RT journal article
T1 External Validation of a Predictive Model for Thyroid Cancer Risk with Decision Curve Analysis
A1 Fernández Alba, Juan Jesús
A1 Carral, Florentino
A1 Ayala Ortega, Carmen
A1 Santotoribio Camacho, José Diego
A1 Castillo Lara, María
A1 González Macías, María del Carmen
A1 Santotoribio Camacho
A2 BiomedicinaBiotecnología y Salud Pública
A2 Materno-Infantil y Radiología
A2 Medicina
K1 thyroid cancer
K1 predictive model
K1 decision curve analysis
K1 papillary thyroid carcinoma (PTC)
K1 thyroid nodules
K1 external validation
K1 artificial intelligence (AI)
K1 machine learning (ML)
K1 diagnostic tools
AB Thyroid cancer ranks among the most prevalent endocrine neoplasms, with a significant rise in incidence observed in recent decades, particularly in papillary thyroid carcinoma (PTC). This increase is largely attributed to theenhanced detection of subclinical cancers through advanced imaging techniques and fineneedle aspiration biopsies. The present study aims to externally validate a predictive modelpreviously developed by our group, designed to assess the risk of a thyroid nodule beingmalignant. Methods: By utilizing clinical, analytical, ultrasound, and histological datafrom patients treated at the Puerto Real University Hospital, this study seeks to evaluatethe performance of the predictive model in a distinct dataset and perform a decision curveanalysis to ascertain its clinical utility. Results: A total of 455 patients with thyroid nodularpathology were studied. Benign nodular pathology was diagnosed in 357 patients (78.46%),while 98 patients (21.54%) presented with a malignant tumor. The most frequent histological type of malignant tumor was papillary cancer (71.4%), followed by follicular cancer(6.1%). Malignant nodules were predominantly solid (95.9%), hypoechogenic (72.4%), withirregular or microlobed borders (36.7%), and associated with suspicious lymph nodes(24.5%). The decision curve analysis confirmed the model’s accuracy and its potentialimpact on clinical decision-making. Conclusions: The external validation of our predictivemodel demonstrates its robustness and generalizability across different populations andclinical settings. The integration of advanced diagnostic tools, such as AI and ML models,improves the accuracy in distinguishing between benign and malignant nodules, therebyoptimizing treatment strategies and minimizing invasive procedures. This approach notonly facilitates the early detection of cancer but also helps to avoid unnecessary surgeriesand biopsies, ultimately reducing patient morbidity and healthcare costs.
PB MDPI
SN 2075-4418
YR 2025
FD 2025
LK http://hdl.handle.net/10498/37881
UL http://hdl.handle.net/10498/37881
LA eng
DS Repositorio Institucional de la Universidad de Cádiz
RD 10-may-2026