Evaluating the diagnostic efficacy of multigene testing in non-diagnostic thyroid nodules by fine-needle aspiration cytology: A prospective cohort study

Patients eligible for this study were aged ≥18 years, had a solitary thyroid nodule, were evaluated for the first time, had nodules graded Thyroid Imaging Reporting and Data System (TI-RADS) 4 or 5 by ultrasound, remained Bethesda I on second FNAC (4), agreed to provide specimens for molecular testing, and had undergone thyroid surgery. FNAC was performed using a 25-gauge needle following institutional protocols. Samples for molecular testing were collected by rinsing residual material from all aspiration passes into a preservative solution tube and stored at –20°C. Multigene testing was performed using next-generation sequencing technology. In total, 41 molecular alterations associated with thyroid cancer, including 16 gene mutations and 26 gene fusions, were examined. Ribonucleic acid (RNA) and deoxyribonucleic acid (DNA) were co-extracted from FNAC samples using microRNA and DNA co-extraction technology. Extracted nucleic acids were qualitatively analyzed to construct a sequencing library, and sequencing was performed using the S5 sequencer (Life Technology™, USA). Comprehensive data quality control was applied to obtain mutation information, including low-frequency DNA mutations and RNA fusions. Multigene testing was performed in collaboration with Beijing Fanshengzi Genetic Technology Co., LTD. According to the commercial test report, samples harboring specific gene mutations (e.g., BRAF V600E) with mutation frequencies > 30% or gene fusions were classified as positive, while all other samples were considered negative. Data were analyzed using R statistical software, v4.4.1 (R Foundation for Statistical Computing; https://www.r-project.org/). The “rms” package was employed for logistic regression algorithms, nomogram construction, and calibration curve construction. Least absolute shrinkage and selection operator (LASSO) regression was used to identify a significant combination of risk factors for the diagnostic predictive model. Receiver operating characteristic (ROC) curves were plotted using the “pROC” package. Decision curve analysis (DCA) was performed using the “dca.R” function. Based on the sample size of the training set, the batch size was set to 20. Each cycle consisted of 10 iterations, and a total of 10 cycles were completed, resulting in 100 iterations overall. All statistical tests were two-tailed, with P < 0.05 indicating statistical significance.