|
Abstract:
|
Clinicians and patients may use a treatment only if it is predicted to highly benefit combination of multiple outcomes. For example, patients with primary progressive aphasia may choose to use tDCS (transcranial direct current stimulation) to treat their affected brain areas, but only if tDCS also improves their writing difficulties. In principle, methods to characterize such patients, can compare the new (tDCS ) vs a standard treatment for different covariate profiles, and select the profiles whose effects meet the patients’ needs. Standard methods to do this first, estimate a prediction model, and, second, use the model to select profiles with high effect; however, estimation is not informed about the ultimate goal. In previous work where the outcome is a simple measure, we have shown that standard methods can dramatically misrepresent high-benefit patients; and we have proposed better methods that directly link the clinical goal to the construction of the method. In this work, we use these ideas to construct methods to characterize patients who highly benefit in a combination of outcomes.
|
