Editorial on the Research Topic
Cancer biology, immunotherapy, and aging converge in ways that challenge traditional clinical paradigms (
Several studies in this Research Topic deploy machine learning (ML) to address gaps in traditional clinical prediction models. The differentiated thyroid cancer (DTC) study demonstrates the strength of ensemble algorithms, particularly XGBoost, for capturing nonlinear interactions between clinical variables. Similarly, the breast cancer axillary lymph node burden analysis highlights the value of combining radiomics with ML to extract predictive imaging signatures not discernible through human assessment. These methods show clear advantages in handling high-dimensional data, improving accuracy, and enabling personalized risk estimation. However, a shared limitation emerges: ML models rely heavily on retrospective datasets, are vulnerable to overfitting, and often lack interpretability despite tools like SHAP. External validation, present in some studies but absent in others, remains essential before these tools can be clinically deployed.
The study identifying T-cell exhaustion and macrophage polarization genes in breast cancer leverages multi-omics data integration and machine learning to uncover immune-related prognostic signatures. This contrasts with imaging-driven methods but shares the goal of revealing biological complexity underlying variable clinical outcomes.
While this approach offers mechanistic depth not available through radiomics or clinical modeling, it introduces its own constraints, such as variability across datasets, bulk RNA-seq limitations (cell-type averaging), and challenges in translating genetic signatures into actionable clinical decision-making.
The glioma and colorectal neuroendocrine neoplasms (CRNEN) studies use traditional regression-based nomograms, long valued for their transparency and clinician usability. Unlike ML models, nomograms allow straightforward bedside application and highlight effect sizes for individual variables. Yet, these model types assume linearity and may oversimplify complex interactions, a trade-off between usability and nuance.
Notably, the CRNEN models incorporate a dual predictive-prognostic framework, illustrating how classic methodologies continue to evolve when embedded within large, high-quality datasets.
The TAS-102 case report, while not methodologically complex, fills a crucial evidence gap: real-world therapeutic guidance for frail, multimorbid elderly patients. Such clinical narratives complement computational studies by grounding innovations in practical, human-centered contexts, an essential dimension often absent from data-driven work.
Across tumor types, these works emphasize that aging fundamentally shapes cancer risk, microenvironment, treatment tolerance, and clinical trajectory. Yet, only a minority of the included studies explicitly stratify by age or examine age-specific biological mechanisms, revealing a critical research gap.
Radiomics, machine learning, and transcriptomics each offer a different “lens” through which to view tumor behavior. When integrated, these modalities could enable more accurate, multi-dimensional prediction models that reflect both tumor biology and host factors, particularly in older patients.
Nearly all studies rely on retrospective datasets such as SEER or single-center cohorts. While rich in sample size, these datasets lack functional biomarkers, treatment toxicity data, and geriatric assessments. Prospective, age-inclusive cohorts are urgently needed.
Despite their contributions, the studies share several overarching limitations: Retrospective designs introduce selection bias and limit causal inference. Lack of functional aging measures: chronological age alone inadequately captures biological aging or treatment tolerance. Limited longitudinal data: Few studies examine dynamic changes over time, especially relevant for aging-related biology. Sparse diversity in many datasets underrepresents low-income, rural, and minority populations. Translational gap: predictive models often lack pathways for real-world implementation in clinics.
Addressing these limitations will be essential to ensure that computational advances translate into meaningful clinical impact.
Models incorporating frailty indices, functional status, inflammatory markers, and patient-reported outcomes could dramatically improve prediction accuracy for older adults.
Future studies should explore models that combine: Clinical data Radiomics Genomics Immune profiling Aging biomarkers
Such integration could capture the full biological and clinical complexity of cancer in aging populations.
Predictive models must move beyond validation toward prospective testing, where predictions actively inform treatment choices.
As ML models become more prevalent, transparency, fairness, and bias mitigation will be critical, particularly in elderly populations disproportionately affected by health disparities.
The studies in this Research Topic collectively demonstrate how emerging analytic tools and biological insights can refine cancer characterization and improve clinical decision-making across diverse tumor types. Yet they also highlight persistent challenges: inadequate age-specific data, limited external validation, and gaps in real-world applicability. Bridging these divides will require interdisciplinary collaboration, prospective validation, and a deeper commitment to studying cancer through the lens of aging biology.
By embracing these directions, the field can move toward genuinely personalized, age-inclusive oncology where innovations in cancer immunotherapy, computational science, and geroscience converge to benefit all patients across the lifespan.
SK: Conceptualization, Data curation, Formal Analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing – original draft, Writing – review and editing. JGW: Writing – review and editing, Methodology, Formal analysis, Validation, Visualization. PBS: Writing – review and editing, Methodology, Visualization, Software.
The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
The author(s) declared that generative AI was not used in the creation of this manuscript.
Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.