IS VERY EARLY HTA POSSIBLE? A MATHEMATICAL FRAMEWORK OF USING ECONOMIC EVALUATION MODELS TO INFLUENCE THE DIRECTION OF DRUG DISCOVERY

Purpose:

Drug discovery is a time consuming and costly process. Mathematical models were commonly used to facilitate the process. For example, pharmacoinformatics models enable researchers to screen millions of molecules and select important ones for synthesis; economic evaluation models facilitate the process of regulatory approval and reimbursement.

Incorporating economic evaluation models to enhance the drug discovery process has gained much attention in the past few years_.  However, detailed methodology has yet been developed. The objective of this research is to develop a mathematical framework that model the relationship between the molecular descriptors and the outcomes of economic evaluation, such that this method can be used to screen for molecules that would be both cost-effective and have high affinity in the early stages of the drug discovery process.

Method:

Our mathematical framework (Figure) can be described in five steps. The first two steps perform a quantitative-structure-activity-relationship (QSAR) analysis. In the first step, we generate a vector space model molecular descriptor (VSMMD) for each molecule in the training set. In the second step, we use a kernel function to map each VSMMD to a feature space typically used for classification. The third step is to generate a new point in the kernel feature space using a kernel feature space algorithm (i.e. Design a new point that is cost-effective in the feature space). In the fourth step, we map this point from the feature space back to the input space using a pre-image approximation algorithm. In the last step, the molecular structure template will be built by the molecule recovery algorithm.

Result:

VSMMD, a constitutional descriptor that provide feature counts related to the two dimensional structure of a molecule was developed. VSMMD based on a vector space model was suitable for kernel studies in QSAR analysis.  We have shown that the VSMMD is able to capture affinity and binding mode information intrinsically using kernel methods.  In additional, through multi-task statistical learning with kernel methods, VSMMD not only can predict the outcomes of the activity level but also whether that molecule is cost-effective or not.

Conclusion:

Although others have already introduced the concepts of bringing in health technology assessment and cost-effectiveness considerations at an early stage of drug development, this research is the first to provide a mathematical framework to automate the process.