Prediction models for postoperative recurrence in papillary thyroid carcinoma: a systematic review and critical appraisal.

Background: Prediction models for postoperative recurrence in papillary thyroid carcinoma (PTC) have increased substantially in recent years. However, recurrence outcomes are inconsistently defined across studies, particularly with respect to structural and biochemical recurrence, and the quality and clinical applicability of existing models remain uncertain. Objective: To systematically review and critically appraise multivariable prediction models for structural postoperative recurrence in pathologically confirmed PTC and to evaluate their predictive performance, methodological quality, and risk of bias. Methods: PubMed, Embase, and the Cochrane Library were searched from inception to February 2026. Studies developing or validating multivariable prediction models for structural recurrence in adult patients with PTC were included. Data extraction was guided by the CHARMS checklist, and risk of bias was assessed using PROBAST. Findings were synthesized narratively, and an exploratory meta-analysis of discrimination performance from validation studies was conducted where appropriate. Results: Thirteen retrospective studies met the inclusion criteria, all of which were conducted in East Asian populations. Reported discrimination was generally acceptable, with most AUC or C-index values exceeding 0.70. However, all studies were judged to have a high overall risk of bias, primarily due to limitations in the analysis domain, including inadequate handling of overfitting, insufficient sample size justification, limited reporting of missing data, and reliance on internal validation. Only two studies performed external validation. Exploratory pooling of validation AUCs suggested moderate predictive performance but substantial heterogeneity across studies. Conclusion: Current prediction models for structural recurrence in PTC show promise for individualized risk estimation but remain limited by methodological weaknesses, heterogeneous modelling approaches, inadequate assessment of calibration, and scarce external validation. Future studies should adopt standardized recurrence definitions, improve reporting transparency, and prioritize robust external validation before routine clinical implementation can be recommended.