Effectiveness of Head-and-Neck Molecular Imaging Reporting and Data System Criterion in Head-and-Neck Squamous Cell Carcinoma PostConcurrent Chemoradiotherapy

Introduction

Two out of three patients with head-and-neck squamous cell carcinoma (HNSCC) presented with locally advanced (LA) cancer and were treated with concurrent chemo radiotherapy (CRT) with a 5-year survival rate of around 50%.[12] Response to treatment has been crucial in determining subsequent management.[3] Due to simultaneous postradiation alterations, the standard morphological response approach has been deemed problematic. Consequently, numerous ¹⁸F-fluorodeoxyglucose positron emission tomography/computed tomography (FDG PET/CT)-based molecular response approaches have been devised and found to be efficient in preventing needless neck dissection.[45] Nonetheless, inter-class heterogeneity, a poor positive predictive value (PPV), time dependence, and a lack of validation studies were the most significant issues with these techniques.[6] Consequently, a specialized, well-structured criterion was required in this case. The purpose of this study was to propose and evaluate a semi-quantitative molecular response criterion for these patients.

Materials and Methods

In this retrospective analysis, LA-HNSCC patients treated with CRT during 2011–2018 were screened. Patients with an FDG PET/CT study at 3–4 months post-CRT with a minimum of 2 years of follow-ups were included. A baseline PET/CT scan was not a requirement. Patients with a previous history of surgery, chemotherapy, second malignancies, and nonsquamous cell carcinoma histology were excluded. Nasopharyngeal carcinomas, indeterminate lung nodules, and only radiotherapy treatment were the other exclusion criteria. This study was approved by the Hospital Institutional Review Board and has granted a waiver from consenting (IRB-BHR/92/2022, July 28, 2022).

Results

Out of screen 314 patients with head-and-neck malignancies, 83 patients were included for final analysis [Figure 1]. Patient characteristics were tabulated [Table 2]. FDG PET/CT scan was done at a mean of 3.2 months post-CRT (range 3.0–3.9 months). Posttreatment mean (range) SUVmax of T, N, background, mediastinum, and liver were 4.3 (16–1.6), 3.0 (26.0–1.0), 2.3 (5.0–1.0), 1.9 (4.0–1.0), and 2.8 (6.0–1.4), respectively. Metastasis was found in 11/83 patients (13.3%) at 3 months. Out of which 7 patients were proven by histopathology. Of remaining 72 patients, 39, 2, and 31 patients had a CR (Scores 1, 2, 3A), equivocal response (score 3B), and PR (Score 4, 5) respectively as per HAN-MI-RADS criterion. At 3 months, 33/72 patients’ histopathology was available. Per patient sensitivity, specificity, PPV, NPV, and accuracy at 3 months were 95.8%, 44.4%, 82.1%, 80.0%, and 81.8%. Out of 72, at 1 year 40 and at 2 years 47 patient's histopathology was available. For remaining patients at one and 2 years, clinical and imaging findings were used for disease status. Diagnostic statistics for predicting loco-regional disease up to 1 and 2 years were 93.3%, 92.5%, 90.3%, 94.9%, 92.9%, and 84.9%, 91.9%, 90.3%, 87.2%, and 88.6%, respectively. For the Hopkins score, of 72 patients, 25, 16, and 31 patients had a CR (Score 1, 2), equivocal response (score 3A, 3B), and PR (Score 4, 5) respectively. Indeed, 14 patients of the CR group by HAN-MI-RADS criterion (score 3A) become equivocal in the Hopkins criteria. This was because the Hopkins score 3 category included diffuse uptake more than liver (score 3A, 3B). For Hopkins score, patient-wise sensitivity, specificity, PPV, NPV, and accuracy at 3 months were 95.8%, 28.6%, 82.1%, 66.7%, and 80.7%. Whereas, similar diagnostic statistics for predicting loco-regional disease up to 1 and 2 years were noticed for it and which were 96.5%, 88.9%, 90.3%, 96.0%, and 92.9%. One year and two years of DFI for each HAN-MI-RADS score showed a statistically significant difference [Table 3]. DFI hazards ratio (HR) for score 3A was near to score 2, while it was near to score 4 for score 3B. Hence, we found that diffuse uptake more than the liver but <2 times, had a lesser risk of harboring disease. One year and two years of OS for each HAN-MI-RADS score showed no statistically significant difference [Table 4]. One year and two years of DFI and OS of each response group (CR, PR, and equivocal) for both HAN-MI-RADS and Hopkins scores were calculated [Tables 5-8]. Statistically, a significant difference was found for DFI but not for OS on intra-comparison for response groups for both. On inter-comparison, by Log-Rank test, the P value to compare the DFI of Hopkins and HAN-MI-RADS was 0.5897. It was nonsignificant. The ROC curve analysis showed a statistically significant difference (area under the curve [AUC] difference 0.184, P = 0.02, Z statistic 2.327) in the AUC of HAN-MI-RADS (AUC 0.884, 95% confidence interval [CI] 0.785–0.948) and Hopkins (AUC 0.699, 95% CI 0.578–0.803) which suggested better disease outcome predictability of HAN-MI-RADS [Figure 2]. Overall, 2 years recurrence rate was 59.0% (49/83), while 19.3% (16/83) patients developed distant metastasis.

Close

Figure 1 STARD flowchart for patient selection. HNSCC: Head and neck squamous cell carcinoma, FDG: ¹⁸F-fluorodeoxyglucose, PET/CT: Positron emission tomography/computed tomography, CRT: Chemo radiotherapy

Export to PPT

Close

Figure 2 Receiver operating characteristic curves for the two response criteria. HAN-MI-RADS: Head and neck molecular imaging-reporting and data system

Export to PPT

Discussion

In oncology, the evaluation of a patient's response to initial treatment was crucial for determining the next course of treatment. In clinical practice, RECIST 1.1 was recognized as the gold standard for solid tumors.[10] However, the situation was more difficult in head and neck cancers due to underlying radiation-induced alterations. Recent research has employed a limited number of molecular qualitative response criteria with different degrees of effectiveness.[11] A Deauville score of five points compares tracer activity with mediastinum or Liver.[12] However, it did not account for the type of uptake (diffuse or focal), which is crucial in a postradiation context. Neck imaging reporting and data systems (NI-RADS) was a criterion developed by the American College of Radiology that required validation.[13] The criteria classified the pattern of tracer uptake as mass-like or nonmass-like and the level of tracer uptake as moderate to moderately intense, although neither the mediastinum nor the Liver was included as comparators. This will be essential for a criterion's objectivity and reproducibility. In our criterion, we endeavored to account for all the missing details.

Already, there was a drift of less aggressive management of patients with negative PET/CT post-CRT in HANSCC following the PET-NECK trial.[4] This was a prospective randomized controlled trial that concluded that PET/CT guided surveillance was noninferior to planned neck dissection, sparing approximately 80% of futile neck dissections with similar survival and more cost-effectiveness. Although, PET/CT response criterion utilized here was qualitative, limited to the neck node, and not well elaborated. Normal mucosa in the local disease site shows significant postradiation changes, hence, more clarity on the objective comparator was needed. Hopkins score was the first well-defined criteria for response assessment in this setting.[8] A prospective multicenter “The ECLYPS” study (n = 125) suggested a reliable FDG PET/CT-based surveillance at 12 weeks with sensitivity, specificity, PPV, NPV, and accuracy of 65.2%, 91.2%, 62.5%, 92.15% and 86.4% respectively with a gradual decrease in sensitivity with time.[13] Hence, an additional surveillance scan at 1 year was suggested. Recently, posttreatment FDG-PET/CT Hopkins criteria showed an accuracy of 85% for predicting locoregional recurrence after definitive radiotherapy for oropharyngeal squamous cell carcinoma.[14] In our study, HAN-MI-RADS sensitivity was near similar at 3 months, and 1 year with an accuracy of 88.6% up to 2 years. Hopkins score showed no significant time dependency in our study and a better accuracy of 92.9%.

We used SUVmax measurement rather than purely visual inspection for score interpretation, which may have impacted this differently than the original work done by Marcus et al.[8] Further, we used background (contralateral normal structure in the neck), and mediastinal blood pool (arch of the aorta) for scores 1 and 2 allocations, rather than internal jugular vein, However, because the mean liver background was higher than both background and mediastinum, hence this have not made a change in score category. We believed that this newly devised system will have a better inter-observer agreement. The main difference in HAN-MI-RADS and Hopkins scores was the equivocal response category. In our experience, a diffuse moderate metabolic activity (> liver but ≤2 times of liver) has a low chance of harbouring disease while a severe diffuse metabolic activity (>2 times of liver) needs caution, hence equivocal in nature. Indeed, category 3A was found to have HR close to Category 2. Hence, we found that splitting category 3 into two (3A, 3B) was worth better risk stratifications and planning further strategy of follow-up.

In addition, several shortcomings were observed in our study. Primarily, there was a retrospective character, a small patient population, and heterogeneous main locations. We eliminated nasopharyngeal carcinoma because it behaves differently than other cancers of the head and neck and is related to a different oncovirus.[15] In addition, human papillomavirus has emerged as an important prognostic marker in oropharynx cancers.[16] We did not have access to this information during our data analysis. Only two patients were presented in the ambiguous HAN-MI-RADS group (Score 3B), which was too small to be statistically meaningful and may have impaired the overall picture. The OS HR was greatest for score 3B [Table 4], which we felt was incorrect due to the limited sample size. In this study, the majority of patients belonged to stages T3 and N2, hence additional research is required to determine the applicability of these results to all stages. In addition, we evaluated the data per patient, and diagnostic statistics for T and N separately require additional large-scale research. Not only, HPE was not accessible in all instances, but it was also impractical.

Conclusions

A five points HAN-MI-RADS criterion was found to be promising for response assessment with 88.6% diagnostic accuracy for predicting loco-regional disease recurrence up to 2 years. Comparative analysis showed less equivocal results and statistically significant higher AUC than Hopkins. Further, prospective studies analysis with larger homogeneous patients will provide clinical validity.