Multi-domain Longitudinal Modeling

Abstract

Individual domains of multi-domain, patient-reported outcome instruments are typically analyzed independently, without considering inherent interdependency. The authors assessed correlations across time (longitudinal) and across domains of a multi-domain instrument, simultaneously, in a single, unified, and cohesive integrated model. International Index of Erectile Function (IIEF) data from a trial of sildenafil in erectile dysfunction were used. For the multi-domain longitudinal modeling, covariance of response was constructed using (1) the Kronecker product of an unstructured covariance matrix to model across domains, with an unstructured covariance matrix to model across time, and (2) a completely general, unstructured covariance matrix. Treatment effects for all IIEF domain scores calculated using individual or multi-domain modeling were similar, reflecting the robust application of the integrated model to these data. In conclusion, modeling correlations simultaneously across domains and across time more rigorously address the natural interrelationship between domains and can provide a more accurate representation of treatment effects.