Cluster analysis and principal component analysis variability of data in cost evaluations: methods and applications in oncology.

OBJECTIVES: In the context of today’s highly globalized environment, the interest in the transferability of data of cost evaluation in health care has strongly intensified. A methodology is proposed to explore the similarity versus dissimilarity of data and hence their transferability across locations. METHODS: The main steps are (i) the definition of the objects (e.g. countries), the identification of potential variability factors, the selection of final variability factors, and the construction of variability areas (e.g. unit cost of personnel); (ii) the measure of distances between the objects, the determination of clusters and the construction of the hierarchical tree using the cluster analysis (CA); (iii) the projection of the objects into factorial planes and the linkage between objects and areas of variability using principal component analysis (PCA). Suggested methods are applied to an international cost evaluation performed within the European network of excellence CONnectiveTIssuesCAncersNETwork. RESULTS: 12 objects and 16 areas of variability were defined. CA shows four clusters meaning that data belonging to different clusters are dissimilar (i) chemotherapy in France, (ii) follow-up with relapse in Italy, (iii) diagnosis, surgery, chemotherapy, radiotherapy, and follow-up without relapse in Italy, (iv) diagnosis, surgery, radiotherapy, follow-up without relapse, and follow-up with relapse in France. PCA opposes (i) follow-up with relapse in Italy to diagnosis, radiotherapy, and follow-up with relapse in France; (ii) chemotherapy in France to follow-up without relapse in France. In sarcoma patients, transferability is then limited for chemotherapy during the initial treatment in France and the follow-up with relapse in Italy. Diagnosis cannot be transferred either between France and Italy regarding the quantities and unit costs of the biopsies. CONCLUSIONS: Using CA and PCA enables health care professionals to rapidly emphasize the variability of data and therefore to determine the transferability of cost evaluations across locations.