This systematic review and meta-analysis was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines (17). The protocol was developed a priori and followed established recommendations for prognostic factor research synthesis.

A comprehensive literature search was performed on September 23, 2024, across PubMed, Scopus, and Web of Science. The search strategy combined controlled vocabulary and free-text keywords related to young age, oral and tongue cancers, and prognostic or regression-based outcomes. The exact search queries used for each database are provided in Supplementary Table S1. No restrictions were applied on publication year. Only English-language publications were considered due to feasibility constraints. All retrieved records were imported into EndNote for reference management, and duplicates were removed prior to screening.

Studies were eligible for inclusion if they met the following criteria:

Studies were excluded if they:

Two reviewers independently screened titles and abstracts and subsequently reviewed the full texts of potentially eligible articles. Disagreements were resolved through discussion until consensus was achieved. The study selection process is summarized using a PRISMA flow diagram (Figure 1).

Data were extracted independently and in duplicate using a standardized form. Extracted variables included study characteristics (country, year, design), sample size, demographic and clinical characteristics, tumor site and stage, treatment modality, and all reported prognostic estimates. Overall survival (OS) was predefined as the primary outcome for quantitative synthesis. Adjusted estimates for disease-specific survival (DSS), disease-free survival (DFS), and other oncologic outcomes were extracted when available but were summarized narratively due to limited and heterogeneous reporting. For each prognostic factor, adjusted hazard ratios (aHRs) were extracted preferentially; unadjusted estimates were collected when adjusted values were unavailable and are presented narratively in Supplementary Table S2. When studies presented multiple adjusted models, the model with the most complete adjustment for confounders was selected.

Methodological quality of included studies was evaluated using the National Institutes of Health (NIH) Quality Assessment Tool for Observational Cohort and Cross-Sectional Studies. Each study was rated as good, fair, or poor quality based on criteria such as adequacy of confounding control, outcome assessment, loss to follow-up, and analytical transparency.

Quantitative synthesis focused on adjusted hazard ratios for overall survival (OS). Meta-analyses were performed using a random-effects model with restricted maximum likelihood (REML) estimation to account for between-study heterogeneity. For each prognostic factor reported in at least two studies, log-transformed hazard ratios and their standard errors were pooled and back-transformed for interpretation. Heterogeneity was quantified using the I² statistic. Adjusted prognostic estimates reported for DSS, DFS, or locoregional outcomes were not pooled and are summarized narratively in Supplementary Table S3.

Unadjusted effect estimates, or those reported only once in the literature, were not pooled and instead summarized narratively. Subgroup analyses based on country, geographic region, tumor location, or clinical stage could not be performed due to insufficient stratified reporting across included studies. Publication bias was not assessed because no meta-analysis contained ten or more studies, consistent with recommendations that funnel plot asymmetry tests are unreliable under this threshold.

All statistical analyses were conducted using STATA V18.

The database search yielded a total of 3,307 records, including 793 from PubMed, 2,053 from Scopus, and 461 from Web of Science (Figure 1). After removal of 717 duplicate entries through EndNote, 2,590 unique records proceeded to title and abstract screening. Of these, 2491 were excluded, leaving 99 reports for full-text retrieval. One report could not be retrieved, resulting in 98 articles being assessed for eligibility. Following full-text evaluation, 74 reports were excluded for the following reasons: prognostic analyses not specific to the young patient group (n = 61), absence of prognostic factors (n = 5), inclusion of fewer than 20 cases (n = 1), limited to salivary gland cancer (n = 1), or lack of regression-based effect estimates (n = 7). A total of 23 studies met the inclusion criteria and were incorporated into the qualitative/quantitative synthesis (18–40).

The summary of included studies’ characteristics can be found in Table 1. Most evidence came from the United States (9 studies, 37.50%) followed by India (5 studies, 20.83%), and South Korea (2 studies, 8.33%), respectively. All of included studies were retrospective cohort in design. Overall, 6965 young oral cancer patients were included, with male predominance (1925 patients, 56.97%) in studies reporting gender data. Among studies reporting cancer stage-related data, patients with stage I were the most predominant (931 patients, 33.1%), followed by stage II (786 patients, 27.9%), stage III (612 patients, 21.7%), and stage IV (486 patients, 17.3%), respectively. As for location-based cancer distribution, tongue cancer was the most frequently reported site (Table 2).

A detailed description of each study’s methodological quality is provided in Table 3. Overall, nine studies had good quality while the remaining 15 studies had fair quality. The main drawback was the lack of confounding control either in the design or analysis phase.

Across demographic variables (Figure 2), age was not significantly associated with survival (three studies; aHR = 0.80, 95% CI 0.34–1.25; I² = 99.79%). Female sex similarly showed no significant effect (two studies; aHR = 1.08, 95% CI 0.97–1.20; I² = 0%). Black race demonstrated a significantly higher risk (one study; aHR = 2.79, 95% CI 1.40–5.56). Lack of insurance was evaluated in one study only (aHR = 1.54, 95% CI 0.88–2.63), and thus this observation remains inconclusive. Smoking was not significantly associated with prognosis (two studies; aHR = 0.66, 95% CI –0.16 to 1.47; I² = 92.01%).

Among primary tumor measures (Figure 2), tumor greatest dimension did not show a clear prognostic association (one study; aHR = 1.02, 95% CI 0.27–3.80). Tumor size per centimeter was significantly associated with worse outcomes (three studies; aHR = 1.01, 95% CI 1.00–1.03; I² = 96.13%). Tumor thickness was examined in one study (aHR = 1.50, 95% CI 0.19–11.61), which does not allow firm conclusions. Depth of invasion was significantly associated with poorer survival (two studies; aHR = 1.03, 95% CI 1.01–1.05; I² = 0%).

Histologic grade (Figure 2) showed a non-significant association in the single study reporting it (aHR = 4.19, 95% CI 0.97–18.07). High-grade tumors (grade 3–4) demonstrated a significantly increased risk (two studies; aHR = 2.15, 95% CI 1.52–2.77; I² = 0%). Moderately differentiated tumors were assessed in one study (aHR = 2.08, 95% CI 0.79–5.52), and moderate-to-poor differentiation likewise relied on a single, non-conclusive estimate (aHR = 2.80, 95% CI 0.75–10.42). Poorly differentiated tumors, evaluated through one study, showed a strong significant association with worse outcomes (aHR = 7.75, 95% CI 2.61–23.01).

Patterns of invasion were examined in one study only (aHR = 2.115, 95% CI 0.79–5.61), providing insufficient evidence for definitive interpretation. Lymphovascular invasion was borderline and non-significant (two studies; aHR = 2.15, 95% CI –0.12 to 4.42; I² = 0%). Perineural invasion (three studies) did not reach statistical significance (aHR = 1.52, 95% CI –1.46 to 4.51; I² = 0%). Extension beyond the organ of origin was evaluated in one study (aHR = 1.86, 95% CI 1.00–3.44), suggesting a possible association but requiring replication. Positive surgical margins were assessed across three studies and showed no statistically significant association (aHR = 1.58, 95% CI 0.99–2.16; I² = 0%).

Advancing T stage (Figure 3) was significantly associated with poorer survival (two studies; aHR = 1.77, 95% CI 1.23–2.31; I² = 0%). Similarly, pT3–4 tumors demonstrated an elevated but non-significant risk (three studies; aHR = 1.06, 95% CI -0.49 – 2.60; I² = 0%). Higher N stage was significantly associated with worse outcomes (two studies; aHR = 1.24, 95% CI 1.11–1.37; I² = 0%), and the presence of nodal metastasis (N+) conferred a markedly increased risk (three studies; aHR = 2.05, 95% CI 1.24–2.85; I² = 0%).

Stage IVB was assessed in one study only and showed a substantially elevated hazard (aHR = 8.37, 95% CI 2.92–23.98), although this finding remains non-conclusive due to reliance on a single dataset. Disease recurrence, likewise reported in one study, demonstrated a strong association with worse prognosis (aHR = 11.05, 95% CI 3.8–31.52). Distant metastasis was analyzed across two studies and was not significantly associated with survival (aHR = 1.91, 95% CI 0.41–3.41; I² = 0%).

TERTp mutation status (Figure 3) was evaluated in one study and showed a significant association with poorer outcomes (aHR = 3.00, 95% CI 1.03–8.752); however, conclusions remain limited given the single-study evidence base. PRKCA mutation also demonstrated a significant association (one study; aHR = 3.57, 95% CI 1.53–8.56), but similarly should be interpreted cautiously due to non-replication across studies.

The choice of primary treatment modality showed notable associations with outcomes (Figure 3). Surgery alone was associated with increased risk (one study; aHR = 1.97, 95% CI 1.07–3.64), although this estimate is based on a single study. Conversely, omission of surgery showed a comparable magnitude of increased risk (one study; aHR = 1.98, 95% CI 1.01–3.88), also non-conclusive due to single-study origin. Receipt of adjuvant therapy was strongly associated with poorer survival across three studies (aHR = 8.16, 95% CI 2.85–13.48; I² = 0%), a finding likely reflecting treatment selection for patients with more advanced disease. Lack of radiotherapy was evaluated in one study and showed no significant association (aHR = 1.43, 95% CI 0.97–2.09).

Our finding that conventional tumour factors (size, depth of invasion, T/N stage, nodal positivity, high histologic grade) are associated with poorer survival in younger cohorts is concordant with multiple prior systematic reviews and population studies that emphasize stage and pathological features as primary determinants of outcome across age groups (2, 3, 41). Where prior pooled work has yielded apparently conflicting conclusions about whether younger patients do better or worse overall, those differences are often explained by differences in outcome definition, the inclusion of oropharyngeal subsites, and whether analyses used adjusted estimates. For example, Tagliabue et al. (3) specifically demonstrated that unadjusted comparisons frequently mask the adverse prognostic weight of comorbidity and stage, and that after adjustment older age was associated with worse mortality in tongue cancer, while younger patients had higher local recurrence risk. Similarly, Panda et al. (15) observed better crude overall survival for younger patients but worse disease-free survival and higher recurrence and distant metastasis rates in unmatched analyses, highlighting the importance of confounder control and subsite composition.

The consistent prognostic effect of nodal disease and higher T stage in our pooled analyses mirrors well-established biologic reality and staging data (e.g., AJCC/TNM) and is reinforced by contemporary prognostic-marker reviews for tongue cancer that show depth of invasion and nodal status to be among the strongest clinicopathologic predictors of outcome. Importantly, our pooled effect sizes for nodal metastasis (aHR ≈2.05) and advancing T stage (aHR ≈1.77) are clinically meaningful and concordant with those primary studies and meta-analyses that focus on oral tongue tumors (3, 16).

Molecular and biomarker data in young cohorts remain preliminary. Single-study associations identified in our review (e.g., TERTp and PRKCA mutations) indicate potential biological differences in some tumors arising at younger ages; however, these findings are not yet replicated across independent cohorts and therefore cannot be considered established prognostic markers. This cautious conclusion aligns with recent narrative syntheses and biomarker-focused meta-analyses that call for multicenter validation before adopting such markers clinically (14, 16).

This study pooled adjusted hazard ratios wherever available and prioritized regression-based, confounder-controlled estimates; that approach reduces bias from simple crude comparisons and from treatment selection effects. By focusing on adjusted estimates and using REML random-effects pooling, the synthesis emphasizes effect sizes that are more likely to reflect independent prognostic contributions. Our findings are therefore complementary to prior syntheses that pooled mainly unadjusted or mixed estimates (2, 15).

Despite methodological strengths, the available evidence has important limitations that must temper interpretation. Although the present review focused on OSCC, which represents the dominant histology in young-onset oral cancer, heterogeneity in molecular drivers within OSCC could not be systematically explored due to limited reporting. Most included studies were retrospective, observational, and hospital-based; only a minority reported comprehensive confounder adjustment, and residual confounding (for example by comorbidity, socioeconomic status, access to care, or detailed treatment factors) remains likely. Several prognostic associations were based on single-study estimates (e.g., specific mutations, some histologic subcategories), precluding confident generalization. The NIH quality assessment applied across studies highlighted frequent deficiencies in controlling for confounders and in analytic transparency; these concerns mirror the assessments reported in prior reviews.

Heterogeneity across studies was another important limitation. Studies used different age cutoffs to define “young,” varied in subsite inclusion (oral cavity vs. oropharynx; tongue-only analyses versus mixed oral sites), and differed in stage distribution and treatment approaches, all of which contribute to statistical heterogeneity and limit pooled inference. Several earlier systematic reviews have underlined the same source of heterogeneity and recommendation for standardized definitions (3, 14, 15).

In addition, although several studies reported adjusted associations with disease-specific or disease-free survival, the limited number of studies per outcome and inconsistent reporting precluded quantitative synthesis; these findings are therefore presented narratively in Supplementary Table S3.

Finally, several potentially important prognostic domains (e.g., HPV status in subsites, comprehensive genomic signatures, immunologic markers, and detailed margin and nodal-yield metrics) were either inconsistently reported or absent in most studies; this limited our ability to perform subgroup or meta-regression analyses that might identify effect modification by these factors. Recent narrative syntheses and methodological reviews reach similar conclusions about evidence gaps in molecular and prognostic modelling for younger patients.

For clinicians, the principal implication is that established tumour factors (size, depth, stage, nodal metastasis, poor differentiation) should remain central to risk stratification and treatment planning in younger patients rather than using chronological age alone as the justification for de-escalation or escalation of therapy. Our pooled results suggest that young patients with adverse pathologic features carry similar or higher risks as older counterparts and therefore warrant guideline-concordant staging, neck management, and adjuvant decision-making based on established pathologic indicators (2, 3).

From a research standpoint, several priorities emerge. First, prospective, multi-institutional cohorts with pre-specified age thresholds and harmonized reporting of clinicopathologic, treatment, and molecular variables are needed to validate single-study molecular signals (e.g., TERTp, PRKCA) and to allow robust multivariable modelling. Second, prognostic model development for young-onset oral cancer should follow best-practice guidance (transparent reporting, adequate events per variable, internal/external validation) because current prognostic models for head and neck subsites frequently suffer high risk of bias and limited external validation. Finally, studies that separate oral cavity subsites (tongue vs. non-tongue) and that stratify by HPV status where relevant will reduce heterogeneity and produce clinically actionable stratification tools (14).