The impact of Human Papilloma Virus status on the prediction of head and neck cancer chemoradiotherapy outcomes using the pre-treatment apparent diffusion coefficient

Objective: To determine the impact of Human Papilloma Virus (HPV) oropharyngeal cancer (OPC) status on the prediction of head and neck squamous cell cancer (HNSCC) chemoradiotherapy (CRT) outcomes with pre-treatment quantitative diffusion-weighted magnetic resonance imaging (DW-MRI). Methods: Following ethical approval, 65 participants (53 male, age 59.9 ± 7.86) underwent pre-treatment DW-MRI in this prospective cohort observational study. There were 46 HPV OPC and 19 other HNSCC cases with Stage III/IV HNSCC. Regions of interest (ROIs) (volume, largest area, core) at the primary tumour (n = 57) and largest pathological node (n = 59) were placed to analyse ADCmean and ADCmin. Unpaired t-test or Mann–Whitney test evaluated the impact of HPV OPC status and clinical parameters on their prediction of post-CRT 2 year locoregional and disease-free survival (LRFS and DFS). Multivariate logistic regression compared significant variables with 2 year outcomes. Results: On univariate analysis of all participants, the primary tumour area ADCmean was predictive of 2 year LRFS (p = 0.04). However, only the HPV OPC diagnosis (LFRS p = 0.03; DFS p = 0.02) predicted outcomes on multivariate analysis. None of the pre-treatment ADC values were predictive of 2 year DFS in the HPV OPC subgroup (p = 0.21–0.68). Amongst participants without 2 year disease-free survival, HPV-OPC was found to have much lower primary tumour ADCmean values than other HNSCC. Conclusion: Knowledge of HPV OPC status is required in order to determine the impact of the pre-treatment ADC values on post-CRT outcomes in HNSCC. Advances in knowledge: Pre-treatment ADCmean and ADCmin values acquired using different ROI methods are not predictive of 2 year survival outcomes in HPV OPC.


INTRODUCTION
Head and neck squamous cell cancer (HNSCC) is the seventh commonest cancer worldwide. 1 Concomitant chemoradiotherapy (CRT) is the standard of care for advanced locoregional disease at most head and neck tumour sites. However, treatment fails in more than 30% of advanced stage tumours, and salvage surgery for locoregional recurrence has a potentially poor outcome. 2 Identifying tumours that may not respond to CRT provides prognostic information, guides more proactive surveillance, and prompts more intensive or modified treatment plans such as radiotherapy dose escalation or surgery.
One potential imaging biomarker is the pre-treatment tumour apparent diffusion coefficient (ADC) on diffusionweighted MRI (DW-MRI). A number of studies have assessed the value of pre-treatment ADC mean in predicting HNSCC treatment response.  Whilst many have demonstrated that that a high pre-treatment ADC mean predicts a

Participants
Participants were recruited for a prospective single centre cohort observational study between 2014 and 2017 (http://www. controlled-trials.com/ISRCTN58327080) following Research Ethics Committee approval (REC reference 13/LO/1876) and informed consent.
Participants were eligible if there was histologically confirmed Stage 3 or 4 primary HNSCC without distant metastases, an axial section delineating an 1 cm 2 area of measurable primary tumour and/or pathological nodal tumour on standard clinicoradiological staging, and curative primary CRT was planned. Data collection included demographic and clinical information (age, gender, tumour subsite and tumour stage). Exclusion criteria were prior chemoradiotherapy, an ECOG performance status >2, lack of capacity to understand the patient information sheet or inability to provide informed consent, and known allergy to gadolinium-based contrast medium or eGFR <30 ml/min.
Sample size calculation was based on 70% of patients demonstrating 2 year disease free survival (DFS). Based on previous reports, 5,7,11,12,19 the ADC measurements were predicted to have a standard deviation of 40 and to demonstrate a difference of 50 × 10 −6 mm 2 /s between participants with and without 2 year survival outcomes. Using a 5% significance level and 80% power, it was calculated that at least 40 subjects would be required for the study without any accounting for loss of data.
HPV status, biopsies and treatment HPV status was analysed for all OPC as per standard of care. Non-OPC HNSCC was not routinely tested for HPV status, according to international guidelines. 32 HPV status was evaluated with p16 testing using an immune-stain or high-risk HPV DNA testing using in situ hybridisation. Biopsies at diagnosis were obtained from the primary tumour (n = 56), lymph node (n = 7) or both (n = 2). Intensity modulated radiotherapy (IMRT) was delivered as per the standard of care which was 7 Gy in 35 fractions; 2 Gy per fraction delivered once daily, 5 days a week. Concomitant intravenous cisplatin at a dose of 35 mg/m 2 every 7 days, starting on Day 1 of radiotherapy, was administered in all patients with adequate GFR and no contraindications (n = 47) with carboplatin (n = 16) or radiotherapy alone (n = 2) being used in the remaining participants.
The cohort was subsequently divided into HPV OPC and all other HNSCC (including HPV negative OPC) for analysis as in previous studies. 4,13 This approach was adopted, since HPV OPC is a distinct entity with unique histopathology and favourable treatment outcomes relative to all other HNSCC and the prognostic impact of p16 positive status is restricted to OPC only. 33,34 MR imaging Protocol and technique Participants underwent 1.5 Tesla MRI (Magnetom Aera, Siemens Healthcare, Erlangen, Germany) using a surface phased array 20 channel neck coil. The routine MRI protocol (Table 1) was supplemented by a research axial echo planar diffusion-weighted sequence with multiple b-values (0, 50, 100, 800 and 1500 s/mm 2 ) and the following scan parameters: repetition time 5900 ms, echo time 60 ms, two signal averages, FOV 240 x 240 mm, slice thickness 4 mm with a 0.5 mm slice gap. Mono-exponential ADC maps were calculated from the b = 100 and b = 800 values.

MRI processing and analysis
The ROIs were placed by a radiologist (MA, 3 years of experience) under the supervision of another radiologist (SC, 21 years of experience). The first 22 primary tumours and 24 nodes were also independently analysed by another radiologist (AG, 5 years of experience) to assess interobserver agreement.
Free hand ROIs were placed using OsiriX v. 8.0.2, open source Mac-based medical image processing software with the images magnified to a standard 300%. They were defined on the DWI b = 800 s/mm 2 map, but with access to the post-gadolinium fat saturated T1 axial sequence. Three separate ROIs were placed individually within the primary tumour and/or largest pathologically appearing lymph node (Figures 1 and 2): • Volumetric ROI (vROI) placed on multiple sections to encompass the whole lesion volume. • Area ROI (aROI) placed around the largest single crosssectional area of the lesion. • Representative area ROI (rROI) placed on the largest crosssectional area in the core of the lesion. This focused on an area of increased DWI signal on the b = 800 s/mm 2 map but excluded any areas of necrosis defined by cross-referencing to areas of either high signal on the b = 0 map or absence of gadolinium enhancement.
Areas of peritumoural inflammation characterised by high signal on the b = 0 map and avid gadolinium enhancement were avoided on all ROIs. An ADC map was generated from the b100 and b800 s/mm 2 images to mitigate the perfusion effects. ADC mean and standard deviation were recorded with ADC min calculated as ADC mean -one standard deviation (SD). 21 The vROI was also used to define the volume of the primary tumour and/or largest lymph node using the summation of areas technique. A ROI was also placed within the cervical spinal cord on the ADC map, and ADC mean was calculated as a reference for any non-patient related variability or artefact.     BJR HPV status and prediction of head and neck cancer outcomes with ADC A standard of care 12 week post-treatment positron emission tomography computed tomography (PET CT) study was used to guide clinical management. A quantitative evaluation of the posttreatment 18 F-FDG PET-CT was also obtained. This comprised a 6 mm diameter volume of interest (VOI) being placed at the site of most intense FDG uptake within either the primary lesion and/or the largest lymph node. If there was reduced uptake on the post-treatment images relative to background, a 6 mm VOI was placed at the same site as the post-treatment MRI ROI. If necrosis was identified within a lesion, the area of necrosis was excluded. The SUV max was calculated with semi-automated software on a Hermes workstation (Hermes Gold 3, Stockholm).

Statistical analysis
Statistical analysis was performed using Stata (v. 15.1) with a p value of < 0.05 being considered statistically significant.
Descriptive statistics summarised the patient demographics, HPV status, tumour site and staging. Pearson's correlation coefficient was used to measure the relationship between pre-treatment ADC mean or ADC min values and the 12 week post-CRT 18 F-FDG PET-CT SUV max for all participants.
The Intraclass correlation coefficients (ICCs) for each of the vROI, aROI and rROI at primary tumour and nodal locations were evaluated in the sample analysed by two observers.

Participant and tumour characteristics
The participant flowchart is summarised in Figure 3.

Analysis of ADC values and 2-year outcomes
The ADC mean and ADC min values obtained at the primary tumour site were significantly lower in HPV OPC participants than other HNSCC participants for all ROI analyses (p < 0.001). However, the the nodal ADC values did not differ between HPV OPC and other HNSCC subgroups (p = 0.3-0.61).
On univariate analysis of ADC parameters and 2 year survival outcomes for all participants, the only statistically significant association was that demonstrated between the primary tumour aROI ADC mean and 2 year LRFS (p = 0.04, Figure 3) (Table 3). HPV OPC status was also predictive of 2 year LRFS (p = 0.03) and DFS (p = 0.02) on univariate analysis (Table 4).
The subgroup analysis did not demonstrate any association between HPV OPC ADC values and 2 year DFS (p = 0.21-0.68); however, due to the limited sample size 2 year LRFS analysis was not possible. Similarly, ADC values were unable to predict 2 year DFS in other HNSCC (p = 0.66-0.85), although the sample size was limited. In participants without 2 year DFS, the ADC values were markedly lower in HPV OPC compared to other HNSCC (Table 5). For instance ADC mean at the primary tumour site was 863 ± 74 x 10 −6 mm 2 /s for HPV OPC (n = 4) whilst 1138 ± 179 x 10 −6 mm 2 /s for other HNSCC (n = 6), however, the sample size without 2 year DFS was too small for statistical comparison.
There was no significant correlation (p < 0.05) demonstrated between the pre-treatment ADC mean or ADC min values and the 12 week post-CRT 18 F-FDG PET-CT SUV max when applying any of the different ROI methods at either primary tumour and nodal locations.

DISCUSSION
This study provides data concerning the impact of HPV OPC status on the ability of pre-treatment ADC values to predict survival outcomes. On multivariate analysis, the HPV OPC status was the only independent variable predicting both 2 year DFS and LRFS, and the apparent association between the primary tumour aROI ADC mean and 2 year LRFS detected on univariate analysis was no longer evident. The pre-treatment primary tumour ADC values were lower in HPV OPC than other HNSCC, both overall and in those without 2 year survival. In this study cohort, it was not possible to demonstrate any prognostic ability of pre-treatment ADC values, when stratified by HPV OPC status.
Many previous studies have shown that an increased pre-treatment ADC is associated CRT treatment failure in HNSCC. [6][7][8]11,12,15,[18][19][20][21] It has been proposed that this may relate to the histological findings of higher stromal content, lower cellularity and micronecrosis, all of which are known to be associated with greater treatment resistance. However, a broad range of ADC thresholds have been reported for the prediction of treatment response and some studies have shown no significant association between the ADC values and outcome. [3][4][5]9,10,13,16,[22][23][24][25] Furthermore, one study has revealed the conflicting finding of poorer outcomes in tumours with lower ADC Previous data also suggest that HPV OPC is associated with lower pre-treatment ADC values compared to other HNSCC, and this finding is corroborated by our data. [27][28][29][30] Despite this, studies correlating ADC values with post-CRT survival outcomes generally comprise heterogenous cohorts and only a few provide data on HPV OPC status 15,17 or analyse HPV OPC cohorts separately. 4,13,24,35,36 There are only four studies which have considered HPV OPC status as a co-variate. 4,14,15,24 Ravanelli et al 24 conducted a retrospective study comparing pre-treatment ADC mean at the primary oropharyngeal tumour site in both HPV positive and negative OPC. In keeping with the results of this study, they also demonstrated that primary tumour aROI ADC mean predicted survival outcomes on univariate analysis but not on multivariate analysis. 24 Cao et al 4 and Martens et al 14,15 adopted a similar approach to our study and compared HPV OPC with other HNSCC at both primary tumour and nodal sites. Cao et al 4 concluded that pre-treatment ADC values in HPV OPC were not associated with treatment failure; however, unlike the present study, they did demonstrate a correlation in their larger cohort of other HNSCC. Martens et al 14,15 performed both retrospective and prospective studies and did not find pre-treatment ADC mean to be predictive of survival outcomes in a multivariate analysis which included HPV status. The current study adds to the existing literature by analysing pre-treatment ADC values in HPV OPC and other HNSCC cohorts with a more comprehensive methodology, which included multiple approaches to contouring ROIs and the application of both ADC mean and ADC min .
The favourable treatment response of low ADC tumours in previous studies [6][7][8]11,12,15,[18][19][20][21] has been considered counterintuitive, since low ADC values are generally found in poorly differentiated HNSCC, which would be expected to have a poorer prognosis. However, it could be postulated that the association of lower ADC values with HPV OPC status is primarily responsible for the apparent improvement in outcomes. Such a hypothesis is supported by our data, since multivariate analysis indicated that primary tumour ADC mean was no longer of prognostic significance, and it was only HPV OPC status that was an independent factor associated with 2 year survival.
An awareness of HPV OPC status is clearly of importance in order to interpret the significance of pre-treatment ADC values. The markedly lower primary tumour ADC values in HPV OPC non-survivors compared to other HNSCC non-survivors would preclude the prediction of treatment failure without information on HPV OPC status. Similarly, the wide range of the primary tumour pre-treatment ADC mean values in participants that did respond to treatment in HPV OPC (e.g. aROI 686-1684 x 10 −6 mm 2 /s) would also limit the prediction on an individual basis without this knowledge.
Another factor to consider in the interpretation of previous literature, is that different methods of contouring ROIs have been variably applied. Few studies have applied multiple methods in the same cohort and assessed the impact on the prognostic value of measurements obtained. Whilst smaller ROIs are potentially less representative, they may benefit from the ability to exclude macroscopic necrosis, so obtaining a value for the "viable tumour" and avoiding partial volume effects. Of the previous studies investigating the prognostic ability of ADC values according to HPV OPC status, one study measured aROI 24 whilst the other evaluated vROI. 4,13 Our data indicate that the choice of ROI analysis may potentially influence the ability to predict survival outcomes.
A further aspect of the methodology in the current study was the acquisition of both ADC mean and ADC min parameters. Histological heterogeneity within tumour tissue may not be accurately reflected by the ADC mean . Whilst ADC histogram analysis is being explored, 10 simple ADC parameters such as ADC min may probe the different tumour cell populations. Since our vROI and aROI tumour contours were potentially influenced by partial volume effects from the air-filled lumen, the ADC min was calculated as one standard deviation below the mean. However, this study did not demonstrate any benefit to using ADC min in predicting treatment outcomes. Interestingly, Martens et al demonstrated the maximum ADC value to be a valuable prognostic pre-treatment parameter. 13 There are potential criticisms concerning the study design, most of them fundamentally relating to erroneous assumptions prior to the power calculation. Whilst the sample size was comparable to other similar studies in the literature, the prospectively recruited study population comprised an unexpectedly high prevalence of HPV OPC (46/65). The dominance of the HPV OPC subgroup led to a smaller number of HNSCC participants which limited the scope of robust statistical comparisons with survival outcomes. Moreover, it is recognised that HPV OPC is associated with a lower rate of treatment failure than the 30% applied to the power calculation 2 with a wider standard deviation of ADC values. 25 This introduces the potential for Type 2 errors, thus also limiting the ability of pretreatment ADC values to predict treatment failure in the HPV OPC subgroup. However, the similarity between ADC values in HPV OPC participants with and without 2 year LRFS and DFS (Tables 3 and 5) would argue that the inability to predict survival represents a true negative finding. Nonetheless, it is proposed that larger cohorts are required to adequately power future similar studies of both HPV OPC and other HNSCC participants. 22 Finally, the assessment of interobserver agreement statistics was suboptimal since it was only performed in a proportion of the participants; however, this was representative of the whole sample with respect to tumour site and HPV status.
In conclusion, this comprehensive methodology adds to the recent data concerning the impact of HPV OPC status on the ability of pre-treatment ADC values to predict survival outcomes. The apparent association between lower pre-treatment primary tumour ADC mean and 2 year LRFS, was not confirmed on multivariate analysis of all participants or in the subgroup analysis of the HPV OPC participants. A potential explanation for this is that the HPV OPC status, which is widely accepted to carry a favourable prognosis, also happens to demonstrate lower ADC mean values. Larger cohorts with separate analyses of subgroups stratified by HPV-OPC status are required in order to advance our understanding of the optimal utilisation of DW-MRI in the evaluation of treatment responses in HNSCC.