Molekylär klassificering av tjocktarmscancer : PAM-klusteranalys för identifiering av undergrupper

Sammanfattning: The main objective of this study is to divide a number of colorectal cancer cases into subgroups based on their molecular features using cluster analysis. The data used is supplied by a research group at Pathology, the Department of Medical Biosciences, Umeå University, and consists, after some preparation, of 455 observations which is a larger data set than many similar studies. The molecular variables that the clustering is based on are CIMP (CpG Island Methylator Phenotype), MSI (Micro Satellite Instability), BRAF- and KRAS-mutations. These are categorical variables and consequently the clustering method used is PAM (Partitioning Around Medoids) which is particularly useful with data on diverse variable level. The final analysis results in four subgroups that are represented by different combinations of attributes on the aforementioned variables. The disparity between the clusters are then evaluated by, for instance, comparing the survival time for their pertaining patients and it appears that two of the clusters are significantly different in this aspect. Other patient related and tumor specific characteristics are also linked with the separate cancer types and tested if they occur in varying extent. The locations of the tumors in the colon are for instance significantly different between the groups. Cluster analyses are exploratory tools so the choice of useful variables and subsequent interpretation of the results can be complicated and require relevant subject knowledge.