Statistical Method for Development of Composite Index in Clinical Research |Biomedgrid

In clinical research, a medical predictive modelling is often performed using a multivariate set of risk factors to predict the performance of clinical outcome for an effective disease management. Using a well-established and validated medical predictive model, our goal is to develop a composite index of several dependent predictors to better inform the disease status and/or treatment effect with more accurate and reliable assessments. In practice, since each of the multiple predictors may be positively or negatively and/or linearly or nonlinearly correlated to the clinical outcome or response, an ideal composite index should be able to account for positively/negatively and/or linearly/non-linearly associations with the clinical outcome or response. In this article, criteria and a statistical approach for development of an ideal composite index are proposed. Under the proposed criteria and procedure, statistical methods are also derived. The proposed procedure for development of the composite index is evaluated both theoretically and via a clinical trial simulation.