HNSCC STUDY

The definitive results of the HNSCC study were published in 1997 (29). This report is based upon the data updated to 1 March 1997. Clinicians taking part in the HNSCC study were also invited to take part in the cell kinetics study. CR analysis with the covariates was performed using data from patients with the information on the molecular markers and at the time of this analysis there were a total of 309 specimens assessable for all molecular markers.

Tumour outcome analysis was performed using 309 patients who had biological specimens for marker studies. Tumour outcome endpoints in this subset of patients were compared with the whole patient population.

Overall survival was defined as the time from randomisation to the time of death and if alive, patients were censored at the time of their last follow-up.

Local and nodal failures were recorded at the time when clinically definitive tumor growth was detected in the primary site or in the nodes. Distant failure was defined as appearance of metastatic disease outside the irradiated volume. All times were calculated from the date of randomisation.

Kaplan-Meier curves were formed for each endpoint and compared across treatment arms by the log-rank test.

2.2.1 Prognostic variables

All available prognostic factors, age, sex, WHO performance status, treatment, T stage, N stage, histological grade, site, Ki-67, p53, CD31, bcl-2, cyclin D1 scores and proliferative pattem, were assigned numerical scores and entered into the model. Optimisation of the variables was carried out by creating dummy variables and incremental coding where necessary. Cyclin D1 and age were the two continuous variables.

2.2.1.1 Molecular markers

Histologic material

Histologic material was obtained, retrospectively, from the referring hospitals. Each Pathology Department was requested either to provide the original blocks for processing at Mount Vemon Hospital or to cut up to twelve 4pm sections mounted onto poly-l-lysine coated slides. Each specimen was examined by Paul I. Richman at the pathology department to confirm the presence of SCC tumour and that the specimen was assessable. George D. Wilson and his group at the Gray Cancer Institute carried out the molecular analyses. As previously mentioned a total of 309 specimens were assessable for all molecular markers.

Immunohistochemical (IHC) staining

Sections of 4 pm were dried overnight at 37°C. Before antibody staining for each marker, the slides were blocked for endogenous peroxidase activity with a 3% solution of hydrogen peroxide in methanol for 30 minutes. Microwave irradiation was used to unmask the binding epitopes using either lOmM citric acid (ph 6.0) for Ki-67, p53, bcl-2 and CD31 or 1 mM EDTA (pH 6.0) in the case of cyclin D1. A cycle of 3 five-minute irradiations was used for all antibodies except bcl-2, which required only two cycles. The slides were then left to stand for 10 minutes in buffer at room temperature before being washed thoroughly in tap water. After three washes in Triethanolamine Buffered Saline (TBS), the slides were incubated in the different antibody in TBS containing 1 drop per ml of Dako Serum Free Protein block (Dako Ltd, High Wycombe, X0909) for Ih r at room temperature. The monoclonal antibodies to p53 (clone DO-7, Dako Ltd.) were used at a 1:75 dilution, CD31 (PECAM-1, Dako

Ltd.) at 1:30, bcl-2 (clone 124, Dako Ltd.) at 1:40 and cyclin D1 (Novacastra, Peterburgh, clone NCL-Cyclin D1-GM) at a dilution of 1:75. The antibody against Ki- 67 was a rabbit polyclonal (clone MIB-1, Dako Ltd.) and used at 1:75 dilution. After three further washes in TBS, biotinylated rabbit anti-mouse antibody (Dako Ltd E0354) or, in the case of Ki-67, biotinylated swine anti-rabbit antibody (Dako Ltd.) diluted 1:400 in TBS was applied for 1 hour at room temperature. After three further washes, ABC complex (Dako Ltd K0355) was added for 1 hour at room temperature. The staining was visualised by adding diaminobenzidine (Vector Labs. DAB kit SK 4100) for 5 minutes at room temperature. Cyclin D1 was stained at a later date and Dako Envision polymer kit (Dako Ltd., K4006) was added for 30mins. Following three further washes in TBS, Envision DAB (Dako Lt d) was added for five minutes. All slides were washed well in tap water and counterstained with Mayers Haematoxylin for lOsecs-lmin and then dehydrated, cleared and mounted in Dextropropoxyphene (DPX).

Assessment of staining

All slides were inspected for the presence of assessable tumour and semi- quantitatively analysed by a consultant pathologist (P.I.R) except in the case of cyclin D1.

For Ki-67 the slides were visually scanned and assigned to one of three scores; 1 = less than 20% positive cells, 2 = 20 to 40% positive and 3 = greater than 40% positivity. In addition, the proliferation pattem was assessed as previously described (30) where 1 = marginal (most organised), 2 = intermediate (mainly organised), 3 =

mixed (more than one pattem usually including random) and 4 = random (diffuse, disorganised staining).

Slides stained for bcl-2 were scored as negative if less than 5% of cells were stained and positive otherwise.

p53 protein was assessed in two ways 1) pattem of expression (whole specimen looked for staining) and p53 was classified as 1 =negative (less than 5% positive cells), 2=sporadic (5-75% cells positive) and 3=all (>75% positive cells); 2) variation in intensity was semiquantitated as strong, moderate or weak in those specimens showing positivity.

The vessel count identified by CD31 staining was classified in 10 high power fields as 1 =up to 35 vessels, 2=35-55 vessels and 3=greater than 55 vessels.

Cyclin D1 was assessed by manual counting aided by an in-house image acquisition system incorporating a CCIR-format 3-CCD chip colour camera coupled to a PC- based video-rate frame grabber. Software routines for image acquisition, normalisation and storage have been developed as have grids and manual counting data recording and export to spreadsheets. At least 10 high power fields (X40 objective) were counted for each specimen.

As an informal validation of the extraction of endpoints from the CHART databases, we reproduced the published results for tumour outcome endpoints by performing a conventional KM analysis and the log-rank test to compare the outcome across the treatment arms.

CHAPTER 3

COMPETING RISKS MODELLING OF FAILURE-SPECIFIC