The recognition that cancers of the same site, stage and morphology have divergent natural history and respond differently to therapeutic interventions have inspired many to search for predictive markers of response to therapy. Until recently, a large body of research on HNSCC has focused on identifying surgical-pathologic features associated with poor outcome, such as the number and location of lymph node involvement, presence of extra-nodal extension, positive surgical margin, and perineural invasion (35
). Although clustering of such surgical pathologic adverse features could identify patients who did relatively well after radical surgery alone or followed by adjuvant radiotherapy (40
) with or without chemotherapy (6
), further work is needed to guide clinical research and practice. Finding prognostic-predictive markers is particularly important for personalizing non-surgical therapy, such as radiation with or without chemotherapy or new class of agents, because surgical-pathologic data are not available in this setting.
Quantification of EGFR expression has been controversial with semi-quantitative scoring system ranging 0-3+. However, our previously study (13
) on a well-defined cohort of patients with locally advanced HNSCC randomized to receive standard radiotherapy alone revealed that pretreatment EGFR expression level of the primary tumor measured by image analysis-based IHC assay was a robust predictor for tumor response to radiotherapy, more so than the T-stage. This data is consistent with other image-guided quantitative scoring systems (17
). Before promoting any biomarkers for clinical application, it is crucial to determine the reproducibility of the assay, to validate its predictive value for radiation response, and to assess whether and to what extent this assay added value to the known clinical prognostic variables. This follow-up study was conducted using a well-defined group of patients enrolled into the same prospective phase III trial but randomized to receive a different radiotherapy regimen. These results establish that the EGFR assay was remarkably reproducible and confirmed the lack of correlation between the levels of EGFR expression with clinical parameters, and validated its value in predicting tumor response to radiotherapy when measured quantitatively.
Currently, there are many proposed laboratory biomarkers that are prognostic of clinical outcomes. However, there are only limited studies in HNSCC to build upon the clinical prognostic factors that are already known to be robust and often are better than expensive laboratory assays, and few that combine multiple laboratory biomarkers. To examine this issue, we pooled the current clinical and EGFR data with those of the previous series and added the results of additional three molecular assays (p53, Ki-67, and MVD) for further analysis after dividing the study population into training and test sets. Furthermore, with recent data regarding the HPV status as a strong prognostic marker and to put our results in perspective, we assayed the remaining unstained oropharyngeal carcinoma slides for p16, a surrogate marker for HPV, and determined the positivity to be approximately 33% (or approximately 20% of total number of tumor samples), probably due to lower prevalence of HPV infection in the early 1990's (unpublished data). Additional analysis incorporating the HPV data is still ongoing. Out of four markers further analyzed, only EGFR was found to have added value when combined with known robust clinical prognostic factors in identifying three classes of patients with significantly different clinical outcomes. Although p53 staining associated with worse PFS and LRR, it did not significantly contribute to the prognostic model probably because there are biological and clinical differences among the disruptive and non-disruptive mutations within the gene that are not distinguished by IHC staining (41
This type of comprehensive prognostic data is critical for enriching patients with well defined tumor features and patterns of relapse as well as to prioritize clinically relevant laboratory biomarkers for enrollment into clinical trials and select appropriate correlative studies. Class III patients, having a 10% chance of survival and an 85% risk of developing LRR at 5 years, are ideal candidates for testing a vigorously intensified local therapy regimen. On the other hand, given the relatively favorable outcome achieved with radiation alone, it is prudent to test an agent with fairly low toxicity profile in combination with radiation for patients belonging to Class I. In addition to prognosis, this type of analyses can be applied to further examine the biomarkers for predictive potential. To ascertain EGFR expression levels as a potential predictive biomarker for either of the treatment arms (SFX versus AFX-C), we evaluated Cox models including assigned treatment (1 if SFX, 0 otherwise), EGFR (1 if > median, 0 otherwise), and treatment by EGFR interaction (as well as T stage, N stage, KPS, and primary site). The hazard ratios (95% CI for interaction) for OS, PFS, and LRR were 0.75 (0.45-1.25), 0.77 (0.45-1.31), and 0.88 (0.46-1.71), suggesting that patients with EGFR expression levels below the median may receive more benefit from AFX compared to patients with EGFR above the median. Although it did not reach statistical significance, it still generated an interesting hypothesis for future clinical trials. Thus, its use for stratifying or selecting patients for enrollment into clinical trials testing specific therapeutic strategy is recommended.
The results of the current work provide the impetus to perform other molecular marker assays using the samples in this series of patients with mature outcome data and conduct similar types of analyses to further refine risk classification. It is also desirable to incorporate the biomarkers to currently existing clinical prognostic information, and extend this approach to well defined group of patients enrolled into prospective trials testing the combination of radiation with chemotherapy or with novel agents when the clinical outcome data become available in the near future. With this goal in mind, many centers and cooperative groups have increased tissue banking efforts, including collection of samples suitable for high through put assays, for correlative studies. Such integrated investigations are changing the design of trials in the quest for realization of personalized medicine.