis and registration of the protocol on PROSPERO; 2.
Identification of records and selection of studies through databases with the removal of duplicates performed manually or by software (EndNote 8.0), performed independently and subsequently comparison of selected studies and decision of studies to be included; 3.
Independently performed table data extraction and subsequent data comparison to minimize the risk of error in reporting information.

Sources of Information, Research, and Selection
The keywords used were microRNA AND HNSCC, LSCC AND MicroRna, OSCC AND MicroRna, and signature microRNA AND HNSCC.
The search was conducted on 3 databases, namely Science Direct, SCOPUS, and PubMed, and one registry, the Cochrane Library.Additionally, Google Scholar (keywords microRNA), gray literature sources such as Open Gray (keywords microRNA), and references from previous systematic reviews on miRs and HNSCCs were searched.
Particularly, the following are all the keywords used in the PubMed search: Search: (signature microRNA OR miR) AND (HNSCC OR LSCC OR OSCC OR oral cancer) Sort by: Most Recent.
The literature search was completed on 20 February 2023.The data to be extracted included the first author of the study, the publication date, the country in which the research was conducted, the type of squamous cell carcinoma, the number of patients involved in the study, the clinical characteristics of the patients and tumors included in the studies, data on the positivity to the HPV virus and exposure to risk factors such as smoking and alcohol, as well as clinical data on the staging of patients included in the studies and on the average or maximum follow-up, risk of bias tools, the studied miRs, the value or type of risk rate (RR) or hazard rate (HR) for various prognostic survival indices: overall survival (OS), disease-free survival (DFS), progression-free survival (PFS), relapse-free survival (RFS), and cancer-specific survival (CSS).

Risk of Bias, Bioinformatic Analysis
Furthermore, the data of a cohort of patients with HNSCCs (N ≈ 512) extracted from the Cancer Genome Atlas (TCGA) database were analyzed to obtain the HR values and associate the prognosis indices with the expression of the signature of miRs created and selected by the authors.
The risk of bias in the individual systematic reviews was assessed by two authors (M.D. and A.B.).The ROBIS (Risk of Bias in Systematic Reviews) was used as an assessment tool specifically developed to assess the risk of bias in systematic reviews.Studies with a high risk of bias were excluded from the review [104].
Clinical studies with the risk of bias were evaluated by two authors (M.D. and A.B.), and the tool used for the assessment of the parameters was derived from the reporting recommendations for prognostic studies of markers (REMARK).Studies with a high risk of bias were excluded from the analysis [105].

Discussion
In the last 40 years, a considerable number of new cancer biomarkers have been identified, but only a few have managed to be effectively used in clinical practice.
Many biomarkers pass validation very well, with concordant and reproducible results across trials; nevertheless, these biomarkers lack the capacity to decisively contribute to patient care, except to provide some additional information on prognosis.Therefore, they are considered by clinicians to be not fundamental in the therapeutic choice.
Physicians often show a tendency to overtreat specific patients, rather than relying on prognostic biomarkers that offer less than precise predictions.Using these flawed prognostic biomarkers could result in fewer patients receiving overly aggressive treatments (true positives) but, at the same time, could also increase the chance of not treating some patients who may actually benefit from therapy (false negatives).Therefore, from a clinical point of view, with the exclusion of fraudulent situations and sensationalist discoveries, the prognostic biomarkers detected are not so promising, and their failure is due to their inadequate performance in clinical practice.
Hence, in the process of transitioning a promising biomarker from Phase 1 studies to clinical implementation, meticulous consideration should be given to the study's design, aiming to mitigate bias in the utilization of sensitive, specific, and precise analytical methodologies.This involves the careful selection of suitable samples, both in terms of quantity and quality, as well as appropriate patient subgroups for the purpose of validation.Furthermore, it is imperative to apply statistically robust and rigorous methods to prevent the occurrence of data overfitting.
The data present in the literature demonstrate how miRs are stable, and the results deriving from the studies are consistent and reproducible, making miRs potential promising biomarkers for diagnosis and prognosis [106].
Prognostic biomarkers including miRs could have a significant impact in helping clinicians improve the quality of life and health conditions of HNSCC patients, providing useful information for oncologists in terms of the most appropriate therapeutic choice, according to life expectancy and neoplasm aggressiveness [107].
Knowledge of the prognostic potential of biosignatures could be useful for clinicians after the diagnosis of HNSCCs to define prognosis by formulating predictive models of individualized prognostic risk.Bringing this model back into clinical practice in patients with HNSCCs who have unfavorable prognostic biosignatures (with low RFS or OS), a more or less aggressive therapy or surgical treatment could be recommended, with a tailored therapeutic approach in the context of personalized medicine [108].
The discovery of the miRs' prognostic value presents critical insights with potential biases that must be taken into consideration before, during, and after the execution of retrospective studies or clinical trials, but also during the data meta-analysis.The choice of variables can significantly affect the results as well as the overall validity of the analysis.
Factors such as sample size, the heterogeneity of patient populations, virus positivity (HPV and EBV), and the choice of statistical analysis can affect the results.
In the context of HNSCCs, taking, for instance, HPV positivity as a variable, the role of papillomavirus as a risk factor in a subset of head and neck cancers [109], mainly oropharyngeal and laryngeal cancers, has been established, with different epidemiological, clinical, and molecular characteristics compared with HNSCCs, starting with HPV positivity, which was associated with distinctly different and more favorable prognostic survival values [110].
Therefore, the inclusion or exclusion of some clinical variables (smoking, alcohol, age, and gender) may alter the results of prognostic values for the associations observed between miR signatures, including the related survival results [111].
The meta-analysis size of the sample can also be addressed by performing a trial sequential analysis (TSA) to verify the power of the results as a function of the sample, with an effect achieved in terms of RR [112].
In addition, some laboratory study phases can be biased, making the detection of biomolecular signatures in biological samples difficult.In fact, possible biases can be identified in the sample selection (e.g., fresh tissue, fixed tissue, and biological fluids), RNA extraction, and sample quality control.In addition, miR profiling can also be affected by variability in the technical platform (instruments and software), which is an important source of bias that affects not least the data analysis [113].
In addition, the results of tissue miR expression seem to be influenced by the tissue preservation technique (frozen or in formalin); in fact, to reduce the heterogeneity of the data, it is recommended to aggregate data during a meta-analysis by conducting a subgroup analysis also based on the category of tissue preservation [114].
The difficulty in determining the prognostic value of miRs is due to the complexity of biological systems and the multiple roles of miRs in the regulation of gene expression.It is important to remember that microRNA expression patterns can vary between different cancer types, and even within subtypes of the same cancer, making it difficult to establish universal prognostic markers [113].
Furthermore, many miR biosignatures are currently being developed using algorithms and machine learning based on the search for associations between expression and disease outcomes.Therefore, causality is often not considered, and algorithms can generate signatures that are not biologically expressive, despite their statistical significance [115].
In this context, the execution of systematic reviews with the inclusion of Phase 2 prognostic studies can lead to improvement in these studies by better highlighting the most reliable and predictable results while not overlooking data without statistical significance or evidence (publication bias).The present systematic review aims to refine the design, execution, and reporting of Phase 2 studies [116,117] and provide useful knowledge in guiding Phase 3 clinical studies aimed toward finding a prognostic model [118].
In light of the data reported in the medical literature, and from the preliminary research conducted in the field [6][7][8][9]13,14,119], we carried out our review after registering it on Prospero, which was written following the indications of PRISMA.A meta-analysis of the data was not carried out due to the excessive heterogeneity of data and histological subtypes of HNSCCs, and thus a TCGA analysis was instead used to test possible microRNA biosignatures that emerged from the data extraction and qualitative analysis of the studies.
In this systematic review, we identified 64 miRs from 15 systematic reviews, whose altered expression was correlated with prognostic indices.The miRs mainly investigated were miR-21, miR-155, and miR-375.HR values for OS in miR-21 ranged from 1.29 to 1.72, and these values were 1.59 for miR 375 and 1.40 for miR-155 (considering only the results of meta-analyses reporting HR values aggregated for individual miRs); the HR values for several miR panels ranged from 2.65 to 1.10 (Table 1).
By selecting the three miRs that, based on our research, were the most investigated in HNSCCs, and performing a survival analysis using these three miRs on the patient cohorts present in the TCGA, an HR of OS equal to 1.28 was found.From these preliminary data, it is evident that the existing results in the literature are still insufficient to clearly define a prognostic microRNA biosignature, and the retrospective statistical analyses performed using the TCGA in an attempt to further validate the findings do not fully achieve this purpose.In fact, by considering only miR-21, three meta-analyses report an aggregate HR value of about 1.7 as the difference between high and low expression levels, while using the TGCA, considering a follow-up period of 60 months, miR-21 presented an HR (high and low expression) equal to 1.27 95% CI: [0.95, 1.71] (Figure 4).Considering instead the HR data of miR-21, miR-155, and miR-375 using the TCGA and combining them in a single prognostic signature, the value of HR was 1.28 95% CI: [0.95, 1.72].

Conclusions
In conclusion, we can state that although prognostic survival biomarkers have been identified that possess a discrete potential consisting of a miR signature, in the current state of knowledge for head and neck tumors, there are no studies that fully validate the results.Nevertheless, it is crucial to emphasize that additional validation is necessary before we can definitively establish their practicality.While some miRNA studies have revealed noteworthy findings related to their influence on patient survival, the limited number of studies that have been agregaded to derive these results diminishes their relevance in clinical contexts.Hence, there is a clear need for more extensive and long-term patient studies that specifically investigate these miRs.

Figure 1 .
Figure 1.Flowchart describing the mechanisms of screening miR studies and including several databases and records.

Figure 1 .
Figure 1.Flowchart describing the mechanisms of screening miR studies and including several databases and records.

Figure 3 .
Figure 3. Automatically generated cut-off plot using the Kaplan-Meier plotter web application, http://kmplot.com/analysis/,accessed on 10 May 2023.Significance vs. cut-off values between the lower and upper quartiles of expression are presented, with the red circle indicating the best cut-off.

Table 3 .
The values of HR (95% confidence interval) and RR for the different prognostic indices of survival are shown in the table; overall survival (OS); disease-free survival (DFS); recurrence-free survival (RFS); cancer-specific survival (CSS); progression-free survival (PFS); relative risk (RR); high versus low expression (H-L); low versus high expression (L-H); infinite (inf).

Table 6 .
Data on resistance to chemotherapy and radiotherapy in relation to altered expression of microRNAs.

Table 8 .
Assessment of the risk of bias; REMARK.