Utility of magnetic resonance imaging in determining treatment response and local recurrence in nasopharyngeal carcinoma treated curatively

Background To determine the optimal timing of the first Magnetic Resonance Imaging (MRI) scan after curative-intent radiotherapy (RT) for nasopharyngeal carcinoma (NPC), and evaluate the role of MRI in surveillance for locoregional recurrence (LRR). Methods Patients with non-metastatic NPC treated radically who had at least one post-treatment MRI (ptMRI) done were included for analysis. ptMRI reports were retrospectively reviewed and categorised as complete response (CR), partial response/residual disease (PR) or indeterminate (ID). Patients with LRR were assessed to determine if initial detection was by MRI or clinical means. Univariable and multivariable Cox proportional hazard regression analysis were performed to identify independent factors associated with CR on ptMRIs. Results Between 2013 and 2017, 262 eligible patients were analysed, all treated with Intensity Modulated Radiotherapy (IMRT). Median time from end of RT to the first ptMRI was 93 days (range 32–346). Of the first ptMRIs, 88 (33.2%) were CR, 133 (50.2%) ID, and 44 (16.6%) PR. A second ptMRI was done for 104 (78.2%) of 133 patients with ID status. In this group, 77 (57.9%) of the subsequent MRI were determined to be CR, 21(15.8%) remained ID and 6 (4.5%) PR. T1 tumour stage and AJCC stage I were associated with increased CR rates on first ptMRI on multivariable analysis. ID status was more likely at 75–105 days (3 months +/− 15 days) vs 106–135 days (4 months +/− 15 days) post RT (OR 2.13, 95% CI 1.16–4.12, p = 0.024). LRR developed in 27 (10.1%) patients; 20 (74.1%) were initially detected through MRI, 3 (11.1%) by nasoendoscopy and 2 (7.4%) by PET-CT. Conclusion MRI is useful for detecting local recurrence or persistent disease after curative-intent treatment. Most patients will need more than one ptMRI to arrive at a definitive status. The rate of ID ptMRI may be reduced by delaying the first scan to around 4 months post RT.


Background
Nasopharyngeal Carcinoma (NPC) has a distinct ethnic and geographical distribution, and is common in Southern Chinese and South East Asian populations [1]. Nonmetastatic NPC is treated definitively with radiotherapy (RT) with or without chemotherapy. Treatment response has been closely associated with prognosis [2,3]. Magnetic Resonance Imaging (MRI)'s role in the initial staging of biopsy-proven NPC is well established [4][5][6][7]. However, the utility of MRI in the assessment of response to treatment and disease surveillance is less welldefined.
The time course of histological NPC regression has previously been demonstrated through serial biopsy to be around 12 weeks post treatment with 3-dimensional conformal RT (3D-CRT) [8]. As such, the first assessment of treatment response is typically scheduled at this time-point through a combination of clinical and endoscopic examination with cross-sectional imaging. In practice, owing to their invasive nature, biopsies are usually reserved for cases where there are suspicious endoscopic or radiological findings. This imparts greater importance for imaging modalities such as MRI to pick up persistent or recurrent disease in a timely and accurate fashion.
Through this retrospective study, we explore the reporting patterns of post treatment MRI (ptMRI) for NPC patients, and aim to determine the optimal timing for the first ptMRI in a real-life clinical setting. In addition, we review current evidence and evaluate the ability of MRI compared to other clinical or radiological surveillance modalities in detecting locoregional recurrences (LRR).

Patients
Approval for this study was obtained from the Institutional Review Board (IRB). The NPC database, comprising all patients with histologically-proven NPC treated in two tertiary hospitals in Singapore, was retrospectively reviewed. Patients with non-metastatic NPC of any histological subtype treated with curative intent by RT alone, concurrent chemoradiation (CCRT) with or without induction chemotherapy between February 2013 and July 2017 were included for analysis. Pre-treatment evaluation for all patients included a complete history and physical examination, endoscopic assessment with biopsy, and staging scans. Computed Tomography (CT) or MRI was used for local staging, positron-emission tomography (PET)-CT was done to exclude metastatic disease. The staging system used was the American Joint Committee on Cancer (AJCC) 7th edition. Histology was classified according to the World Health Organisation system. For inclusion, at least one post-treatment MRI (ptMRI) needed to be done within 1 year of RT completion. Patients who did not complete the prescribed course of RT or have ptMRIs done and reported in other local/overseas institutions were excluded.

Treatment
According to institutional guidelines, patients with stage I and node-negative stage II disease were treated with RT only. Node-positive stage II, stage III to IVB patients were treated with CCRT; those with T4 or N3 disease were also offered 2-3 cycles of induction chemotherapy. All patients received Intensity Modulated Radiotherapy (IMRT). CT simulation was done with administration of intravenous contrast; fusion with pre-treatment MRI was done for planning wherever possible. Treatment was carried out with patients in a supine position immobilised by a thermoplastic shell. Our treatment protocol closely follows that used in the Radiation Therapy Oncology Group (RTOG) 0615 trial [9]. All patients were prescribed a total dose of 69.96Gy in 33 fractions. Clinical Target Volume (CTV) is designated as the Gross Tumour Volume (GTV) with a circumferential margin of ≥5 mm; where tumour is in close proximity to critical structures such as the optic apparatus or brain stem this margin may be reduced accordingly. CTV 70 included gross disease in the nasopharynx and any overtly involved lymph nodes. High risk and low risk subclinical regions as outlined in the RTOG 0615 radiotherapy schema were prescribed 59.4Gy and 54Gy respectively. Concurrent weekly cisplatin-based chemotherapy (30-40 mg/m 2 ) was administered during RT. Treatment breaks were not specifically recorded.

MR imaging, interpretation and timing
Timing of ptMRIs was calculated from the last day of RT. As per institutional practice, all scans were either reported or verified by a radiologist with a special interest in head and neck imaging. There were minor variations in the MRI protocol between the two tertiary hospitals but in all patients, four key sequences were performed: axial T1-weighted, axial T1-weighted contrast-enhanced fat-suppressed, axial T2-weighted and diffusion-weighted series.
The first ptMRI reports were reviewed and coded as 'complete response' (CR), 'partial response' (PR) or 'indeterminate' (ID). For the first ptMRI, CR is defined as complete resolution of disease or absence of any residual tumour in both primary and nodal sites in the radiology report. In subsequent ptMRIs, CR also included stable posttreatment changes seen on later imaging. PR is defined as the presence of residual disease in either the primary and/ or affected nodes. Where the report indicated inability to definitively distinguish between residual tumour and post-treatment changes, or suggested clinical/endoscopic correlation, or repeat imaging, it was coded as ID.
To determine the optimal timing of first ptMRI, we arbitrated a cut-off point of 4 months +/− 15 days (106-135) compared with 3 months +/− 15 days (75-105). Based on current institutional practice, the first ptMRI is typically done at 3 months post RT. However, in our experience the rate of indeterminate outcomes is high when the scan is performed at this time point. We hypothesised that delaying the timing of the first ptMRI by 1 month (or 30 days), the proportion of ID ptMRIs may be reduced. This particular value was selected so as not to deviate too far from the common practice of 3month post-RT scans, whilst keeping in mind that further delays to first ptMRIs may compromise timely detection of residual disease.

Follow-up
Depending on extent of acute toxicity, patients were reviewed at weeks 2 and 4 post RT. Subsequently the follow-up schedule would be 2 to 3-monthly until end of year 2, 4-monthly in year 3 and 6-monthly in years 4 to 5. Post-treatment review comprises clinical and endoscopic examinations, and may be shared between the patient's Ear, Nose & Throat (ENT) surgeon or medical oncologist where appropriate. Although there is a lack of specific guidelines with regards to the schedule of post treatment imaging, the first ptMRI was typically done at around the 3-month mark. Subsequent ptMRIs were not mandated if the first scan was reported as a CR; but were ordered where clinically indicated, for example in patients with PR/ID on first ptMRIs, or those with suspicious clinical or endoscopic findings. Patients with LRR based on ptMRIs were identified and reviewed to determine the timing of recurrence, how it was first detected (clinically versus radiologically), and how it was managed. The follow-up period was calculated from the last day of RT to day of last medical encounter or death.

Statistical analysis
Descriptive analysis was done using frequency with percentage and median. Univariable Cox proportional hazard regression analysis was performed to look for association between various patient, disease and treatment characteristics with achieving CR on first and subsequent ptMRIs. Covariables with P-value of ≤0.05 in the univariable analysis were included in the multivariable Cox proportional hazard regression analysis to identify independent factors associated with CR. For all analyses, two-sided P-values of < 0.05 were considered to be statistically significant. Statistical analyses were performed using STATA version 14.0.
Multivariable analysis showed that T1 stage and AJCC Stage I were significantly associated with achieving CR on the first ptMRI (Odds Ratio [OR] 2.96, p = 0.036 and OR 1.78, p = 0.046 respectively). Receiving CCRT was approaching significance (OR 2.46, p = 0.054) ( Table 2).

Discussion
Outcomes for patients with NPC have improved over the years with the introduction of chemotherapy and IMRT. However, local failure in the form of residual or recurrent disease still occur in 10-30% of cases [10,11]. The assessment of treatment response and clinical surveillance after definitive therapy for NPC is important, in order to permit earlier recognition of local failure and initiation of salvage therapies. Typically, patients are assessed clinically with cranial nerve examination, neck palpation and endoscopic inspection, in combination with imaging such as CT, MRI or PET-CT. Any suspicious lesions are then biopsied to obtain histological confirmation.
Given that radiation can lead to anatomical distortion within the treatment field, identification of residual or recurrent disease is often challenging. Palpation for cervical lymph nodes may be limited by fibrosis of neck musculature. Post radiation endoscopic examination may only reveal subtle mucosal changes such as fullness of postnasal space (PNS), or a mass which may represent fibrosis, crust or slough rather than residual tumour [12], and submucosal or deep-seated recurrences may be missed. Sensitivity of endoscopic examination in detecting persistent disease after RT is only 40.4%. Similarly, endoscopic biopsies run the risk of sampling errors as residual tumour cells are often scattered in small clusters, resulting in a sensitivity at 6 weeks post RT of 59.3% [13].
Radiological assessment faces similar difficulties, and to date there is no consensus regarding the optimal imaging modality in the post treatment setting. The utility of MRI, whilst well established in the initial staging of biopsy-  [14,15]. The identification of skull base erosion is improved with contrast-enhanced fat-suppressed sequences [5]. However, when compared to PET-CT, MRI may be limited in its ability to distinguish between post RT changes often seen in the irradiated nasopharynx and neck e.g. tumour necrosis, tissue fibrosis and inflammation, from true viable tumour. Conversely, changes in tissue metabolism may precede changes in tumour morphology or volume. Liu et al in a systemic review concluded that PET-CT, with its ability for combined functional and anatomic assessment, had superior pooled sensitivity, specificity and overall diagnostic accuracy when compared to both CT and MRI [16]. PET-CT also has the added benefit of uncovering any systemic metastases within the same examination, which can impact on goals of further treatment.
Nevertheless, there are drawbacks to relying solely on PET-CT to uncover residual or recurrent NPC. Disease at the primary site is more accurately demonstrated on MRI rather than PET-CT (92.1% vs. 85.7%) according to Comoretto et al. [17]. The latter produced false negative findings especially where there is intracranial extension of disease. False positive results have also been associated with PET-CT when it is done too early post RT due to the residual inflammatory effects causing apparent increased glucose metabolism. It has been suggested that the PET-CT should be done 6 months or later post RT for optimal accuracy (sensitivity and specificity are 92 and 100% at 6 months or later vs 33 and 64% within 5 months) [18]. Additionally, cost and resource availability can be limiting factors for the prevailing use of PET-CT. In view of these considerations, it is likely that MRI and PET-CT should be complementary, in order to improve overall diagnostic accuracy for recurrent or residual disease [17]. Our data indicates that, within the boundaries of our institutional practice, most LRR are detected by MRI rather than non-MRI radiological modalities or clinical examination. If salvage surgery or RT is planned for LRR, the superior ability to determine extent of tissue invasion with MRI makes it preferable to PET-CT in guiding resectability or extent of re-treatment required. Another issue to address is the optimal timing of MRI in view of the potential diagnostic uncertainties post RT. Guidelines suggest varying timing of post treatment imaging ranging from 3 to 12 months [19][20][21]. The time course of NPC regression after definitive treatment has been studied histologically and radiologically. In 1999 Kwong et al followed 803 NPC patients treated with 3D-CRT with or without induction chemotherapy, through 2weekly endoscopic biopsies. 93.2% achieved histologic remission by 12 weeks post RT. [8] This formed the basis upon which many subsequent studies relied on when scheduling post treatment imaging. Li et al in 2017 challenged this paradigm with data from 556 NPC patients treated definitively with IMRT and followed up with serial MRIs [22]. All MRI scans were reviewed independently by two radiologists with extensive experience in head and neck cancer imaging. In this group, 83.3% of patients achieved a clinical complete response (cCR)defined as no evidence of residual tumour or nodal disease based on examination with MRI and flexible nasoendoscopy -at 3-4 months (early cCR), a figure which increased to 91.4% at 6-9 months (delayed cCR). Interestingly, prognosis of patients with a delayed cCR was no different to those with an early cCR, leading the authors to conclude that 6-9 months may be the best time point for assessment of maximal tumour response to IMRT.
The increasing use of IMRT as standard of care, as well as CCRT has been associated with delayed primary tumour regression through mechanisms which to date remain unclear [23]. Whilst results from both Li's and our study both suggest a lag time between histological and radiological tumour regression, a significantly lower proportion of patients in our series (32.8% vs 83.3%) achieved a radiological CR on the first ptMRI. This may be explained by the methodology of our study, which looks at the real-world radiology reports, rather than having one or more radiologists retrospectively reviewing imaging and coming to a binary outcome regarding the absence or presence of residual disease. We believe our method reflects real-life clinical practice more closely, where limits of certainty in imaging interpretation is accepted, and collective decision-making is undertaken in a multidisciplinary setting for indeterminate cases. In spite of this uncertainty, having the first ptMRI done at an earlier date may still be worthwhile in order to provide a crucial baseline which can inform future scans. In addition, there may be a window period between 3 and 4 months and 6-9 months where further investigations can lead to earlier diagnosis of residual or recurrent disease, permitting prompt initiation of salvage therapies. Indeed, our study offers evidence that the optimal timing of the first ptMRI should be 4, rather than 3 months to reduce the rates of uncertainty in radiology reports.
We acknowledge that this study has some limitations. Firstly, the retrospective design meant that patients were not followed up based on a standardised protocol. ptMRIs were done at the clinicians' discretion, resulting in a wide date range for the first (32-346 days) and subsequent ptMRIs. Of note, a proportion of patients who had ID/PR status did not go on to have subsequent ptMRIs, reflecting the pattern of real-life clinical practice. Depending on overall clinical suspicion of residual or recurrent disease, these patients may have undergone investigation with other imaging modalities such as PET-CT or had biopsies which confirmed or excluded presence of disease, thereby sidestepping the need for further serial MRIs. Secondly, the determination of ptMRI status was based primarily on the authors' review of actual radiology reports rather than centralised review by dedicated radiologist(s). Although the ptMRIs are all reported by radiologists with a special interest in head and neck imaging, their level of experience may be differing.

Conclusion
MRI has an important role in NPC surveillance compared to other imaging modalities and detected a majority of loco-regional recurrence in our series. However, most patients will require more than one ptMRI to reach a definitive status. The rate of ID scans may be reduced by delaying the first ptMRI to 4 months post-RT.