First pan-endocrine meta-analysis identifies MIR-221 as a cross-tumour transcriptomic biomarker: a machine learning framework for precision surgical oncology

Abstract Background Surgical decision-making in endocrine malignancies is often confounded by heterogeneous biology, multifocality and unpredictable recurrence. In this context, microRNAs (miRNAs) have emerged as non-invasive biomarkers with translational potential. Their integration could refine diagnostic and prognostic pathways across thyroid, adrenal, parathyroid and neuroendocrine tumours. Although tumour-specific miRNA signatures have been reported, no unifying biomarker has been validated across endocrine cancers. This study presents the first transcriptomic, machine learning-integrated meta-analysis with in-silico pathway mapping to identify dominant miRNA classifiers, building on prior findings in a 30-patient indeterminate thyroid nodule cohort. Method A systematic review and meta-analysis of 34 studies (n = 3412) was performed. Pooled diagnostic metrics (Area Under Curve [AUC], sensitivity) and prognostic estimates (hazard ratios) were derived using DerSimonian–Laird and Bayesian hierarchical models. SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) assessed feature attribution. Logistic regression evaluated additive effects. Validated targets were mapped to canonical pathways using support vector machine (SVM)-ranked miRDB, miRWalk and TarBase databases. Risk of bias was evaluated using Newcastle-Ottawa. Results miR-221 emerged as the highest-ranking biomarker (AUC 0.841;HR 10.10;P < 0.001), improving multi-miRNA classifiers by +0.07 AUC. SHAP and LIME confirmed dominance over miR-146b (AUC 0.60) and miR-375 (AUC 0.66). In thyroid cancer alone, miR-221 achieved 88% sensitivity, outperforming cytology (60–75%) and ultrasound (52–85%). Logistic regression validated it as the top individual classifier (AUC 0.77). Validated targets were enriched in MAPK and PI3K–Akt. Conclusion We propose miR-221 as the first cross-endocrine molecular marker with both diagnostic and prognostic value. A translational pipeline is planned, including prospective validation in a 30-patient indeterminate thyroid cohort.