Biological predictors of treatment response in affective illness

Clinicians and investigators have sought better means of matching patients to treatments since the advent of modern psychopharmacology. Attempts to use clinical features to predict which patients will respond to specific medications have met with limited success (see article by Nierenberg in this issue). Researchers also have sought biological tests that might discriminate between likely treatment responders and nonresponders or identify those at greater risk for relapse. Investigating techniques have ranged from urine neurotransmitter metabolites to electroencephalogram (EEG). Most recently, the widespread availability of neuroimaging technology has led to new efforts to apply these techniques to predict treatment response in psychiatry. The search for response predictors is driven by two related goals. The first is identifying the underlying pathophysiology associated with a given disease. Just as the utility of selective serotonin reuptake inhibitors in major depression provided evidence for serotonergic dysregulation in some affective illnesses, the discovery that changes in the metabolism of certain brain regions predict treatment response would implicate those structures in the process of disease development or recovery. These findings could guide future drug development. A second goal is to allow more targeted and focused clinical interventions. For example, while first-line antidepressant treatments have response rates of 50% or more, many patients still fail to respond to multiple interventions [1,2]. As each new pharmacotherapy is tried, patients are exposed to additional cost, adverse effect burden, and the potential for loss of function and suicide. Importantly, these two goals imply different criteria for what makes a response predictor useful. From a research perspective, identifying any association would be valuable in guiding future studies. From a clinical perspective, however, the requirements are more stringent and rarely have been achieved. Such factors as positive and negative predictive value and cost and availability of technology become critical in determining the utility of a predictive test (see article by Nierenberg in this issue). Attempts to identify biological response predictors generally have met with limited success, particularly where the goal is to develop clinically useful indices. This article reviews the biological approaches to predicting treatment response, beginning with neuroendocrine studies and EEG analysis and concludes with structural and functional neuroimaging. Where relevant, the authors identify studies correlating change in a measure during treatment; though these studies identify correlates of response rather than true predictors, they also may point the way to useful predictive measures. The article describes the designs of typical studies to aid in interpreting their results and concludes by addressing some of the problems and limitations associated with these approaches and suggesting future directions for this research. Original reports or reviews included in this article were identified by conducting a MEDLINE search with the terms "response," "predictor," and "treatment," combined with "depression," "bipolar disorder," or "affective illness." References in the publications identified were reviewed manually to locate additional relevant publications. Of note, pharmacogenomics, which shows great promise for response prediction in psychiatry, is discussed elsewhere (see article by Pickar in this issue).