Detection and Oncologic Outcomes of Head and Neck Squamous Cell Carcinoma of Unknown Primary Origin

Discussion

Finding the optimal approach to diagnosis and treatment of SCCUP is uniquely challenging because of the lack of prospective studies and definitive evidence in this disease. Often, clinicians rely on clinical judgement to determine the most appropriate treatment plan (13). The American Society of Clinical Oncology has published guidelines for the diagnosis and treatment of SCCUP that insightfully summarize the current data in the field (14). Our study, like previous studies, found a great majority of these tumors in the cryptic tissues of the oropharynx, with a heavy majority showing association with HPV (15).

In our study, the detection methodology was largely dependent on the individual caregivers. Generally, the most common approach, after initial in office examination, was a PET/CT scan followed by direct laryngoscopy including biopsies of candidate sites with or without surgery to remove the tonsils and lingual tonsils. This is in line with the most current American Society of Clinical Oncology guidelines (14). Approximately half of the 133 patients who received a PET/CT demonstrated suspicious findings which directed biopsy sampling. With regards to biopsies, the most commonly biopsied sites were the oropharyngeal structures. The surface epithelium of the palatine and lingual tonsils is composed of deep invaginations into the lymphoid stroma, creating crypts that increase the surface area by nearly 700% (16). Within these crypts, the superficial cell layer is interrupted by migrating non-epithelial cells (such as lymphocytes) which cause breaks in the otherwise continuous barrier. This exposes the basement membrane, leaving it prone to deposition of viral particles (16). Therefore, these structures are increasingly susceptible to HPV infection. Unfortunately, the same cryptic organization also makes detection more challenging as small tumors may hide in the tonsillar folds, escaping detection.

Our primary detection rate of 51.1% is in line with previous studies. In a study of 110 patients, Ryan et al. identified the primary sites of 53% of tumors using a stepwise approach similar to the detection techniques used at our institution. Similar to our study, direct laryngoscopy identified a primary site in 31% of patients, while the rest were identified by palatine or lingual tonsillectomy (6). In our multivariable analysis, a lack of suspicious PET/CT findings was associated with lower odds of finding the primary site, perhaps indicating that these tumors are exceedingly small and difficult to locate. Previous studies have shown varying levels of utility for PET/CT scans. Noij et al. found that quantitative analysis using SUV_max resulted in a sensitivity and specificity of 81.3% and 93.3% respectively, in a small sample of 31 patients (17). However, a larger analysis of 190 patients found a sensitivity and specificity of 73.1% and 68.9%, respectively (18). In our study, sensitivity and specificity of PET/CT in ultimately detecting a cancer was 62.5% and 65.2%, respectively. Nevertheless, further studies are needed in order to determine whether PET/CT is a useful tool in ultimately affecting treatment decisions and outcomes. It cannot be used as the definitive tool to determine a primary site given the data, and suggestive findings should be confirmed histologically.

Lingual tonsillectomy has arisen as a tool with great promise in detecting unknown primary sites. In our study, four of eight diagnostic lingual tonsillectomies were positive (50%). All of these patients had their lingual tonsillectomy performed at the time of DL. In the majority of cases, the surgeon would preemptively discuss with the patient that a lingual tonsillectomy may be performed if no obvious tumor was discovered elsewhere. In other cases, a suspicious PET scan was the reason for lingual tonsillectomy. Ryan et al. reported eight primaries detected in 14 lingual tonsillectomies (57%) and an increase in the total detection rate from 44% to 66%, after introduction of lingual tonsillectomy to their institutional approach (6). In another study, Nagel et al. performed lingual tonsillectomy in 14 patients with negative biopsies, and detected eight primaries (57%) (19). Together, these data suggest a role for TORS lingual tonsillectomy in patients where a primary is not detected after biopsies and palatine tonsillectomy, even when base of tongue biopsies are negative. However, it is important to keep in mind that lingual tonsillectomy is associated with significant morbidities including postoperative pain, decreased swallowing function, and high risk of readmission (19, 20).

OS and DFS were excellent in our cohort, with five-year OS and DFS of 85% and 77%, respectively, driven by the high percentage of HPV associated disease. As with many previous studies, we found that HPV associated tumors carried a far better prognosis than HPV negative tumors (15, 21-23). DNA and mRNA evaluation in tumor samples has shown that a majority of HPV associated oropharyngeal tumors are HPV driven and that HPV DNA and p16 status are independent markers for prognosis (23).

The treatment of SCCUP consists of varying permutations of chemotherapy, radiation therapy, and neck dissection. Many previous studies have suggested the equivalency of treatment across modalities (24-26). We found significant differences in OS and DFS across treatment methods in our cohort. Treatments with CRT offered the highest OS and DFS among treatments in our cohort. We noted that there seemed to be a clear separation in DFS between treatment regimens that included CRT with or without LND versus those that did not (Figure 3B). In fact, five year DFS among patients that received CRT was 83.2% compared to 59.3% in those that did not (p<0.001, log-rank test). However, it is important to keep these findings in context. At our institution, CRT is the standard of care for all patients who meet appropriate criteria. In patients who did not receive CRT, it must be presumed that there was a reason (perhaps a medical or other contraindication). Therefore, this group is likely to have done worse. Similarly, the most aggressive treatments were offered to the most robust patients. We were also unable to control for other factors such as performance status in our multivariable analysis, which is a well identified prognostic factor. Therefore, this finding warrants further investigation.

To analyze the effects of all of these variables on OS and DFS, we used multivariable Cox proportional hazards regression models. When controlling for N stage, HPV status, primary site detection, and treatment modality, we found that cN3 disease predicted worse OS and treatment with CRT predicted better OS. Similarly, we found that HPV positive status (HR=0.26) and treatment with CRT (HR=0.38) predicted better DFS (Table II). Together, these data confirm that HPV associated tumors have a substantially better prognosis as compared to HPV negative tumors, and that treatment of SCCUP with CRT offers excellent outcomes in patients who are eligible.

The definition of SCCUP is difficult to pinpoint; the most recent ASCO guidelines do not explicitly define head and neck SCCUP (14). Therefore, we performed our analysis using two different definitions of SCCUP: one being the lack of primary site after in-office fiberoptic examination, and the other being lack of primary site prior to treatment. Knowledge of a primary site prior to treatment allows for a more focused treatment, such as limiting the radiation field to the known tumor instead of a larger field. This may lead to decreased complications due to treatment. However, we found that our outcomes held true regardless of SCCUP definition, resulting in increased robustness and validity of our findings.

It is important to note the limitations associated with our study. Due to the inherent limitations of the retrospective study design, we were unable to control for many factors that factored into the choice of detection and treatment methods. For example, choice of treatment was likely influenced by a number of factors such as functional status of the patient or physician experience, which could have led to substantial selection bias in our cohort. Similarly, choice of biopsy location was likely dependent on surgeon experience and patient history not covered by our variables. Given the nonrandom nature of these variations, our results must be assessed in this context. Further studies, ideally in prospective form, should aim to further investigate the associations we have found in this study.