Multivariate regression analysis to predict progression of oral premalignant lesions

B138

Survival rates for oral cancer patients have remained unchanged in the past several decades largely because of the late identification of the disease and the high rate of local recurrence after treatment. We present research based upon an approach that uses automated quantitative microscopy technology in conjunction with molecular approaches to create an integrated plan for patient assessment and management that would be rapid and cost-effective.
 >Objective: The aim of this study was to assess the potential of Quantitative Tissue Phenotype (QTP) and its correlation with conventional histopathology, allelic losses for prediction of cancer development from OPLs.
 >Methods: A total of 138 oral mucosa lesions were analyzed. The distribution of the pathology grades was as follow: normal- 30, hyperplasia -21, mild dysplasia -13, moderate dysplasia-10, sever dysplasia/CIS- 35, and invasive Squamous Cell Cancer -29. Thoinin-Feulgen stained sections, adjacent to the H&E sections used for pathological assessment, were analyzed using Our in-house imaging Software Getafics (BCCRC). A Region of Interest (ROI) was manually selected and all intact-in-focus nuclei were collected for image processing and feature extraction. Samples from normal oral epithelium and samples from SCC were used to generate a cell-by cell nuclear morphometric index, a linear combination of 5 features measuring the shape, size of the nuclei, as well as the DNA chromatin organization changes. This 5-feature function best discriminates cells from normal and cancer specimens. A Nuclear Morphometric Score (NPS) was then generated for each epithelial specimen based upon the frequency of cells with specific nuclear morphometric indices. All samples were assayed for LOH on 7 arms (3p, 9p, 4q, 8p, 11q, 13q, and 17p). The highest risk of progression was associated with LOH at 3p &/or 9p in the presence of LOH at any other arm (RR = 33.4; n = 34, MR3). The time course data for these patients has been stratified into those which develop cancer (29 lesions) and those which have not developed cancer after at least 10 years of follow-up (15 lesions)
 >Results: We investigated the potential of Quantitative Tissue Phenotype (as measured by the NPS), to recognize severe dysplasia/carcinoma in situ (CIS) (known to have an increased risk of transformation into invasive cancer) and to predict progression of hyperplasia/mild/moderate dysplasia (termed HMD). Using the NPS in this pilot data we can correctly identify 94% of the high-grade OPLs while maintaining a specificity of 74%. The NPS can be used in this pilot study to identify patients which progressed to cancer 77% of the time while correctly identifying the patient which do not progress 78% of the time. There was significant correlation between lesions with High NPS and lesion with genetic damage. From a Cox model the most predictive factors for cancer risk are LOH (p =0.0008) and NPS (p = 0.0007). In the multivariate Cox model, LOH (p =0.01) and NPS (p = 0.07) are the strongest predictors for cancer development. All analyses showed that pathology was not a predictor for HMD progression to cancer.
 >Conclusions: These data support the potential utility of Quantitative Tissue Phenotype as a marker associated with molecular damage and with cancer development. QTP could be used to identify lesions that require molecular evaluation.