Key points are not available for this paper at this time.
Background/Objectives: Occult cervical nodal metastasis drives prognosis in oral cavity squamous cell carcinoma (OCSCC), yet current tools for risk stratification in clinically node-negative (cN0) patients are imperfect. Machine learning (ML) models have been proposed to refine selection for elective neck dissection (END), but their diagnostic performance and generalizability are unclear. Methods: We performed a diagnostic test accuracy systematic review and meta-analysis of ML models predicting occult nodal metastasis in adults with cN0 OCSCC. Eligible studies evaluated an ML-based model, used pathologic nodal status as the reference standard, and reported or allowed reconstruction of sensitivity and specificity. Internal and external validation were distinguished; quantitative synthesis was restricted to non-overlapping external validation cohorts. Diagnostic performance was synthesized with a bivariate random-effects hierarchical summary receiver operating characteristic (HSROC) model, with prespecified sensitivity analyses restricting to lower-risk patient-selection cohorts, models using only preoperative predictors, and non-outlying cohorts. Results: Thirteen retrospective studies (4730 patients) met inclusion; pooled occult nodal metastasis prevalence was 23.6% (crude 20.4%). Eight studies reported only internal validation; five provided six external validation cohorts. Across these external cohorts, pooled sensitivity was 0.79 (95% CI, 0.64–0.89) and specificity 0.84 (95% CI, 0.70–0.93). Negative predictive values were consistently high (0.94–0.98), whereas positive predictive values were modest (0.39–0.85). Sensitivity analyses yielded similar summary estimates with wider confidence intervals. Conclusions: Externally validated ML models for predicting occult nodal metastasis in cN0 OCSCC show promise but remain insufficiently validated to guide END in routine practice.
Hughes et al. (Wed,) studied this question.