Age-specific risk of malignancy in pediatric thyroid cytology: Reframing ROM based on pre-test probability

Pediatric thyroid nodules are uncommon but have a higher malignancy rate than adult nodules. Existing Bethesda risk-of-malignancy (ROM) estimates are not age-stratified and are affected by verification bias. We aimed to generate age-specific ROM, likelihood ratios (LRs), and post-test malignancy probabilities for pediatric and young adult thyroid cytology.

We analysed 2728 thyroid fine-needle aspirations (FNAs) from patients aged 0-25 years across multiple tertiary centers (2000-2023). All cases were classified or reclassified using the 2023 Bethesda System and grouped into four age bands (0-8, 9-14, 15-18, and 19-25 years). We calculated lower-bound ROM (ROM overall; assuming nonoperated FNAs were benign) and surgery-only upper-bound ROM (ROM surgery; 991 of 2728 cases with histology). Age-specific pretest probabilities and Bethesda-category LRs were used to compute post-test malignancy probabilities with Bayes' theorem.

Among operated nodules, malignancy prevalence decreased with age (84.2% at 0-8 years to 64.6% at 19-25 years). Verification bias was substantial: malignancy prevalence was 24.6% (671/2728) under the lower-bound assumption but 67.7% (671/991) among surgical cases. Upper-bound ROM values were 2-10 times higher than lower-bound values across categories. Benign cytology showed low LRs (0.07-0.15), reducing post-test malignancy probability to 0-25%. Indeterminate categories showed age-related variation; for example, a follicular neoplasm diagnosis carried a ∼71% post-test risk in a 9-year-old versus ∼49% in a 24-year-old. AUS subtyping (nuclear vs other atypia) did not consistently separate ROM.

Age substantially modifies both pretest and post-test malignancy probabilities in pediatric thyroid cytology. An age-stratified LR framework helps quantify verification bias and provides individualized risk estimates to guide decisions about surgery, molecular testing, and follow-up.