Meta-analytic estimates of Risk of Malignancy (ROM) of cytopathology reporting systems: Lacunae, limitations and a proposed solution with an example from the IAC Yokohama system for breast cytopathology

BACKGROUND: Cytopathology reporting systems classify fine-needle aspiration biopsy (FNAB) specimens into ordered diagnostic categories; each associated with a category-specific risk of malignancy (ROM). ROM is widely used to guide clinical management; however, it is a post-test probability that depends on disease prevalence and category-specific likelihood ratios. Despite this, most meta-analyses report pooled ROM values without explicitly defining the underlying pretest probability, resulting in statistically invalid and clinically misleading estimates. METHODS: We reanalyzed data from a previously published meta-analysis of 22 studies evaluating the International Academy of Cytology (IAC) Yokohama System for reporting breast FNAB. A Bayesian multinomial mixed effects model was applied to estimate category-specific likelihood ratios. These were used to calculate post-test probabilities (ROM) across a wide range of pretest probabilities (5-95%), rather than reporting a single pooled ROM. The inadequate category was excluded from the primary analysis and included in sensitivity analyses. Model performance was assessed using a principled Bayesian workflow, including convergence diagnostics and posterior predictive checks. RESULTS: The model demonstrated good convergence and fit. Category-specific likelihood ratios showed a clear and ordered gradient across diagnostic categories. ROM varied substantially with changes in pretest probability, particularly for intermediate categories, highlighting the limitations of fixed ROM estimates. Sensitivity and specificity estimates at different diagnostic cutoffs were comparable to those of the original meta-analysis. CONCLUSION: ROM is not an intrinsic property of cytopathology categories but a context dependent post-test probability. Meta-analyses should report category-specific likelihood ratios or ROM across defined pretest probabilities to ensure statistical validity and clinical usefulness. The proposed approach provides a robust framework for interpreting multicategory cytopathology systems and should be adopted in future diagnostic accuracy studies.