Comparative analysis of diagnostic ultrasound and histopathology for detecting cervical lymph node metastases in head and neck cancer

This noninterventional, prospective, monocentric clinical study included patients who were referred to our internal US department from 01/01/2007 to 06/15/2016. All patients included had a histologically confirmed carcinoma of the head and neck region or a relapse thereof and were scheduled for examination of the presence of cervical LN metastases by US. The obtained findings were to be correlated with the histopathologic results, requiring at least excision of one LN up to and including comprehensive ND as part of the diagnosis or treatment of the underlying disease.

Exclusion criteria have been: diagnosis other than head and neck cancer, inability to passively participate in US (duration 30–45 min), age less than 18 years, induction treatment by radiation or drug therapy and treatment without LN dissection.

We planned to enroll among the study cohort at least 200 patients with therapy-naïve primary head and neck squamous cell carcinoma (HNSCC) who were assigned for surgical therapy upon consultation of the multidisciplinary tumor board. The study protocol was reviewed and approved by the ethics committee of Heinrich-Heine-University Düsseldorf (# 4231). All patients provided written consent for the sampling of LNs and data collection.

US examinations were performed by KCS (level III sonologist for Head and Neck ultrasound of the German Society for Ultrasound in Medicine (DEGUM)) on the day before surgery using the Acuson Antares System Premium Edition and the VFX 13-5 Multi-D linear probe (both Siemens Medical Solutions®, Mountain View, CA, USA) on individually tailored preset “lymph nodes”, which had been further optimized for the individual AAO-HNS neck LN levels I, II, and III (Robbins et al. 2002), each as described previously (Heusch et al. 2014; Sproll et al. 2017; Sproll et al. 2021). In the ultrasound (US) examination, each lymph node (LN) was individually evaluated based on its imaging features and anatomical location. The measurements were carefully documented, capturing four specific distances to provide a comprehensive geometric description of the LN. The descriptions of the four distances measured are the following: Longest Cross-Sectional Diameter (l): This is the maximum distance measured across the lymph node in the plane that yields the largest cross-sectional area. Essentially, it is the longest axis that can be drawn through the lymph node in a single 2D plane. Longest Perpendicular Diameter (b): once the longest cross-sectional diameter (l) is determined, the longest perpendicular diameter (b) is the maximum distance measured at a 90° angle to “l” within the same 2D plane. Longest diameter perpendicular to the plane (d1): This distance is measured by identifying the longest diameter that lies perpendicular to the plane containing the longest cross-sectional diameter (l) and the longest perpendicular diameter (b). Longest diameter perpendicular to d1 (d2): After establishing the longest diameter perpendicular to the plane (d1), the distance d2 is the maximum length measured perpendicular to d1, essentially capturing the third dimension of the LN if one considers l and d1 as the other two dimensions. These measurements collectively aim to provide a multidimensional assessment of the LNs size and shape, assisting in further diagnostic evaluations (Fig. 1). The LN was then depicted schematically on a diagram containing the six AAO-HNS neck LN levels (Suppl. Fig. 1a) (Som et al. 1999, 2000). Additionally, subjective US findings were recorded for each LN. For more details on the subjective assessment of the US results, please refer to Suppl. Doc. 1.

The result of the subjective LN examination was an assessment of the LN as either containing metastasis or tumor-free. Patient treatment was carried out as described in Suppl. Doc. 1 (NCCN 2018; Wolff 2019).

ND specimens were processed in the operating room to retrieve each LN identified by US, and each LN was recorded on the patient’s LN map, as described previously [Supporting information No. 1 in (Sproll et al. 2017)]. Each LN was then freed from surrounding fibrofatty tissue, placed in an individually labeled container with Formaldehyde Solution 4% phosphate buffered for histology (neoFroxx, Einhausen, Germany) and sent for routine histopathological assessment (Suppl. Fig. 2). The remaining ND specimen was dissected according to the AAO-HNS neck LN (sub) level (1–6) and assessed for each neck LN level by histopathology (Robbins et al. 2002). Details of histopathological evaluation are given in Suppl. Doc. 1. pTNM staging was performed according to the sixth and seventh editions of the TNM classification of malignant tumors, as amended at the time (Sobin et al. 2011). If metastases were found in the histopathological assessment, their maximum diameters (in mm) were measured in two planes.

Study endpoints were the agreement of sonographic and histopathological findings as well as the optimal cutoff values for the individual diameters and the roundness indices in the prediction of metastatic disease. All patient-related data (clinical course, US and pathology) were stored in a special GCP-compliant individually designed database we have named “Lymphsono” based on the OmniComm® Electronic Data Capture Software broadly used in our study center (OmniComm Systems, Inc.; Fort Lauderdale, FL, USA) (Suppl. Fig. 3). Access was controlled to ensure that the data could be viewed and edited only by authorized personnel. The database has been continuously adapted and improved, and changes have been documented in the program history to keep a record of the development of the program.

Count data are presented as the total number of LNs in the patient cohort and the mean and median (range) number of LNs per patient. Data were extracted from the OmniComm database, and all statistical calculations were executed in R (R Development Core Team 2020). Fisher’s exact test for contingency tables larger than 2 × 2 based on FEXACT (Mehta and Patel 1986) improved as described by Clarkson et al. (1993) was used. Two indices were used to describe roundness: the two-dimensional index (the Solbiati index), namely, 2D roundness = \(\frac{l}{b}\) (Solbiati et al. 1992) (Fig. 1) and a three-dimensional index, namely, 3D roundness = \(\left(\frac{l}{b}\right)\)² × \(\left(\frac{l}{d1}\right)\)² × \(\left(\frac{l}{d2}\right)\)². Both indices assume a value of 1 for a sphere and increase as l increases compared to the other distances, but the 3D index suggested a greater spread and may have better performance.

For analysis of count data, exact binomial tests and Fisher’s exact tests were employed (both using the statistical package). OCP analysis was carried out using the pROC package in R (Altman and Bland 1994a, b; Robin et al. 2011). The unweighted Cohen kappa index (κ) was calculated using the “irr” package in R (Cohen 1960).

For the subjective ultrasound findings, we calculated the sensitivity, specificity, positive, and negative predictive value (PPV and NPV, respectively) and likelihood ratio (LR) (Altman and Bland 1994a, b). For individual distances l, b, d1, and d2 (Fig. 1) and the two roundness indices, optimal cutoff points (OCPs) were determined by the maximal Youden index (YI) (t_Y = max_c (Se (c) + Sp (c) − 1)) for each individual neck level using the pROC package in R (Robin et al. 2011). The area under the curve (AUC), YI, sensitivity, specificity, PPV, and NPV were also determined using the OCP. Furthermore, the performance of the subjective ultrasound findings, the two roundness indices and the individual distances (l, b, d1, and d2) were determined depending on the sizes of the metastases detected by histopathological analysis (≤ 2 mm, > 2 and ≤ 4 mm, > 4 and ≤ 6 mm, > 6 and ≤ 8 mm, > 8 and ≤ 10 mm, and > 10 mm). On the basis of the OCPs of individual distances l, b, d1, and d2, the two indices for each individual neck level (1–6) (Robbins et al. 2002) and the subjective ultrasound findings for individual LNs, a clinical N stage (cN stage) was determined for each patient and compared to the pathologically determined N stage (pN stage). These classifications were carried out both for individual N stages (Sobin et al. 2011) and for the summary categories “metastasis present or not present” (N + vs. N0). Cohen’s kappa index was determined for each individual distance, the two roundness indices and the subjective ultrasound findings. The Landis and Koch classification was applied to systematize data with values between 0 (worthless) and 1 (perfect) (Landis and Koch 1977).

In total, 235 patients were enrolled, comprising 97 women (41.28%) and 138 men (58.72%). The average age at the time of tumor surgery was 64.9 years (median 66.0 years). 202 of them suffered from primary therapy-naïve HNSCC (lip mucosa to oropharynx), 3 from cervical CUP syndrome (2 of SCC and 1 of melanoma), 4 with verrucous carcinoma of the oral cavity, 1 with undifferentiated sinus carcinoma, 6 with salivary gland adenocarcinoma, 1 with carcinoma of the Meibomian glands, 1 with malignant melanoma, 1 with recurrence of mucoepidermoid carcinoma, 1 with recurrence of melanoma, and 15 with recurrence of HNSCC (lip mucosa to oropharynx). Among these patients, there were 248 surgical treatment sessions with varying extents of ND.

Overall, 4539 (1–42, mean 18.38, median 18 per patient) LNs were identified by US preoperatively. Of these, 265 (5.84%; 0–16, mean 1.07, median 0 per patient) of the identified LNs were classified as suspicious of malignancy according to subjective sonographic findings. The remaining LNs were categorized as reactively enlarged and tumor free. Based on our subjective assessment of LNs, each patient was assigned a cN status preoperatively.

During subsequent tumor surgery, 7237 (1–126, mean 29.30, median 24 per patient) LNs were removed from tissue obtained by ND and analyzed by histopathology. Among these, metastases were found in 443 LNs (6.12%; 0–57, mean 1.79, median 0 per patient). The pN classification was based on histopathological results and used as a reference for the abovementioned cN classification. A total of 2684 LNs were evaluated using both methods, and their results compared. Of the LNs evaluated by both methods, 2304 (85.8%) were identified individually from the dissected neck specimens in the operating room (OR), marked on the neck diagram of the corresponding patient analogously to the US findings and then sent in individual containers for 1:1 histopathological assessment (Suppl. Fig. 3). An additional 380 LNs (14.2%) were sent along with the specimens, which were still dissected in the OR, allotted according to the AAO-HNS neck LN level and analyzed separately per neck LN level (Robbins et al. 2002). Of the 2684 LNs, 259 (9.65%) were classified as suspicious for metastasis by US, but 299 (11.14%) were found to be positive by histopathology. Discordant findings were found in 86 cases (23 false positive (Suppl. Fig. 4a–c) and 63 false negative (Suppl. Fig. 5a–c) results by US). According to histopathology, pN status was pN0 in 133, pN1 in 41, pN2 in 2, pN2a in 3, pN2b in 45, pN2c in 19, and pN3 in 5 cases.

To maximize the performance of the objective criteria in predicting tumor involvement, the OCPs for individual diameters and indices per level (I–VI) and sublevel (IA–B, IIA–B, and VA–B) were determined by ROC analysis (Table 1). There was a clear spread among the values across neck LN levels. For example, the 2D roundness value for level IA was 1.5 and that for level III was 2.01. For diameter l, the OCP was 6.1 mm for level IV and 14.95 mm for level IIA. For individual diameters, there was a general spread from the smallest to largest value by a factor of two (Table 1). Determination of the individual OCPs per level reflects the shape variations of the LNs in the individual levels and led to an improvement in the prediction of metastatic disease.

To determine the performance of the objective criteria and the subjective ultrasound findings as a function of the size of the metastasis, we assessed metastasis size by measuring the diameters of the tumor deposits inside the LNs in two dimensions microscopically, and the longest diameter was used for evaluation. Because of the otherwise insufficient number of LNs, the same OCPs (the respective ideal values for all levels (I–VI), Table 1, under “all”), were jointly used for all LNs from the different neck LN levels except for subjective ultrasound findings. Metastases were categorized based on the UICC classification as follows: isolated tumor cells (ITCs) and micrometastases had a longest diameter ≤ 2 mm (Sobin et al. 2011), and macrometastases, with a longest diameter > 2 mm, were grouped in 2-mm increments up to a maximum diameter of 10 mm, and as metastases with a longest diameter > 10 mm. Among all criteria (indices and subjective US findings), the predictive performance increased (linearly) with increasing size of the metastases (Table 2). The 3D roundness index was not superior to the 2D index. Among the individual distances, d2 showed the best performance with a cutoff point of 6.45 mm (Table 2). The data also showed that subjective assessment is insufficient for the detection of micrometastases, ITCs and macrometastases up to 8 mm.

Beyond analyses based on single LNs, we wanted to determine the contribution of sonography to N-staging of the patients. To do this, each patient was assigned a cN status on the basis of measurements of the individual distances l, b, d1, and d2 (with respective cutoffs) and subjective US findings, and the predictive performance of the cN status was assessed by examining the pathological N status (pN status; Table 3).

In the present study, the clinical relevance of pathological upstaging, specifically from cN0 to cN+, merits particular attention. As delineated in Table 3a, among the cohort of 151 patients initially designated as cN0 based on ultrasonographic evaluation, 128 were histopathologically confirmed as pN0. Conversely, 17 patients were upstaged to pN1, 2 to pN2b, and 4 to pN2c. A comprehensive analysis of the LNs accountable for the diagnostic discordance is elucidated in Suppl. Table 1. The ultrasonographic dimensions of the LNs ranged from 8 to 29 mm (mean = 16.01 mm, median = 15.7 mm), whereas the histological dimensions varied from 4.5 to 20 mm (mean = 10.6 mm, median = 10 mm). Furthermore, the dimensions of the metastatic foci within the LNs spanned 0.3 to 8 mm (mean = 3.67 mm, median = 3.2 mm) (Suppl. Table 1). All deviations of the cN values from the pN values were first included unweighted in Cohen’s κ calculation. Regarding the roundness indices, the 3D index (κ = 0.330) was superior to the 2D index (κ = 0.271). Among the individual distances, d2 with κ = 0.438 remained the best at correctly predicting pN status, whereas l was the worst. Overall, subjective US findings performed best, at κ = 0.713. Because any weighting (e.g., weighted Cohen’s κ (Cohen 1968)) would have led to enhancement of errors, we did not calculate a weighted Cohen’s κ because the performance was already too poor to be reliable. However, the performance improved somewhat by allowing only a binary decision, namely, cN+ vs. cN0. Again, the 3D roundness index (κ = 0.3504) was superior to the 2D roundness index (κ = 0.2517), and among single distances, d2 (κ = 0.4462) was best for making the N+ vs. N0 decision (Table 4). As before, the subjective ultrasound findings were best, with κ = 0.7706, but they were still below the limit of “almost perfect” (κ = 0.8) (Watson and Petrie 2010). In 4 cases, cN status was correctly anticipated despite the wrong LN being categorized as being suspected of containing a metastasis.

In summary, we evaluated objective and subjective morphologic US criteria to predict metastases in neck LNs. They show different values depending on the neck level and their performance in micrometastases is insufficient, but improves with increasing size of metastases.