4.1 Non-diagnostic

According to Faquin et al¹³ non-diagnostic FNA should not exceed 10% to fulfill quality requirements. In our cohort, a total of 8% of FNA and 6% of cases with available histology were non-diagnostic. Significant differences were detected in non-diagnostic rates between FNA performed by non-cytopathologists (29%) and cytopathologists at the UHZ (5%). Since no rapid on-site evaluation was conducted in either external or internal FNA, this can be excluded as cause of the difference. A possible explanation is the varying level of experience. This is also reflected by the fact that cell-poor or artifact-ridden smears were the main findings in non-diagnostic FNA (62.2%). Especially in pleomorphic adenomas (which represented 28.9% in the definite histologic workup in this group), the diagnostic yield can be low in unexperienced hands. In 13.4% cytology showed only normal tissue without lesional components, indicating a sampling error, maybe due to inadequate clinical information or incorrect radiological correlation. The size may also affect the difficulty of FNA. Accordingly, we found that the average diameter was slightly smaller than the average size in other categories, not including cystic lesions. Altogether, this suggests that FNA of SG lesions should be performed in experienced centers to keep the number of non-diagnostic results low. Another reason for non-diagnostic aspirates might be the low cellularity of the lesions themselves. Typical examples are hemangiomas, lipomas, and fibrosing lesions such as scars or atrophic tissue. However, some SG tumors are also rich in fibrous stroma, which can lead to non-diagnostic aspirates.

According to MSRSGC, FNA with exclusive non-mucinous cyst content should be classified as non-diagnostic,^{9, 13} since the spectrum of cystic-transformed lesions is extremely broad. In our daily practice, non-mucinous cystic lesions with complete regression after FNA matching ultrasound and clinical appearance are classified as benign cysts and placed in category II. This approach leads to a low percentage of non-diagnostic FNA. Several previous studies report a higher rate (>10%) of non-diagnostic cases when the MSRSGC criteria are strictly applied.^{6, 14, 15, 21, 23, 24, 27} In our study, unclear cystic lesions represented 24.4% of category I cases, which were finally diagnosed as cystic-transformed SG tumors in histology—most commonly Warthin tumors, but also other tumors with unusual pronounced cystic degeneration (e.g., pleomorphic adenoma, oncocytoma, cystadenoma, and mucoepidermoid carcinoma).

Our ROM in category I was slightly higher (26.7%) compared to original MSRSGC (25%) and previous studies (on average 15.7%, Table 3). This can be partly explained by the fact that we placed non-mucinous cysts—if clinically and radiologically appropriate for a cyst—in category II. Although this kept the rate of non-diagnostic FNA low, it led to an increase in ROM. If we had placed non-mucinous cysts strictly in category I, ROM in this category would have decreased to 20.0% and increased to 7.3% for category II. Additionally, only a few category I cases were resected (26.3%). This represents a selection bias for ROM and RON, which are calculated based only on cases with available histology. In reality, ROM and RON are therefore likely to be lower. A similar effect was observed in the studies of Lubin et al²⁵ and Mazzola et al.²² It is also possible that suspicious lesions underwent direct surgery without repeat FNA, whereas inconspicuous lesions continued to be observed and disappeared during follow-up. Another relevant finding was the higher percentage of non-diagnostic FNA performed by non-cytopathologists (29%) in comparison with cytopathologists (5%), resulting in different ROM values (33.4% versus 23.3%). Finally, a second FNA was performed in 23% of initially non-diagnostic cases, of which 86.5% were eventually diagnostic. This approach also supports the recommendations by MSRSGC to correlate lesions clinically and radiologically and to repeat FNA in doubtful cases.^{9, 13}

4.2 Non-neoplastic

This category included mainly reactive or inflammatory processes, such as abscesses, chronic or granulomatous inflammation, accounting for 32.9%. In our study, 25.7% of cases were cystic non-neoplastic lesions with a broad spectrum in the resected specimens like epidermoid cysts, ranulae, mucoceles, thyroglossal duct cysts, lymphoepithelial cysts, branchiogenic cysts, and others, but also 17.1% cystic-transformed benign tumors. 10% of cases were reactive lymph nodes.

Category II quantitatively accounted for the second largest category (32%) of all performed SG FNA. However, only 10.1% of these lesions were resected, so ROM is probably overestimated. Nevertheless, at 5.7%, our ROM was well below the 10% ROM rate reported by the MSRSGC and below the ROM rate of most previously conducted studies as well (Table 3). Overall, false-negative results were limited to 4 out of 70 cases (5.7%), half of which were lymphomas (misinterpreted as reactive lymph nodes). This reflects the results of the studies by Rossi et al,⁹ Viswanathan et al,¹⁵ and Song et al,¹⁸ which also found B-cell lymphomas as a main reason of false-negative diagnoses. In general, diagnosis and subclassification of low-grade lymphoma are well known limitations of FNAC.²⁸ Moreover, the FNA diagnosis of “reactive or normal intraparotid lymph node” (n = 13) was only accurate in 53.8% of subsequent excisions. Apart from the two already mentioned low-grade lymphomas, two lymph node metastases (one sclerotic variant of a mucoepidermoid carcinoma and one melanoma) were found in the resections—both small and in the marginal area of a lymph node, indicating a sampling error. Furthermore, two Warthin tumors were misinterpreted once as a lymph node and once as a lymphoepithelial cyst. Although the abovementioned cases were cytologically classified as non-neoplastic, they were nevertheless subsequently resected due to persistent clinical and/or radiological suspicion for malignancy.

A striking feature in our study was the high RON of 24.3% in this category. In addition to the aforementioned malignant neoplasms, this category also contained benign neoplasms like Warthin tumors (9/70) and cystadenomas (2/70), which were cytologically interpreted predominantly as cysts and occasionally as inflammatory lesions or lymph nodes. Subtotally cystic regressed neoplasms are inadequately identified in this regard. Previously performed studies that provided data for RON, reported similar or even higher RON rates with 21.4%¹⁷ and 47.6%²¹ but with a smaller number of cases (n = 14 and n = 21).

4.3 Atypia of undetermined significance (AUS)

According to the MSRSGC, this category should not exceed 10%.^{9, 13} Our rates were 5% overall and 6.6% for the cases with correlating histopathological diagnosis, which showed a wide spectrum of benign and malignant processes: Warthin tumor (12/50), lymphoma (7/50), pleomorphic adenoma (5/50), cysts (5/50), reactive lymph nodes (4/50), inflammatory changes (4/50), and squamous cell carcinomas (4/50). One typical cytological finding in this diagnostically inconclusive category is a monotonous population of small lymphocytes that may indicate low-grade lymphoma but may also be present in Warthin tumors, intraglandular lymph node metastasis, or inflammatory processes.⁵ ROM in this category was 34% and thus clearly higher than the provided 20% of the MSRSGC. However, a comparison with previously conducted studies showed that most ROM in the AUS category were above the proposed 20% of MSRSGC (in average 26.9%, Table 3). Finally, a second FNA was performed in 13% of initially as AUS interpreted cases in our study, of which 71.4% were eventually diagnostic. This approach and the relatively high ROM support the recommendations by MSRSGC to repeat FNA or to go for surgery.

4.4 Benign neoplasm

The numerically largest category was category IVA (48.7% of FNAC with histopathological follow-up). Half of the excised lesions in this category were pleomorphic adenomas and 41.4% Warthin tumors. Accordingly, RON was high (99.2%), whereas ROM was very low (1.1%), adhering to the established ROM of <5% by MSRSGC. Overall, 74.5% of pleomorphic adenomas and 80.9% of Warthin tumors could be accurately diagnosed in our study. The high accuracy can be explained by straight forward diagnostic features. Nevertheless, four false-negative results were found, one being a lymphoma and one a mucoepidermoid carcinoma, both misinterpreted as Warthin tumor. The other two false-negative diagnoses were an adenoid cystic carcinoma and an epithelial-myoepithelial carcinoma, which were misinterpreted as pleomorphic adenomas.

A second FNA was performed in 2% of category IVA cases. In most cases, it was performed with an average interval of 2.4 years in Warthin tumors (60%) that were repunctured due to an increase in size during follow-up.

4.5 SUMP

This category includes neoplasms for which no specific subtype can be determined cytologically, and therefore, no classification into benign and malignant can be made.⁹ Smears are mainly highly cellular without striking nuclear atypia, resulting in morphologic overlap between benign tumors and low-grade carcinomas. A typical benign example is the cellular variant of pleomorphic adenoma, which was indeed the most common histology in this category (n = 45, 57.7%), followed by other benign tumors, such as lymphocyte poor Warthin tumor and basal cell adenoma (n = 6, 7.7% each). However, also malignant tumors were diagnosed including three adenoid cystic carcinomas, three epithelial-myoepithelial carcinomas, two myoepithelial carcinomas, two salivary duct carcinomas, two squamous cell carcinomas and others. The ROM value was 21.8%, lower than the MSRSGC data (35%) and lower than most previous studies. Liu et al²⁹ and Mazzola et al²² showed similar low ROM for the SUMP category. They considered a possible selection bias because of their low number of cases in this category: Liu et al²⁹ with 54 cases, Mazzola et al²² with 61 cases (our study with 78 cases).

In this category, MSRSGC recommends conservative surgery as further clinical management to clarify the diagnosis and behavior of the tumor. From a surgical point of view, a definite diagnosis would significantly influence planning and extent of resection. Although category IVB accounted for only 5% in our study, reducing the number of cases in this category should be the goal.

In the last years, an increasing number of SG neoplasms have been characterized on a molecular level.^{30, 31} Therefore, ancillary tests such as immunohistochemistry, fluorescence in situ hybridization or sequencing-based methods will probably further significantly increase the diagnostic accuracy of preoperative cytologic specimens and reduce the number of cases of SUMP, but also in the uncertain categories III and V.^{30, 31} Recently, the SG neoplasm specific next generation sequencing panel (“SalvGlandDx”) has been developed at the UHZ. It offers the advantage of simultaneous detection of aberrant expression, gene mutations, and fusions by RNA extraction. This NGS-based panel can also be applied to cell block material from FNA and is suitable as a useful all-in-one tool for diagnosis and more accurate classification of SG neoplasms.³² A study investigating this aspect is currently underway at the UHZ.

4.6 Suspicious for malignancy

This category contained the fewest cases (3.3% with histopathological follow-up). ROM and RON were both high (92%). The ROM provided by MSRSGC is 60%. In comparison with previous studies, our value lies in the upper range, but all studies showed values above 60% with an average of 84.8% (Table 3).

Half of the definite histopathological diagnoses were different types of lymphomas, which were also correctly classified in the previous FNAC. There were two false-positive diagnoses: In one case, cytology was assessed as “cells with atypia, suspicious for malignancy”, whereas histology revealed a benign cyst with excessive degenerative atypia. The other was diagnosed as “suspicious for a malignant neuroendocrine neoplasm” based on immunohistochemical examination of the cell block. Final histological workup showed a central fibrotic area in otherwise normal parenchyma without evidence of malignancy. The interval between FNA and surgery was approximately 1 month. A potential sampling error was discussed, but the patient was lost for follow-up.

4.7 Malignant

15.7% of all FNAC with histopathological follow-up were classified malignant. ROM and RON were high with 99.2% and 100%, respectively, and thus higher than the provided data by MSRSGC (90%). All reviewed previous studies in the literature reported ROM > 90%. In our study, the most frequently diagnosed malignant histopathological subtypes in this category were squamous cell carcinoma (17.8%), lymphoma (16.9%), and melanoma (14.4%). The most frequently diagnosed malignant tumors originating from the SG were acinic cell carcinoma (9.3%) and salivary duct carcinoma (8.5%). We detected one false-positive case: A Warthin tumor with squamous metaplasia was misdiagnosed as a lymph node metastasis of a squamous cell carcinoma in FNAC.

All histopathologic entities in every MSRSGC category are provided in Table S2.