[Skip to Navigation]
Comment & Response
February 6, 2019

Importance of Variable Selection in Multimodal Prediction Models in Patients at Clinical High Risk for Psychosis and Recent Onset Depression—Reply

Author Affiliations
  • 1Department of Psychiatry and Psychotherapy, Ludwig-Maximilian-University, Munich, Germany
  • 2Institute of Mental Health, University of Birmingham, Birmingham, United Kingdom
  • 3School of Psychology, University of Birmingham, United Kingdom
  • 4Orygen, the National Centre of Excellence for Youth Mental Health, Melbourne, Australia
  • 5Centre for Youth Mental Health, University of Melbourne, Melbourne, Australia
JAMA Psychiatry. 2019;76(3):339-340. doi:10.1001/jamapsychiatry.2018.4237

In Reply We thank the authors for their interest in our Original Investigation published in JAMA Psychiatry.1 Nelson et al raised a query that our analysis was biased toward advocating structural magnetic resonance imaging for the prediction of functional outcomes in the clinical high-risk states for psychosis or in recent onset depression. More specifically, the colleagues were concerned that we had a priori picked just 8 functional baseline variables instead of using the full breadth of the clinical database available in our data. They suggest that the rationale of this strategy was to create a weak “straw man” model that would have been easily outperformed by the combined predictor, allowing us to disregard clinical assessments in favor of more costly magnetic resonance imaging as prognostic tool.

Add or change institution