Fons Slieker

Introduction

Oral squamous cell carcinoma involving the maxilla (MSCC) is a rare subtype of oral cancer [1]. In oral cancer research, MSCC is often grouped with more prevalent subtypes of oral cancer, like squamous cell carcinoma of the tongue, floor of mouth or lower alveolar process [2-4]. Sample sizes for MSCC are usually very small which makes grouping understandable. However, grouping puts interpretation of study results at risk of bias via heterogeneity and confounding. After all, MSCC is quite distinct from other types of oral squamous cell carcinoma, because of the specific maxillary anatomy and the involvement of the midface in advanced stages. Accordingly, this thesis set out to study multiple aspects of care for patients with MSCC to ultimately improve patient-specific care.

Key findings and implications

Detecting bone invasion
MSCC may invade adjacent bony structures and can even grow into the sinonasal cavities and beyond [5]. Surgical resection is the preferred treatment, so that complete removal of the tumour may be achieved [6, 7]. If bone invasion is present, en-bloc resection of the affected bone is necessary to achieve tumour free resection margins. Reliable imaging methods are essential for adequate planning of the surgical resection. Spiral Computed Tomography (CT) and Magnetic Resonance Imaging (MRI) have both been established as reliable imaging methods in oral cancer [8-11]. However, the reliability of spiral CT and MRI was primarily tested in either heterogeneous oral cancer groups in which MSCC patients were marginal subgroups [9, 11], or studies specifically focussed on mandibular invasion [8, 10]. That is why in chapter 1, the diagnostic value of spiral CT and MRI in detecting bone invasion of the maxilla was investigated. In the absence of metallic dental restorations, spiral CT could detect bone invasion more accurately than MRI; although this difference was not statistically significant. Therefore, during preoperative assessment the imaging method of choice may depend upon situational factors. Spiral CT might be preferable if the patient is less cooperative or claustrophobic. Limited clinical access to one imaging modality may also play a role; spiral CT-scans are generally cheaper and take less time to complete. On the other hand, MRI scans have low radiation burden and are less prone to imaging artefacts by metallic dental restorations and might therefore be preferable in certain cases [12-14]. An interesting development in CT-scanning is the Cone Beam Computed Tomography (CBCT). CBCT imaging requires less time, produces lower radiation dosage, generates higher spatial resolution and is available in most outpatient clinics. Furthermore, patients are not required to lay down, but can sit with their head in the natural position during the scanning procedure [15]. The value of CBCT in detecting bone invasion was previously tested in studies focussed on mandibular invasion [16, 17]. Hence, in chapter 2 the value of CBCT in detecting bone invasion of the maxilla was studied.

The results suggest that the accuracy of CBCT for the detection of bone invasion in MSCC patients was high, but observer dependent. There are two main causes of observer dependent differences. The first cause is differences in training and experience of the observers. Repeated training has been shown to improve the interpretation of imaging of several anomalies [18-20]. Joint evaluation of the scans and discussion might improve the diagnostic accuracy as well. The second cause of observer dependent differences may be the (lack of) scoring criteria. Standardised scoring and reporting have been shown to improve the interpretation of scans of the appendix, pulmonary oedema and adnexal masses [21-23] and may help with the correct interpretation of imaging in general [24]. Our results suggest that the use of specific criteria improves the interpretation of CBCT imaging. As of yet, clear peer-reviewed guidelines are lacking for the interpretation and reporting of CBCT images of oral cancer. Formats for structured reporting of spiral CT and MRI images have been widely adopted by radiologists, to describe the location of the primary tumour and its volumetric dimensions, the extent of soft tissue involvement in all dimensions, the extent of bony involvement in all dimensions and the nodal status [23]. Similar formats for CBCT reports are not yet in place. A CBCT report format has been proposed for use by dentists in general practice [24]. This format mentions all anatomical subheadings that may be depicted on a CBCT scan: paranasal sinuses, nasal cavity, airway, cervical spine, temporomandibular joint, dental findings, other findings and recommendations.

Risk factors and treatment outcomes
In chapter 3, a systematic review and meta-analysis of different treatments, risk factors and outcomes of MSCC was discussed. The pooled 5-year local recurrence rate was comparable across the included studies. The pooled 5-year overall survival rate was 53.7%. Some studies had noticeably lower survival rates, because their samples had a substantial proportion of cases with (chemo) radiotherapy as primary treatment [25-27] and a large proportion of cases with advanced tumour stages [26-29]. Furthermore, elective neck dissection was also associated with improved 5-year overall survival (OS) rates [30, 31]. In fact, the subgroup analysis of surgery only vs. surgery with (neo)adjuvant (chemo) radiotherapy resulted in a non-significant difference. This means that current (neo)adjuvant treatment protocols for adverse tumour characteristics successfully seem to improve overall survival rates for MSCC patients. Interestingly, the (neo)adjuvant treatment regimens were slightly different in all three studies of the subgroup analysis, but none were significantly better or worse [32-34]. One risk factor specific to MSCC was associated with lower rates of OS in multiple studies: posterior tumour extension defined as extension into the soft palate, infratemporal fossa, pterygoid muscles and pterygoid process [26, 33, 34].

Local recurrence and salvage treatment
Generally, locally recurrent tumours were the largest group of recurrent tumours in oral MSCC. Due to the complex anatomy, poor visibility and poor access, complete removal of maxillary tumours can be challenging, especially at the dorsal margins. In chapter 4, local recurrence and salvage treatment for MSCC were discussed in depth. Vascular invasion was significantly associated with the likelihood of local recurrence. There is currently no consensus on the optimal salvage treatment strategy for recurrent tumors involving the maxilla. Salvage surgery is often the treatment of choice, but it is frequently at the cost of morbidity and quality of life [35]. Furthermore, the type of salvage treatment was significantly associated with overall survival. Extensive salvage surgery should be considered with caution, as its value in survival may be debatable. Extensive salvage procedures may disturb the appearance and function while quality of life is particularly important in the final period of life.

Development and internal validation of a prediction model that can calculate overall mortality
Clinical parameters and the TNM-classification have been incorporated into prognostic models that calculate personalized survival probabilities of head and neck cancer patients in general, and for specific patient-groups, like octogenarian patients and advanced larynx cancer patients [36-39]. However, there were no prediction models that calculated the survival probabilities for MSCC patients specifically, nor were there prediction models that incorporated histopathological factors as predictors [36-41]. In chapter 5, statistically significant multivariate prediction models were successfully computed to calculate 2- and 5-year mortality probabilities with clinical- and histopathological predictors of MSCC-patients. The prediction models were statistically adjusted for the confounding effect of medical history and (lifestyle-related) comorbidities with the Charlson’s comorbidity index [42]. Both adjusted prediction models had good to moderate results of predictive accuracy-tests and were recalibrated with the heuristic shrinkage factor [43-48]. The risk of distorted analysis results caused by the effects of selection bias and confounding were minimalised by analysing only the MSCC-subgroup and excluding other oral cavity cancer groups. The prediction models can be accessed easily via: mscc.oncologyheadneck.com

Figure 1: QR-code for quick access to the prediction models on mscc.oncologyheadneck.com.

Midfacial defect management – an e-survey on the clinical practice in the Netherlands
The extent of the resection determines the size of the subsequent midfacial defect. If the postoperative midfacial defect is not managed properly, the loss of orofacial function and cosmetic mutilation may lead to severe loss of health-related quality of life [49, 50]. The midfacial defect can be managed by the placement of an obturator prosthesis, or by surgical reconstruction with pedicled flaps or vascularized free-flaps. There are currently no guidelines for evidence-based treatment options of midfacial defects. That is why a Dutch national e-survey was created. Twenty medical specialists in total (otolaryngologists and maxillofacial surgeons) participated and completed the survey. The questions contained in this survey covered various topics, such as: prosthetics, timing of surgical reconstruction, type of surgical reconstruction per class of midfacial defects. Survey participants preferred conventional obturator prostheses, implant-supported obturator prostheses and obturator prostheses with frame reconstruction almost equally. If implant placement is possible, the masticatory functionality and comfort in implant-supported obturator prosthetics are significantly better than conventional prosthetics. Implant-supported obturator prostheses provide better anchorage and thereby improve retention [51-54]. However, results on whether (implant-supported) obturator prostheses improve the quality of life over surgical reconstruction are contradictory [55, 56]. Surgical reconstruction results in improved word intelligibility and masticatory efficiency over obturator prosthetics [57, 58]. Most survey participants preferred primary surgical reconstruction over secondary surgical reconstruction. Secondary reconstruction is associated with higher rates of complications and flap failure [59-61]. The survey participants employed a wide variety of different surgical reconstructive procedures for different Brown classes of midfacial defects. Our results demonstrated that ‘pedicled flaps’ for Brown I and the ‘fibular composite free-flap’ for Brown II – VI were favoured most.

Future perspectives

Detection of bone invasion
Various combinations of imaging modalities have previously been investigated and imaging-algorithms have been developed to detect bone invasion of the mandible [6]. Comparable studies for the detection of bone invasion of the maxilla have never been conducted. A study that compares different imaging algorithms might be an interesting subject for further study to improve the accurate detection of bone invasion of the maxilla. Furthermore, the accuracy of CBCT for detecting bone invasion in MSCC and the interobserver agreement may improve by standardisation of interpretation and reporting of CBCT images. The incorporation of the essential CT/MRI reporting requirements of oral tumours [23] with the CBCT report format proposed by Miles et al. [24] might improve the interpretation and interobserver agreement of CBCT images. This can be studied in a prospective trial with multiple observers scoring and reporting CBCT images for bone invasion in MSCC.

Prognostic prediction modelling in the future
Although the confounder-adjusted logistic models were successfully computed, internally validated and recalibrated, the performance generality should be tested and externally validated in another sample. This will constitute a challenge, because the incidence of MSCC is very low. By using this developmental method of prediction model computation, more prediction models can be developed for other oral squamous cell carcinoma subtypes. Both patients and doctors would be greatly supported in their clinical decision-making if they have access to prediction models tailored to specific subtypes of oral squamous cell carcinoma. These prediction models might be valuable tools in oral cancer care, if certain characteristics, like ease of use, accuracy and regular calibration are prioritised and safeguarded.

General limitations

Researching MSCC is generally limited by the low incidence of MSCC [1]. This generally means two things: most studies have small sample sizes and most studies are retrospective in nature. In this thesis, the largest sample of analysed patients consisted of only 95 cases (chapter 4, 5). Although this number may seem big to some, in actuality a sample of 95 cases is small. These 95 cases were patients that were operated in a time span of 15 years (between 2000 – 2015). In other words, on average 6.3 MSCC patients were eligible for inclusion per year in the University Medical Center of Utrecht. Unfortunately, 6.3 patients per year does not constitute enough patient volume in order to perform prospective studies. Retrospective studies are at risk of information bias and therefore lower in rank of evidence. As a result, multicentre prospective studies are needed. In the Netherlands almost all head and neck cancer patients are treated in the 8 major head and neck centers and 6 preferred partners of the Dutch Head and Neck Society (NWHHT). Within the NWHHT many multicentre studies were conducted, allowing for successful prospective studies also in rare head and neck cancer subtypes, e.g., MSCC. Establishing specialised maxillary cancer care centers with a dedicated maxillary cancer team might improve patient volume and quality of care. After all, high patient volumes in specialized cancer centres are associated with better survival outcomes [62-64]. Accordingly, higher patient volumes might have beneficial effects on treatment outcomes and would facilitate research to be conducted for MSCC patients as well [65, 66].

General conclusions

In this thesis, several aspects of clinical care for MSCC patients were studied. The aim was to ultimately improve MSCC patient-specific care. The main findings are as follows:
• Both CT and CBCT are very accurate imaging methods to detect maxillary bone invasion. However, MRI is a suitable alternative if contraindications for CT are present or MRI is already made for other indications.
• The best treatment of MSCC available today is probably surgery including elective neck dissection, and adjuvant (chemo)radiation in case adverse tumour characteristics are present.
• Salvage surgery prolongs overall survival in case of small recurrence, but might have dubious value in survival regarding larger recurrences infiltrating adjacent facial structures.
• The overall 2- and 5-year mortality probability of MSCC can now be calculated with newly computed prediction models.
• A wide variety of reconstructive procedures are performed, but ‘pedicled flaps’ for Brown I and the ‘fibular composite free-flap’ for Brown II – VI were favoured most among Dutch specialists.

en Discussie

Introductie

Het plaveiselcelcarcinoom van de bovenkaak (de Engelse afkorting is MSCC) is een zeldzaam type mondholtekanker [1]. Omdat MSCC zo zeldzaam is, worden patiënten met MSCC in onderzoek vaak gecombineerd met patiënten die mondholtekanker hebben op andere plekken in de mond, zoals de tong, of de mondbodem [2-4]. Omdat er weinig patiënten zijn met MSCC is dit ook begrijpelijk. Hierdoor kan echter bias ontstaan, waardoor de resultaten van onderzoek over MSCC verkeerd geïnterpreteerd kunnen worden. In deze dissertatie werden meerdere aspecten van het behandeltraject van patiënten met MSCC onderzocht, met als doel om hiermee de zorg voor deze patiënten te verbeteren.

Belangrijkste bevindingen en implicaties

Botinvasie vaststellen
MSCC groeit vaak in naastliggend bot, in de sinonasale holtes van het aangezicht, of zelfs daar doorheen [5]. Chirurgische resectie van MSCC is de eerste keuze van een in opzet curatieve behandeling. Het doel van deze behandeling is om de tumor volledig te verwijderen [6, 7]. Als de tumor in naastliggend bot gegroeid is, zal ook een deel van het bot verwijderd moeten worden. Het is daarom belangrijk vast te stellen of er sprake is van ingroei in het bot [8-11]. In hoofdstuk 1 van deze dissertatie werd onderzocht of met spiraal-CT en MRI kan worden vastgesteld of botinvasie van de bovenkaak aanwezig was. In eerdere onderzoeken was het herkennen van botinvasie met spiraal-CT en MRI onderzocht in heterogene patiëntengroepen [9, 11], of onderzocht op patiënten met kanker van de onderkaak [8, 10]. In hoofdstuk 1 werd aangetoond dat zowel met spiraal-CT, als met MRI, botinvasie van de bovenkaak herkend kon worden. Daarom zal de keuze voor de beeldvorming afhankelijk zijn van situationele factoren. Zo is spiraal-CT meer geschikt voor minder coöperatieve patiënten of voor patiënten met claustrofobie, omdat scans gemakkelijker en sneller gemaakt kunnen worden. Tevens is spiraal-CT goedkoper dan MRI. Het voordeel van MRI is dat het een lagere stralingsbelasting heeft en dat er minder vaak beeldartefacten ten gevolge van dentale restauraties ontstaan, die de interpretatie bemoeilijken [12-14]. Daarnaast geeft MRI meer informatie over uitbreiding van de tumor in de weke delen. Een interessante ontwikkeling in CT-beeldvorming is de Cone Beam CT (CBCT). Het voordeel van CBCT-beeldvorming is dat het vervaardigen van een scan minder tijd kost dan het vervaardigen van een spiraal-CT. Ook heeft CBCT een lagere stralingsbelasting, een hogere resolutie en is bovendien beschikbaar in de meeste kaakchirurgische poliklinieken. Bovendien kan de scan zittend worden gemaakt met het hoofd van de patiënt in een natuurlijke rustpositie [15]. In hoofdstuk 2 van deze dissertatie werd onderzocht of botinvasie van de bovenkaak gedetecteerd kon worden met CBCT [16, 17]. De resultaten van hoofdstuk 2 laten zien dat met een CBCT zeer nauwkeurig botinvasie vastgesteld kan worden. De resultaten waren echter observant-afhankelijk. Er zijn twee