Assessing the diagnostic accuracy for pleomorphic adenoma and Warthin tumor by employing the Milan System for Reporting Salivary Gland Cytopathology: An international, multi-institutional study

Onken and Nishino have suggested that there is a standard method with which to calculate the risk of malignancy (ROM) within a formal diagnostic system. ROM calculation has been a topic of discussion for years for prior cytology reporting systems such as the Bethesda systems for the cervix and thyroid. In addition, the formerly introduced cytologic diagnostic systems generally are organized toward targeting a single diagnostic entity (eg, The Bethesda System for Reporting Thyroid Cytopathology and papillary thyroid carcinoma). The Milan System for Reporting Salivary Gland Cytopathology (MSRSGC) has the unique quality of encompassing many divergent diagnostic entities in each category with a complex navigation. In our experience using MSRSGC, the next stage of development for this unique diagnostic system comes from possibly separating ROM calculations for cytologic diagnoses that can be made with a relatively high degree of confidence (eg, Warthin tumor and pleomorphic adenoma [PA]). The objective of our study was to initiate that process and guide investigators in how data might be collected. Other data collection schemes may demonstrate usefulness. To calculate ROM, the data are collected in such a way as to include the maximum number of each desired subject while surgical diagnosis remains the gold standard whenever cytology and/or histology correlation is required. If the aim of a study is to calculate the ROM for a diagnostic category such as the atypical category, the data can be retrieved in a “forward” fashion, similar to what we did in our previous studiesor with a definitive diagnosis of PA and Warthin tumor as in our latest study. However, if the goal of the study is to calculate the ROM for a diagnostic entity such as PA, the search then must be both forward and backward to include all desired cases. Forward search methods enrich for cases that were more likely to have atypical clinical or imaging features that trigger surgery, perhaps aside from the cytology diagnosis, potentially resulting in a high ROM. In addition, forward searching potentially leaves out those cases that were diagnosed as PA on follow-up but were classified on cytology as another entity and/or differential diagnosis. Leaving out the misinterpreted and/ or variant fine-needle aspiration diagnoses and categories could falsely result in an improved diagnostic accuracy. “Backward” searching enriches for the definitively benign diagnoses that falsely lower the ROM, but it adds value by including a range of cytology diagnoses to gain a better understanding of which categories are being diagnosed for cases definitively classified as PA on histology. Therefore, both forward and backward methods were included to build this data set to optimize the calculation of ROM. The criticism of Onken and Nishino results from a conflation between the formal and the practical. The criticism is fair, but one easily could argue that the suggested forward search method is flawed and should cast doubt on the findings for such a study. The fact that the same methods did not produce such a reduction in the ROM for the Warthin tumor cohort suggests to us that our data were not due entirely to a methodological limitation. Our objective was to make a case for reciprocal evolution in the formal elements of future editions of MSRSGC to account for specific empirical features that are sufficiently generalizable to warrant elevation from the specific to the formal to account for the practical efforts of pathologists.