Transferability of health cost evaluation across locations in oncology: cluster and principal component analysis as...

The transferability of economic evaluation in health care is of increasing interest in today's globalized environment. Here, we propose a methodology for assessing the variability of data elements in cost evaluations in oncology. This method was tested in the context of the European Network of Excellence "Connective Tissues Cancers Network".

Using a database that was previously aimed at exploring sarcoma management practices in Rhône-Alpes (France) and Veneto (Italy), we developed a model to assess the transferability of health cost evaluation across different locations. A nested data structure with 60 final factors of variability (e.g., unit cost of chest radiograph) within 16 variability areas (e.g., unit cost of imaging) within 12 objects (e.g., diagnoses) was produced in Italy and France, separately. Distances between objects were measured by Euclidean distance, Mahalanobis distance, and city-block metric. A hierarchical structure using cluster analysis (CA) was constructed. The objects were also represented by their projections and area of variability through correlation studies using principal component analysis (PCA). Finally, a hierarchical clustering based on principal components was performed.

CA suggested four clusters of objects: chemotherapy in France; follow-up with relapse in Italy; diagnosis, surgery, radiotherapy, chemotherapy, and follow-up without relapse in Italy; and diagnosis, surgery, and follow-up with or without relapse in France. The variability between clusters was high, suggesting a lower transferability of results. Also, PCA showed a high variability (i.e. lower transferability) for diagnosis between both countries with regard to the quantities and unit costs of biopsies.

CA and PCA were found to be useful for assessing the variability of cost evaluations across countries. In future studies, regression methods could be applied after these methods to elucidate the determinants of the differences found in these analyses.